AMD Fusion To Add To x86 ISA

Giants2.0 writes "Ars Technica has a brief article detailing some of the prospects of AMD's attempt to fuse the CPU and GPU, including the fact that AMD's Fusion will modify the x86 ISA. From the article, 'To support CPU/GPU integration at either level of complexity (i.e. the modular core level or something deeper), AMD has already stated that they'll need to add a graphics-specific extension to the x86 ISA. Indeed, a future GPU-oriented ISA extension may form part of the reason for the company's recently announced "close to metal"^TM (CTM) initiative.'"

Comments

Am I the only one?

[man_of_mr_e]

Am I the only that thinks this is a bad idea? Either I change video cards more often than CPU's or CPU's more than graphics cards, but in either case I seldom want to upgrade both at the same time. Although I suppose I wouldn't mind a better GPU "for free" with my CPU, I suspect it won't be "for free".

Re:Am I the only one?

[Anonymous Coward]

I've always upgraded both at the same time

Re:Am I the only one?

[r_jensen11]

I'm guessing that, as with integrated graphics, having (a) shared GPU/CPU(s) would allow having an additional video card. I seriously doubt they're going to remove the PCIe 16x slot from motherboards any time soon.

Re:Am I the only one?

[hawkbug]

Yeah, I thought that same thing at first. However, I don't think we are the target market. I think Laptops and OEMs will be the market for this. Just imagine a mac-mini type computer from Dell or somebody. Onboard video has been around for ages, but if the board could be smaller since the gpu is on the cpu, then you'd save space and power so the machine could be smaller and theoretically cheaper.

Re:Am I the only one?

[Anonymous Coward]

Well that all depends on how much performance can be gained by the integration. I seem to remember this as a weak argument for why numeric co-processors shouldn't be
integrated into the 486DX...

Re:Am I the only one?

[Rosco P. Coltrane]

Am I the only that thinks this is a bad idea? Either I change video cards more often than CPU's or CPU's more than graphics cards, but in either case I seldom want to upgrade both at the same time. Although I suppose I wouldn't mind a better GPU "for free" with my CPU, I suspect it won't be "for free".

Look at it this way: nowadays you can get a computer with a video "card" onboard the motherboard, but nothing prevents you from disabling it and installing a separate video card. Most likely, that's what's gonna happen in this case. But remember, one of the advantage of merging the cpu and gpu is to get around the bus bottleneck, so presumable an embedded gpu may well blow any separate video card out of the water in terms of performance. Unless of course this is just a marketting stunt, or a consumer lock-in scheme, which is just as likely...

Re:Am I the only one?

[rubberbando]

I'm betting that this will be just like how many motherboards come with 'integrated graphics' to fall back on instead of having a dedicated graphics card. All they're doing is shifting the GPU (as well as the cost) to the CPU core from the motherboard. Pricewise, this will probably suck, however performance will most like be greatly increased so it will be a good thing for OEMs and those who buy such machines.

Re:Am I the only one?

[electrofreak]

I think it's a terrible idea. I also only upgrade one of those at a time. Plus, it'd be a waste of money for those that don't even need a high end GPU to go with their high end CPU. Or, even the other way... gamers often buy lower end chips and then overclock them, then they might get a high end GPU. Unless you think that each chip will have a whole spectrum of GPUs to go with it.

Terrible idea.

Re:Am I the only one?

[cnettel]

On the other hand, the real payoff of low latency won't surface if every operation means going through a driver, which only then realizes "oh, I have a single instruction for this thing, let's head back to the caller". This means that game writers will either still need to batch up complex operations, that the driver will then translate into batches of suitable instructions, or that we'll see games/applications with radically different codepaths. Any attempt to benefit optimally from the integrated approach will perform badly on a separate card, while code tuned to a separate card won't come close to harnessing the good points of an underpowered, but lower latency, local graphics implementation.

It's almost like they would add L3 in a non-transparent manner, that is, expecting the developers to write the code moving suitable data into the cache and addressing that data in a radically different manner, while still also supporting the normal style of memory access, where you of course need to care about the cache, but not so explicitly. (The Cell's explicit local RAM for each unit, and the whole design of that beast, comes to mind. At least ALL PS3s will have one, but the expected target market for Fusion-only adaptations is much less clear cut.)

And, yeah, this is quite like the situation almost ten years ago, when 3D cards were hot and new. Writing a pipeline to feed those cards was quite different from rolling your own hacked-up software rendered. (And with T&L and shaders, the move has been even greater.)

But maybe then I'm just speculating a bit too much here. It would make sense that AMD is designing these instructions to fit into the existing driver model (or at least the DX10 one), so that you can get pretty good performance by just doing the relevant translation there.

Re:Am I the only one?

[tomstdenis]

Yeah, except that the die that goes in a mobile and a desktop are not always different designs. Just the ones that can run low voltage make it there. Granted, the turions are 754-pin so they're not the same as the AM2 desktops but it was the case back in the day (e.g. you could throw a turion in a 754-pin desktop and use it there).

So a "laptop chip" is not always a distinct design. Even in the Intel world with the 479-pin Core 2 Duo mobiles and 775-pin desktops they're likely very similar internally (once you get past the front side bus). So adding a GPU to one and not the other is probably a pain in the ass.

If AMD decides to fracture their cpu lines and have totally distinct lines I think they'll suffer. They don't exactly have the mass staff to support multiple product lines (of completely different design). And I can't see them putting GPU extensions in the Opteron line...

Tom

Re:Am I the only one?

[drinkypoo]

All they're doing is shifting the GPU (as well as the cost) to the CPU core from the motherboard.

They're also eliminating all of the components between the CPU core and the GPU. In theory they could have a HT chip that handled all of the I/O and didn't even present a traditional system bus, if they felt they didn't need expansion slots. Thus you could eliminate the PCI/PCI-E bus and all the things needed to support it; at minimum however you are eliminating the bus between the North Bridge and the GPU and all that entails... which is a lot.

Re:Am I the only one?

[P3NIS_CLEAVER]

Realistically, how many computer buyers upgrade their video cards, 1/2%?

Re:Am I the only one?

[mrchaotica]

I seriously doubt they're going to remove the PCIe 16x slot from motherboards any time soon.

What I'd like to see is for AMD to put the CPU and GPU on separate chips, but make them pin-compatible and attach them both to the hypertransport bus. How cool would it be to have a 4-socket motherboard where you could plug in 4 CPUs, 4 GPUs*, or anything in between?

*Obviously if it were all GPUs it wouldn't be a general-purpose PC, but it would make one hell of a DSP or cluster node!

Re:Am I the only one?

[forkazoo]

Am I the only that thinks this is a bad idea? Either I change video cards more often than CPU's or CPU's more than graphics cards, but in either case I seldom want to upgrade both at the same time. Although I suppose I wouldn't mind a better GPU "for free" with my CPU, I suspect it won't be "for free".

People said the same thing about memory controllers, FPU's, on-motherboard audio etc. Nowadays, nobody would go out of their way to get a special chip without an FPU. It simply wouldn't be cheaper to avoid getting it. Because the mass market is buying the all-in-one version, it would be insanely expensive to do a custom run of the "cheaper because they are simpler" versions. The best part of a hybrid CPU/GPU is that you have insane bandwidth between the CPU and GPU. Bandwidth to memory may be mediocre compared to the very specialised setups on a modern graphics card, but you may still come out ahead for many tasks.

Re:Am I the only one?

[Reason58]

Once in mass production I'm sure it wouldn't raise the cost of a CPU by that much, but then again how much does it really cost to add an onboard graphics chip to a motherboard? It seems like a lot of research, development, and production money for little gain. Maybe I'm missing some bigger picture here?

Re:Am I the only one?

[mrchaotica]

Shouldn't this kind of thing be the compiler or library writers' problem, not the application developers'?

Re:Am I the only one?

[timeOday]

in either case I seldom want to upgrade both at the same time.

What if the new architecture's "graphics" pipelines aren't dedicated exclusively to graphics and can be used to speed up other tasks? Or if the total power consumption of the integrated component is 50% of separate components? Or if a non-upgradeable "PC on a chip" with no expansion bus offers great performance at 50% the cost of a traditional PC?

Re:Am I the only one?

[gripen40k]

From what I understand I think the parent is right. If you use OpenGL, you don't worry about pipelining or however else the computer actually 'makes' the graphics, you just code it. Buuuutt.... I guess you would need to compile two versions of the same thing and put it on the same game disk, or figure out some kind of neat system so that translations are done in real time with hardware (much faster than the soft approach).

Re:Am I the only one?

[ruiner13]

Dead on. Think of the power savings for laptops, not needing to have to use energy to drive a pci-e slot with a graphic chip that only gets replaced when the laptop does. It would also allow for really slick interfaces on smaller devices, such as tablets, pdas, etc. It would also have one hell of a bandwidth rate to the processor, including full speed access to the computer's RAM. I don't think they'd give it dedicated memory die to die size, but it sure would beat going over a pci-e bus like today's shared memory integrated chipsets.

Re:Am I the only one?

[Anonymous Coward]

Am I the only that thinks this is a bad idea?

No but probably the minority.

Either I change video cards more often than CPU's or CPU's more than graphics cards, but in either case I seldom want to upgrade both at the same time.

RTFA you dumbass! This will most likely target laptops. How often do you upgrade your laptop's GPU?

Re:Am I the only one?

[mr_mischief]

Don't really think of it as a separate graphics card pushed onto the CPU. MMX, SSE, SSE2, 3DNow!, AltiVec, et cetera have all brought specialized vector processing functions into CPUs with existing sets of opcodes. You know what GPUs tend to be really good at? Vector processing, matrix transforms, and all of that other highly parallel, small-data work. Now, you'll have an extended instruction set that does that well. Hopefully, it'll be done as part of the CPU's instruction set, or as a very tightly-bound coprocessor.

Imagine this scenario: A high-end game runs a physics engine on your CPU. It wants to run on your GPU, but that's busy doing the shading and rendering. So it runs on your second core if you have one, or your second CPU, or maybe on your second GPU, or on a dedicated physics board. Or it just falls back to the CPU, and eats up time better devoted to game logic and the IM you're about to get about your sick aunt from your Mom, letting you know that your aunt is all better and you don't have to quit your game of FarCry 16 or Half-Life 24 or whatever to go see her.

Okay, now imagine you've got a CPU that can do pretty much what a GPU can do besides its own specialty of kicking ass on integer logic and cache handling. It's a dual-core CPU with a built-in GPU to boot. The game logic runs mostly on core 1, the physics runs mostly on the built-in GPU, the rest of the system mostly runs on core 2, and you're a real gamer so your shading and rendering still happens on a dedicated graphics card with its own GPU. You only needed one Awesome Local Gamers Advanced Extreme XII (or "ALGAE-boxcars" to boys over at PC Rag) slot on your motherboard, because one of your GPUs was already on the CPU. That sick aunt? She types up an email about her pending medical malpractice suit on her system that has one chip socket and no ALGAE slot (not even ALAGE-uno, dammit), because all she needed was a driver for her entry-level laptop that basically runs video right from the CPU/GPU through some lead traces and a $1.50 blitter chip.

People who need high-end, specialized hardware to get the performance they want or need are not going to be left without options. People who just want to write a letter, play Tetris-klones, and maybe some low-end RTS games will get a cheaper system by not having to pay a third party for more hardware and more integration of disparate components than they really care about. It's win-win.

It's a cycle of breaking things out of the center of the system for dedication, then bringing them back into the center when the technology reaches a point that tighter integration is possible and helpful. We don't need sound on the CPU because there's no bottleneck, for example. We didn't need floating point on the CPU at one time, but enough people were using it and it got to be such a slowdown to use a separate chip for it, that it got placed on the CPU. AMD did memory controllers on the chip, and noone talked about how horrible an idea it was (except Intel, who will probably eventually do the same). GPUs are just next.

If most PCs sold started to be used for speech recognition instead of keyboard input, we'd see dedicated PCI or PCIe x1 cards for the work involved with doing that really, really well. Then, if there was need, x4 or x8 versions of the boards would appear, and motherboard manufacturers would take that slot need into consideration. Some boards would get speech recognition chips integrated onto them. Then, if it continued to be the primary input device, and there was still a big enough bottleneck at x16, we'd see that integrated onto the CPU once the die sizes shrunk enough that the "traditional" core could use tighter integration with the speech recognition stuff. I don't know that good speech recognition would take that level of integration to be efficient, but I couldn't think of a more hyped example of what seems _perpetually_ on the horizon "in just a few years at the current pace of microprocessor innovation".

People thought discrete sound and video were silly an

Re:Am I the only one?

[man_of_mr_e]

If we're talking about a substandard "onboard" kind of graphics, then I suppose it's not a big deal. You know, like Intel onboard graphics or whatever. But if we're talking about high end video, which is what I would suppose it would be, otherwise there would be no purpose in trying to make it "ultra efficient", then that changes far too fast.

I don't want to pay $500 premium on my CPU for something that will be outdated in 6 months. I mean, I can sell off my old high end video card in 6 months and buy the latest stuff, I can't sell off the integrated GPU.

Re:Am I the only one?

[kramulous]

I don't think that it is necessarily a bad idea. Some of the datasets I need to display are far bigger than what the greatest video card (currently) can handle. The data displayed has too many pixels (to see without resizing) than what the graphics card (with appropriate displays) is capable of. From what I see that the immediate future shows, the graphics card is in a position where it will never be able to catch up (yes, even with parallel pipes). I'll leave it for the active user to prove that it is even worse when considering memory. We've had to go for the software implementation on a cluster.

This all said, it is probably perfect for a notebook/laptop/[insert modern notation here].

What I do love about the idea is to implement functions/methods/[insert language specific notation here] for those specific algorithms that can be written to take advantage of the parallel architecture of the GPU, while the CPU can spend time processing stuff that the GPU cannot do.

Re:Am I the only one?

[Sparohok]

The history of computer architecture is a ceaseless march toward higher integration and higher generality. That is, what was once special purpose hardware in a seperate unit is now implemented in firmware as part of a more flexible, general purpose processor.

This march is littered with those standing by the wayside saying things like, "Who needs floating point in the CPU? Leave it on a seperate chip!" or "I want to be able to upgrade my CPU without buying a new memory controller!" or "If you integrate sound on the motherboard, I'll just have to buy a seperate sound card."

So, while I'm sure you're not the only one who thinks GPU integration is a bad idea, that doesn't mean it is a bad idea.

Re:Am I the only one?

[Rycross]

I see a lot of people commenting that you'll probably be able to disable it and slide in another video card, but I'm not seeing the other obvious suggestion: your integrated gpu becomes a backup gpu for your main video card.

See, the big trend for high-performance video nowadays is SLI, which involves sticking a bunch of identical video cards in a computer and connecting them with a bridge. How much more work would be necessary to have that backup gpu support your main card?

Re:Am I the only one?

[Morphine007]

I know that graphics-guru's used to have to write a lot of inline assembly and other optimizers in order to get certain parts of their code to run a little bit faster. So, the majority of people won't need to delve into that kind of depth... but there are some who push the envelope a bit and who, traditionally, needed to get their hands dirty on the low level stuff. No idea if this is still the case... maybe someone a little more involved in pushing the envelope could answer... Mr. Carmack, are you out there?

Re:Am I the only one?

[Alien Being]

"Buuuutt.... I guess you would need to compile two versions of the same thing and put it on the same game disk, or figure out some kind of neat system so that translations are done in real time with hardware (much faster than the soft approach)."

I don't think so. The application can be linked against a single graphics library. The GL just swaps some function pointers when special hardware is available.

Re:Am I the only one?

[carl0ski]

Yes! Yes you are the only one
The purpose of the Fusion program is not to target users that want to use high end graphics ala stand alone GTX/XT Nvidia/ATI PCI-e graphics cards.
It is to replace the requirement of a seperate chipe for onboard graphic card users.

This includes:
Notebooks
Enterprise Desktop computers
Workstations
Servers
and potentially ultra small devices.

Personally i started with an onboard ATI Xpress200 Graphics card it was as fast as my previous 9600XT. I later upgraded to a higher end Graphics card.

there is no doubt a GPU with direct access to the CPU and potetnially system ram will be faster and more efficient than current onboard solutions available
Hell in a perfect senario it may bridge the compatability gap between windows Gaming PC's and Consoles.
A PS2/XBOX game works on all PS2s/XBOXs
All PC games react different to every completely different PC.

Re:Am I the only one?

[obeythefist]

Remember the characteristics of performance GPUs at the moment - AMD's Fusion technologies are not aimed at performance gaming at all, they're squarely aimed at the imbedded/budget/mobile segments of the market.

Just look at the memory speed of your average CPU (667MHz DDR2) and your average GPU (1GHz DDR3 is good).

Now, hooking your GPU up to the main system memory takes away a huge chunk of your performance. Saves you a lot of money though. If you're not sure, go look at how overclocking video memory affects performance, just about any of your THG/Anandtech/Extremetech sites will tell you.

This is just another method of doing your onboard graphics solution. It will perform much better than current integrated graphics solutions. But there's no way it can outperform a quad-SLI rig, even on paper.

Re:Am I the only one?

[rbanffy]

As added benefits:

- With a public and standard ISA, you will have Linux-compatible drivers shortly

- With a public and standard ISA, people will have a single standard to code against. Library support should be excellent.

- While your über-FPUs/vector accelerators/stream processors (what GPUs are made of) are not GPU-ing something, they can accelerate SSL, physics processing and any other vector-friendly activity you may have. Playing Flash content, maybe.

- GPUs are memory-hungry. The added memory bandwidth will benefit all software, not only graphics-intensive stuff.

- There is nothing that precludes you from using a stand-alone GPU, provided you have the drivers. But your CPU will have a couple high performance units that can give it a hand. Think asymmetric SLI.

We will see how well the idea performs by watching the Cell processor (a CPU with 8 "GPU"s attached) in the PS3. That's roughly the same idea.

In the meantime, I bet it will work just fine.

Re:Am I the only one?

[obeythefist]

And that lovely less-than-average performing integrated GPU is taking up space that could be used with a CPU, so you're sacrificing your SMP for the sake of that.

Unless you get two sockets, so two GPUs and two CPUs. Some bizarre new AMD crossfire monster. But the memory throughput would still be weaker than using PCIe cards. 4x4 would be better performance.

Slightly off topic - has anyone noticed that the Vista EULA only allows two CPUs? What does this mean for Intel and AMD quad cores? Everyone has to buy a Windows Server 2003 license?

Re:Am I the only one?

[Duncan3]

You're failing to take into account that CPU's stubbed their toes a while back and fell off the Moore's law curve. GPU's are ~40x or so faster now, and AMD was damn smart to grab ATI when they did.

What's really happening is that the CPU doesn't matter to anyone, they are fast enough for Word and Excel, and are falling fast in price. A 2.2 Ghz Athlon 64 core is only $242 for 2, and an Intel 3Ghz is about the same $257 for 2.

Gamers are the only thing still driving the market, and they want GPU's with massive GFLOPS. Everyone else wants low power cheap CPUs with almost no profit margin, because it's good enough.

Re:Am I the only one?

[man_of_mr_e]

I'm not entirely sure I agree. MMU's or FPU's are central processor functions. People didn't typically upgrade these regularly, though sometimes they would buy machines without them and want to add them later.

The video, on the other hand, is something that is in constant flux and is an incredibly competitive market. Sound is ok on the motherboard, as is video for "typical" users, because they don't have high demands. Putting it in the CPU, on the other hand, seems kind of silly, since the CPU is the single most expensive component in the computer in most cases.

Re:Am I the only one?

[man_of_mr_e]

That really makes no sense. Why put it on the CPU unless you were interested in high performance, low latency gaming? I could see it being used in a console, sure, where you don't upgrade the equipment. Possibly Laptops, too, but not Desktops. What's wrong with being on the chipset? Oh, right, AMD doesn't make chipsets anymore, but I guess ATI does...

Re:Am I the only one?

[man_of_mr_e]

Gamers aren't the only ones driving the market. You'll want a fast dual core processor for Vista, for instance. Software will always expand to fill all available CPU Cycles. I just see software becoming more powerful, not just games. Thinks like more AI in real-world apps, star trek level stuff.

Re:Am I the only one?

[Anonymous Coward]

Either I change video cards more often than CPU's or CPU's more than graphics cards, but in either case I seldom want to upgrade both at the same time.

Yeah, tell me about it! I used to change engine and transmission at different times; now I just grumble and buy a new car.

Re:Am I the only one?

[TheRealMindChild]

Except AMD bought VIA who had the MediaGX line which is exactally this, only about 8 years ago

Re:Am I the only one?

[afidel]

Except GPU's are at least as high transistor count as CPU's these days and based on transistor count growth GPU's will exceed CPU's in the next one to two generations. That's a lot different then all of your examples.

Re:Am I the only one?

[hackstraw]

What I'd like to see is for AMD to put the CPU and GPU on separate chips, but make them pin-compatible and attach them both to the hypertransport bus.

Maybe I'm just projecting, but what I think that you want, and many others out there want is simply a better bus, bridge, magic, glue or whatever you want to call it between the major parts of a computer.

There are many other xPUs out there that people want to add to computers, but the price/performance, especially if the data must go out to the peripheral and back to the main memory, CPU, or whatever simply does not pan out. I'm talking about crypto units, physics units, custom VLSI units, FFT units, TCP offloaders, there are tons of things that could make use of a better bus on a computer, but in recent history these things have mostly failed because the time that the price/performance seems reasonable CPUs progress to the point that they can just brute force the processing about as fast as the specialized addon can.

Right now we have FPUs, CPUs, and GPUs, and GPUs were able to sneak around the price/performance barrier because they are pretty much uni-directional devices.

Now regarding AMD's thrust here. I could be very wrong, but I believe that its either a waste of time or its something that will revolutionize computing, but I simply don't see the latter. A combo CPU/GPU either only targets the very low end generic computer or a specialized graphics type of computer. With the failure of the other addons to computers, I don't see the advantage. Graphics simply don't matter in the server market.

Re:Am I the only one?

[Duncan3]

No, you want a fast GPU for the eye candy.

From what I've read Vista with the eye candy off takes no more CPU or RAM then XP did.

.

Re:Am I the only one?

[SirKron]

Your thinking is so two years ago. Why do you need 4 sockets? Did you not see this coming when AMD announced purchasing ATI and an eight core processor for desktop users?

Re:Am I the only one?

[byteframe]

Integrating CPU with the GPU? That would be cool, however in the long term I don't really see it happening, or most anything else mentioned here. Rather this 'fusion' of technology will most likely just end up as a new processor, based around a concept similiar to Cell, but probably without an emphasis on the differant 'processing units'. We'll probably end up with a new cpu design, hopefully x86 compatible, (although if they could create a really really good cpu, I mean the next big thing, they really shouldnt feel like they have to work with the x86 ISA), with all these smaller units (not cores) doing everything. Think of the latest 8800 GTX, it has 128 unifed 'Streaming Processors',that were previously known as pixel shader units (or a variation thereof). Graphics cards also have/had vector shader units, and the ability to use geometry shaders, however these (at least for g80) are all unified now, in the sense that these 128 'Streaming Processors' (notice the name change AND it's significance) can on-the-fly change their function to either of the three categories. Cool idea for the GPU, but it also hints at where things are going. The Voodoo2 is more of a 'graphics processor' than g80, IMHO, in the sense that gpus are just essenestially becoming 128 'tiny'-cored general purpose coprocessors, that area really good at floating point math, (essential for graphics). As far as today's graphics cards go we can run C code on these things. Sounds more like an auxillary, differant, type of CPU, that's all today's graphics cards are. check it out... http://developer.nvidia.com/object/cuda.html/ So the wheel of reincarnation has started. It began with GLQUAKE, with an addon card (that was an addon to your 2D card...) and will now end with either a CPU w/ GPU (fusion) or even better, a CPU that incorporates processing concepts from the GPU into a new design that is really really good, especially in the areas needed for Quake VI. PPSh-41: Well still have pci-e express, and better still we'll probably have PCIe x16 slots too. Whether or not the consolidated market will still continue to sell graphics coprocessors is beyond my skill of intuition, however they wont stop selling them for at least 6 years. Another possible scenario is an Intel/NVIDIA buyout (I like nvidia, dont buy them out, in fact just establish a partnership). In that case, I'd venture to say that with each cpu/gpu maker in their own business couplings the GPU as we know it (as of right now) is dead. Remember what tim sweeney said, "something something blah blah blah, what we'll find is that [graphics] is a problem that turns out was just general computing". That definatly seems like the case, when you consider the ps3's graphics (which are probably good, havent checked it out, not that I'd buy a console). If a new (more than just an extention to the x86 spec) cpu comes out of all this technology, I really really doubt well be using a GPU, as it'd be kinda like just having another CPU, but in a pci slot, with a two slot cooler, loud noise, more cords, bigger cases, more heat etc etc. Provided the new cpu specification can be used buy both companies to make compatible cpus, well all see the extinction of the GPU, as well as numerous benefits mentioned a second ago. So bottom line, and in my best hopes, a Slashdot reading Dual Core, Dual GPU fat person, will eventually have a new box in 2012, right before the world ends. It will be running GNU/Linux), and will no more than the size of a mid tower. However the CPU in the system will be a really really fast, and will have hundreds of the 'streaming processors' (that are the basis of out latest gpu's processing ability) that can be used for games, graphics, or anything else, or even porn.

Re:Am I the only one?

[Anonymous Coward]

RE Vista: I'm fairly certain XP considers a dual core cpu to be only 1 CPU for licensing purposes. This definitely applys to Intels' HyperThreading and I think it was extended to dual/quad core chips. Basically 'CPU' in the license refers to the physical item not how many cores are packed into it.

Now Oracle on the other hand...

Re:Am I the only one?

[241comp]

How about the ability to have dual video cards within a laptop? The low-power on-cpu video driver for when you're just browsing the web or whatnot and the latest Nvidia/ATI pci-e power-hog for when you're gaming... think of the battery life you could have when reading your email at the airport without sacrificing the ability to play your favorite games at full res.

Re:Am I the only one?

[atrizzah]

I think everyone's missing the point. Everyone's talking about how this isn't going to be the next big thing for gamers. Well that's obvious. However, I see this is as being huge for anything not built for gaming. The lowest end machines won't even need a slot for a graphics card. Laptops can do away with all of the bus hardware that would have been necessary to support to a graphics chip. I imagine the Fusion will probably have the power to let people run Vista without the necessity of an external graphics card, and that will be enough for probably 95% of PC's sold. Everyone else (i.e. everyone here on Slashdot) will stick with off-chip graphics. I think it's a brilliant move, since it's pretty clear the chip maker's are running out of ideas for what to do with those extra transistors as CMOS gets smaller and smaller. I wonder what the net effect on power consumption will be. If this move reduces overall power consumption, then that's yet another plus.

Re:Am I the only one?

[timeOday]

Software will always expand to fill all available CPU Cycles.

Demand for faster PCs has been falling since the late 90s. It was at that time that the average price of a new computer started falling from its historical (approx 15 year) average of $2500-$3000 down to around $1000 today, and still falling. You can still buy a $3000 PC today if you want to, but very few people do. Why did the average consumer start buying lower-end gear? Because it's good enough. Yet look at the latest gaming consoles, the latest generation is not much cheaper (even adjusted for inflation) than the last two or three generations, because a PS2 does not "solve" the problem of gaming in the way that a 2.4 GHz PC "solves" web surfing and Word.

Re:Am I the only one?

[Lord Crc]

Except GPU's are at least as high transistor count as CPU's these days and based on transistor count growth GPU's will exceed CPU's in the next one to two generations.

Current GPU's are already ahead of current CPU's:
Amd X2 1mbit cache: 233.2 million transistors
Intel Core 2 Quad: 582million transistors
Nvidia 8800 GTX: 681 million transistors

Re:Am I the only one?

[carl0ski]

With the GPU on the same physical Ceramic chip less power voltage is required to communicate, think of it as "How loud would you need to talk for someone 10 metres away to hear you?" "How loud would you need to be if they were standing next to you?" Being on the chipset will probably be favoured for a while but Fusion would lower the cost of manufacturing a mainboard will meet along with AMDs massive push for low power desktop PCs Fusion could also suit casual gamers since The Processor and Graphics cards tend to be out of date simultaneously in some cases. ala Doom3 Battlefield 2

Re:Am I the only one?

[carl0ski]

On the general Desktop Scene however things change
If people want the benefit of a GPU that has direct access to CPU and ram
yet need the oportunity Torrenza is the solution.
Any type of processor can be placed in an AMD Opteron socket including NVIDIA or ATI Graphics chips and have high speed low latency HyperTransport links to the CPU & memory.
The would offer higher communication bandwidth for Crossfire or SLI than PCIe x16 can ever offer.

This could potentially make the PCIe or standard peripheral slot obsolete.

Re:Am I the only one?

[mrchaotica]

A couple of questions for you to consider:

1. Which has more bandwidth, 16x PCIe or HyperTransport?
2. What's stopping AMD from simply putting a memory controller that can talk to 1GHz DDR3 on these chips (remember, AMD CPUs have on-die memory controllers)?

Also, keep in mind that we're talking about a technology in the early stages of development. The current performance of off-the-shelf hardware is irrelevant; the issue is whether the basic technology has the potential to be made to do it.

Re:Am I the only one?

[mrchaotica]

Maybe I'm just projecting, but what I think that you want, and many others out there want is simply a better bus, bridge, magic, glue or whatever you want to call it between the major parts of a computer.

Hmm... you might be right. But an equally important aspect is that I want a socketed GPU and graphics memory, so that it would be modular just like the CPU and system memory.

A combo CPU/GPU either only targets the very low end generic computer or a specialized graphics type of computer. With the failure of the other addons to computers, I don't see the advantage. Graphics simply don't matter in the server market.

You're thinking too small. Modern GPUs don't just do graphics; they are becoming able to do just about any kind of very-parallel computations.

For example, imagine a server of some kind that processes every packet it sends in the same way. Let's say it has 1000 connections, and that each one is handled by a separate thread. Now, imagine that the thread can somehow be implemented as a shader. Then, instead of processing 1 or 2 packets at a time (as on a single- or dual-core CPU), you can suddenly process (tens? hundreds?) of packets simultaneously! Even if the thing is clocked lower than a CPU, I'll betcha it'll still have better overall performance -- assuming you can make it work.

Personally, I think there are quite a few problems like this, that could be parallel even though they aren't usually implemented that way now. I think it's a matter of how most programmers are used to traditional CPUs, and so don't think in a parallel way. I'm actually banking on this becoming a big deal in the future, which is why I'm taking a bunch of graphics, systems and HPC classes (I'm a CS student). And it had better do so, because otherwise computers are going to stop getting faster -- the whole reason everybody's so focused on multi core systems now is that they've hit a wall wrt. the laws of physics.

Re:Am I the only one?

[mrchaotica]

What a silly question! You need 4 sockets so that you can have (8 cores * 4 sockets =) 32 processors!

Seriously, you need to turn in your geek (or indeed, man) card. You can always want MORE POWER!

Re:Am I the only one?

[mrchaotica]

Wow, man, stream-of-consciousness much? Let me introduce you to my little friend, the Paragraph . How's about you two get acquainted?

Anyway, now that I've finished deciphering your post, you make some good points. I agree that GPUs are becoming more generalized, and that they'll become more important for solving mainstream computing problems. I'm not so sure that the traditional CPU will be eliminated, though -- there are some problems that can't be parallelized, and CPUs will still run at higher clockspeeds than GPUs for the foreseeable future.

Re:Am I the only one?

[peawee03]

I dunno, I rather like the notion of a 4 socket 8 core machine (32 cores of awesome). Then again, I maintain a computational cluster. YMMV.

Re:Am I the only one?

[byteframe]

I think I made one entirlely whole, reasonaed, arguement. Please give me rep points.

Re:Am I the only one?

[Calinous]

Full speed access to the high-latency, low-bandwidth memory of the computer's RAM... Let's take into account the AM2 platform - dual DDR2-800 - or some 12.8GB/s total bandwidth. The 8800GTX - the latest and greatest video card - has 86GB/s memory bandwidth. Meanwhile, the 7600GT - which is a gaming budget card - has 23GB/s available. Integrating GPU and CPU is good, but not because of faster access to system memory.

Re:Am I the only one?

[WetCat]

by the way, what ISA is? An old 8-bit bus? Or what?

in topic: also if you add an additional graphical card, software may use builtin card for efficient counting anyway.

Re:Am I the only one?

[Calinous]

Trace length (resistance/capacity) would be one reason not to be able to have a high-bandwidth, low-latency connection (you could still have some kind of high-bandwidth, high-latency connection) Resistance in overall link between chips is greatly increased when using non-soldered contacts (so you would need soldered high-performance memory on mainboards, and not memory slots). For what I know, on the current AMD designs, the PCIe 16x connect to the processor through the HyperTransport bus As graphics performance in games increase linear with the memory bandwidth (if you have enough GPU processing power), no, you will reach the performance level of a NVidia 7300 (5.2GB/s memory bandwidth) with ease. With some improvements, you could reach maybe the 7600GT performance (which uses a 23GB/s memory bandwidth, while your AM2 has at most 12.8GB/s on DDR2-800). More than that? Maybe with something similar to Kyro and Kyro2, but your current/soon to be current designs simply don't have enough bandwidth to feed a powerful GPU

Re:Am I the only one?

[mrchaotica]

Whole and reasoned, yes, but also disordered and needlessly difficult to read. It was bad enough that I almost didn't bother at all. Now, I'm not saying this to insult you; I just figured that most people post because, you know, they want other people to read it. And your lack of organization is an impediment to that.

Re:Am I the only one?

[kosmosik]

How often do you change videocard or CPU in your laptop?

Re:Am I the only one?

[Sparr0]

Having GPU and CPU on the same card isnt just for 'new video card' stuff. It means we can do vector processing with what would normally be shader pipelines, in non-graphical apps. Like the Folding@Home GPU client, or nVidia's physics-on-GPU implementation. This is one step towards an architecture like the Cell, with many specialized intercommunicating cores on a single chip, instead of a single fast-but-generalized core.

Re:Am I the only one?

[smallfries]

As demonstrated below, (one "single-consistent idea" can still be broken into logical pieces:

Integrating CPU with the GPU? That would be cool, however in the long term I don't really see it happening, or most anything else mentioned here. Rather this 'fusion' of technology will most likely just end up as a new processor, based around a concept similiar to Cell, but probably without an emphasis on the differant 'processing units'. We'll probably end up with a new cpu design, hopefully x86 compatible, (although if they could create a really really good cpu, I mean the next big thing, they really shouldnt feel like they have to work with the x86 ISA), with all these smaller units (not cores) doing everything.

Think of the latest 8800 GTX, it has 128 unifed 'Streaming Processors',that were previously known as pixel shader units (or a variation thereof). Graphics cards also have/had vector shader units, and the ability to use geometry shaders, however these (at least for g80) are all unified now, in the sense that these 128 'Streaming Processors' (notice the name change AND it's significance) can on-the-fly change their function to either of the three categories. Cool idea for the GPU, but it also hints at where things are going. The Voodoo2 is more of a 'graphics processor' than g80, IMHO, in the sense that gpus are just essenestially becoming 128 'tiny'-cored general purpose coprocessors, that area really good at floating point math, (essential for graphics).

As far as today's graphics cards go we can run C code on these things. Sounds more like an auxillary, differant, type of CPU, that's all today's graphics cards are. check it out... http://developer.nvidia.com/object/cuda.html/ [nvidia.com] So the wheel of reincarnation has started. It began with GLQUAKE, with an addon card (that was an addon to your 2D card...) and will now end with either a CPU w/ GPU (fusion) or even better, a CPU that incorporates processing concepts from the GPU into a new design that is really really good, especially in the areas needed for Quake VI.

PPSh-41: Well still have pci-e express, and better still we'll probably have PCIe x16 slots too. Whether or not the consolidated market will still continue to sell graphics coprocessors is beyond my skill of intuition, however they wont stop selling them for at least 6 years. Another possible scenario is an Intel/NVIDIA buyout (I like nvidia, dont buy them out, in fact just establish a partnership). In that case, I'd venture to say that with each cpu/gpu maker in their own business couplings the GPU as we know it (as of right now) is dead.

Remember what tim sweeney said, "something something blah blah blah, what we'll find is that [graphics] is a problem that turns out was just general computing". That definatly seems like the case, when you consider the ps3's graphics (which are probably good, havent checked it out, not that I'd buy a console). If a new (more than just an extention to the x86 spec) cpu comes out of all this technology, I really really doubt well be using a GPU, as it'd be kinda like just having another CPU, but in a pci slot, with a two slot cooler, loud noise, more cords, bigger cases, more heat etc etc. Provided the new cpu specification can be used buy both companies to make compatible cpus, well all see the extinction of the GPU, as well as numerous benefits mentioned a second ago. So bottom line, and in my best hopes, a Slashdot reading Dual Core, Dual GPU fat person, will eventually have a new box in 2012, right before the world ends. It will be running GNU/Linux), and will no more than the size of a mid tower. However the CPU in the system will be a really really fast, and will have hundreds of the 'streaming processors' (that are the basis of out latest gpu's processing ability) that can be used for games, graphics, or anything else, or even porn.

Seriously though, you did just hit HTML formatted by accident. Come on you're amongst men now, you can admit it bravely...

Re:Am I the only one?

[name*censored*]

With the constant changing of slot sets (well, maybe GPU slightly less than CPU), eg AMD754->939->AM2->AM3, and AGP->PCI-e, you'd be better off to upgrade your motherboard+cpu+gpu all in one go, since this won't force you to try and squeeze as much out of an older slot set (eg, instead of upgrading your old AGP GF-5400 or R-8550 to an AGP GF-6800 or AGP R-X800 in 6 months, upgrade to a PCIe GF-7900 or PCIe R-X1900 in 12 months). I for one welcome the new fusion technology (overlords), since it'll lower CPUCPU latency and it makes upgrading less of a balancing act (since GPU and CPU power requirements from games more or less scale - that is to say, you'd be hard pressed to find a game that requires a powerful GPU but a weak CPU or visa-versa) budget-wise.

Re:Am I the only one?

[GNious]

Thank you for this usefull information.

Re:Am I the only one?

[clacke]

I'm a nerd, but not a gamer. I just want a reasonably fast thing to compile my thingies on and maybe play something fun once on a while. I like something with 3D capabilities, but it doesn't have to be the latest ATItrox G4FX80. I buy a computer every five years, then keep it until it becomes unbearably slow due to spec rot, i.e. considering the new advances in using up system resources.

"Upgrade" to me means buying new motherboard, CPU and RAM. I'll get a motherboard with integrated audio and graphics, to not crowd my box with unnecessary cards, to not have to choose. They're also cheaper and smaller integrated. Now, if the GPU were integrated into the CPU this would probably mean less chip area, less power consumption and generally good things, so I guess I'm the target market here.

Re:Am I the only one?

[somersault]

Have you never heard of integrated graphics? Some of the crappier computers here don't have AGP slots for graphics. They still have PCI slots, but just for expandability's sake, not for graphics, sound or networking.

I'm hoping that this will spur some kind of move to using faster RAM for processors as well as just graphics cards.

Re:Am I the only one?

[jhfry]

I have made a similar argument, I want a FPGA on the HT bus. A Fully Programmable Gate Array would allow software developers to create hardware "programs" to accelerate their products. For example, Photoshop and Premiere could re-program the FPGA to accelerate processor intensive tasks. Codecs could be implemented on the chip for realtime encoding (for example http://www3.elphel.com/en/products is a camera that does OGG Theora in a FFPGA in realtime at 640x480x90fps). And game writers could optimize any part of their applications they wish. Sure FPGA's are not as fast as a non-programmable processor, but they are infinately more flexable, and if it's sitting on the HT bus with direct access to system RAM, you would see some amazing uses. Accelerated encryption/decryption, email filtering and virus detection, or any other process that is sequential in nature and done repeatedly.

Re:Am I the only one?

[mrchaotica]

Good idea! I wonder, would the development of such a chip have to involve AMD, or could a third party do it?

By the way, the F in FPGA stands for "Field," not "Fully."

Re:Am I the only one?

[ruiner13]

It is still way better than the current shared video memory cards which still access the main system RAM but have to do so through the PCI-E X16 interface. Having it on the CPU would be WAY faster.

Re:Am I the only one?

[Hatta]

Don't be silly, everyone knows PCI is better than ISA.

Re:Am I the only one?

[LWATCDR]

Isn't even the term FPGA kind of old school these days?
I thought they where called PLDs for Programmable Logic Devices.
In embedded devices this is pretty common. I have seen a few PPC embedded chips that had an PLD right on the chip.
Logicaly that would be an interesting next step. Don't hang it the PLD on the a Hyperchannel connection but put it right on the core.
That would probably be the better solution since if you make it an option few people will support it. If you make it standard then support for it will be common.

Re:Am I the only one?

[Calinous]

AMD's implementation is even worse than that - the video card is integrated in northbridge, and access the memory thru the processor (on Intel, the video card is in the northbridge, just like the memory controller - one could easily make a 256-bits connection between the two) By the way, in the Intel's current processors, memory access would be slower if the GPU would be on the CPU (compared to the current situation)

Re:Am I the only one?

[Afrosheen]

I think this is another one of those 'just for the OEMs' stunts. Like if Abit wants to make a board with onboard video..well...you just drop one of these CPUs in and there you go. They can tier them as well, so the cheaper AMD64x2 3200+ will get the equivalent of a 7600 or other low-end chipset. However, I don't know how this will affect heat dissipation/generation. Seems like having the two hottest chips on the board integrated into the same die is kind of bad.

Re:Am I the only one?

[WhiteWolf666]

Hear Hear!

I fit into the category of, "Too difficult to read, so I won't bother."

Your message is lost on me.

Re:Am I the only one?

[zacronos]

So, you want CPUs designed to do many things in parallel, much the way modern GPUs do? Why does it need to be a GPU then (optimized for graphics functions)? We could have Serial Processing Units (SPUs) that are like today's CPUs, Parallel Processing Units (PPUs) that are like today's GPUs but more general-purpose, GPUs, and things in-between (like the AMD Fusion, which would presumably be somewhere in between all three of those). Then I would agree with you in a heartbeat; depending on the system needs, you could set up any combination of SPUs, PPUs, and/or GPUs -- that would be really cool.

Re:Am I the only one?

[644bd346996]

Anyone in the HyperTransport consortium could produce such a device. They might not be able to use AMD's sockets, but there is already the HTX slot for this stuff. Take a look at the membership list for the consortium.

Re:Am I the only one?

[Sparohok]

I've heard arguments like that over and over again concerning functionality that is now implemented in software, sometimes with new dedicated CPU facilities. Graphics will be no different, it is just a matter of time. It may start with "typical" business users and laptop hardware, but sooner rather than later, the architectural and cost advantages of integration and generality will always outweigh putting huge amounts of custom logic on the wrong side of an IO bus.

The handwriting is on the wall if you know where to look. CPU and graphics architectures are converging.

First the CPU side. More and more of the work being done on desktop CPUs is media related: HDTV and DIVX encoding/decoding, digital photography, audio DSP, game physics, software radio. General purpose superscalar CPUs are horribly inefficient for this class of problems. At the same time, new chip processes are providing more transistors than CPU designers know what to do with. At first they just added more cores, but that is already running into scaling problems. They're clearly tempted to add some sort of highly parallel SIMD engine that is better suited for DSP type problems. Look at MMX in all its guises. Look at the Cell architecture. Both are steps in this direction.

Then the graphics side. The clear direction here is away from specialized hardware for each stage of the graphics pipeline, and toward large numbers of more general computation elements in a SIMD architecture. Look at NVidia's latest chip and you'll see something that is completely unlike traditional graphics pipelines. However it's almost exactly like what CPU designers want to put in desktop chips anyway, to solve the problems that superscalar processors are bad at.

Finally, look at some of the other stuff we are putting on the wrong side of the IO bus: Creative Labs X-Fi, AGEIA PhysX. These are also examples of convergence toward highly parallel and relatively general DSP architectures.

Even if power gamers like to upgrade their graphics card every few months, there are other markets which are increasingly driving system architecture. Game consoles are a huge market, and the next generation of consoles is poised to take advantage of the convergence I'm talking about. Vista is a big market for 3d graphics on business PCs, and that market is all about higher integration and lower cost. Neither of these markets will ever upgrade graphics independent of CPU, but they are paying for super high bandwidth IO busses and duplicate memory interfaces which they don't need.

Graphics hardware on an IO bus has about 5 years to live, in my opinion.

Re:Am I the only one?

[farnz]

Already exists, and there's even choice in the market. Admittedly, it's currently Opteron-only, but it could end up trickling down to 4x4.

Re:Am I the only one?

[afidel]

Wow, the 8800 is a beast! The 7950GTX is only 278 million transistors.

Re:Am I the only one?

[jhfry]

Thanks for the link... mmm I wanna get my grubby hands on one of those!

Re:Am I the only one?

[mrchaotica]

Why does it need to be a GPU then (optimized for graphics functions)? We could have... Parallel Processing Units (PPUs) that are like today's GPUs but more general-purpose.

The goal of the computer graphics industry is perfect realism, right? Well, reality is complex. To model it, graphics chips are going to have to be able to handle more complex stuff as well, which means becoming more flexible and generalized (we're already seeing this with shaders, for example, and the Nvidia 8800 is reportedly extremely generalized -- they made a C compiler for it!). In other words, "optimized for graphics functions" and "general-purpose" are (or will be) the same thing. The distinction is meaningless.

Re:Am I the only one?

[jhfry]

The only reason that I can see to not put one on the core is that it would break x86 spec. Though it looks like AMD is considering doing this anyway with their new GPU/CPU combo mentioned today.

As far as FPGA (a type of PLD that is "field programmable" meaning it can be reprogrammed while installed rather than by using a separate programmer) technology being oldschool, I would say it is not.

I believe that they are primarily used in hardware design however, and very few have been used in consumer products as it is cheaper to mass produce a dedicated chip. That and I could argue that hardware manufacturers would rather sell you the next version of the hardware than let you simply download an update for your FPGA.

Considering that an FPGA can be programmed to do specialized tasks at speeds that can be as much as 100X faster than a general purpose CPU, I think it is inevitable that they will eventually become mainstream in some form or another... the HT bus is ideal for such an implementation.

Re:Am I the only one?

[LWATCDR]

Actually I was saying the term FPGA was old school. I keep seeing them called PLDs.
PLDs, CPLDs, and FPGAs are all very interesting and I would even say exciting. I just think that sticking them on the core is an even better idea than on the HT buss.

Re:Am I the only one?

[obeythefist]

And, no matter how fast your integrated memory controller is, you're splitting that memory between CPU and GPU (remember, if you didn't want to share the memory, why would you put the GPU on the same socket as the CPU, why not have it in a seperate socket, or, haha, as a PCIe card?). Sure, you get the benefit of having the CPU shove data directly to the GPU without using the heavily saturated (not) PCIe bus.

It would be highly inefficient to shove DDR3 memory at a CPU - benchmarks consistently demonstrate that additional memory throughput beyond DDR2800 makes little to no difference in performance of a contemporary CPU. Going from DDR2-667 to DDR2-800 is barely noticeable. Why throw even more money at the CPU so you can have a GPU that performs close to the speed of a self-sufficient graphics card?

No no, it is evident this solution is meant for the low end markets. While performance parts make the prestige of a company, it's the volume sales that are the bread and butter. This is why AMD consistently paper launches their high end video cards, there's no sense in producing them at high volume, you just need to make enough for the benchmarking sites to record you as the maker of the fastest GPU, then you get the prestige, then you sell the midrange parts in volume.

Re:Am I the only one?

[Mr2001]

Buuuutt.... I guess you would need to compile two versions of the same thing and put it on the same game disk, or figure out some kind of neat system so that translations are done in real time with hardware (much faster than the soft approach).

Two (or four) words: just-in-time compilation.

Put a metadata tag on each graphics API function that says "this function can be implemented as an opcode". Then, when the game is JIT compiled (at install time, load time, or on demand), invocations of those functions are compiled as single graphics opcodes if the hardware supports it, or library calls otherwise.

Of course, you have to write for a JIT-capable platform to make that work, but there are several reasons why you might want to do that anyway (besides the one or two reasons why you wouldn't).

Yeah that's the future

[Rosco P. Coltrane]

ISA is definitely the future to interface a CPU and a GPU, but I keep hearing about this VLB technology that's even hotter!

Re:Yeah that's the future

[From A Far Away Land]

I was thinking the same thing. I can't wait for them to put a new TLA architecture into my MAC or IBM. Because ISA is totally the VLB of the 2000s. It may even outcool USB and PCI-e.

Re:Yeah that's the future

[Anonymous Coward]

I'll tell you where the future really lies. A PCI-E CGA Card, now that would really bring in those framerates on the games. ;)

Re:Yeah that's the future

[MadEE]

Bah You haven't seen the new ISA EXTREME!

Re:Yeah that's the future

[misleb]

My money is on microchannel. Nobody ever lost money betting on IBM.

-matthew

Re:Yeah that's the future

[LoRdTAW]

VLB.... Bah. Im sticking with EISA.

Re:Yeah that's the future

[Dunbal]

A PCI-E CGA Card, now that would really bring in those framerates

    10 zillion frames per second, but only 4 colours. Oh my eyes!!!

Re:Yeah that's the future

[Anonymous Coward]

I bought IBM stock in the 90's, and let me tell you, I sure didn't lose any money. In fact, I just sold a part of it and bought myself one of these.

brand new I/O model

[Anonymous Coward]

the new string technology is getting hotter, maybe one should go out more often *g*

m10

How long until a physics extension?

[User 956]

'To support CPU/GPU integration at either level of complexity (i.e. the modular core level or something deeper), AMD has already stated that they'll need to add a graphics-specific extension to the x86 ISA.

x86 is a great multi-purpose, but the reason we're seeing greater and greater offload onto a GPU is because that's great at a specific task. So my question is, how long until we see widespread PPU (Physics processing unit) usage, and beyond that, a Physics extension to the x86 ISA? Or will we all just be computing on the grid at that point?

Re:How long until a physics extension?

[GmAz]

There is a PPU available. http://www.ageia.com/ . The problem is that you have to use a specific physics engine in the game to utilize it. And the problem with that is that no company wants to be forced to use a specific physics engine in their game. Some make their own. When this card came out, many games actually ran slower because the physics card was trying to take over and wasn't working well with the physics engine the game was using. On the other hand, the games that did run the right physics engine ran about the same with a small margin of improvement. If the card could be like a video card and be somewhat programable (per pixel shading) by each physics engine utilizing it, it would be much more useful.

Re:How long until a physics extension?

[Wesley Felter]

So my question is, how long until we see widespread PPU (Physics processing unit) usage, and beyond that, a Physics extension to the x86 ISA?

Never, since it looks like physics can efficiently run on GPUs now.

Re:How long until a physics extension?

[volsung]

When I first heard about the PhysX, I was very excited because I thought there might be a way to use the card as a coprocessor in actual physics simulations. Much to my dismay, the PhysX website was totally focused on games, only offered a very-high level Windows SDK which cost tens of thousands of dollars unless your software was a mass-market game, and had no useful information about the specific hardware capabilities of their chip (single precision? double precision? instructions set?). I emailed them to ask if they had any plans to investigate the scientific computing market, and got no response at all, which I guess was all the answer I needed. A Windows-only black box is worse than useless for us, so I have moved on to investigating the general-purpose use of GPUs instead. ATI and Nvidia are far more forthcoming with information, although only slightly. Getting accessed to ATI's GPGPU libraries for Linux required several attempts at getting a researcher account before they finally replied.

Re:How long until a physics extension?

[Anonymous Coward]

ATI and Nvidia are far more forthcoming with information, although only slightly.

Slightly far more forthcoming? Interesting.

Re:How long until a physics extension?

[Anonymous Coward]

x86 is a great multi-purpose

Ahahhahahahaha!

I'm sorry, what were you saying?

Check out AMD's wrongdoing!

[Anonymous Coward]

Here: http://malfy.org/

One unanswered question?

[Chabil Ha']

Can it run Linux? OK JUST KIDDING!

No, the real hting lingering in mind is: Why? If I want to upgrade the GPU, now I have an additional cost because I have to upgrade the CPU as well and vice/versa. So, from economical perspective, is this the best way to go?

Re:One unanswered question?

[zpapasmurf]

no, but SLI isn't economical either. it's not the most bang for the buck, its the most bang.... period

Re:One unanswered question?

[Anonymous Coward]

Can it run Linux? OK JUST KIDDING!

Why joke? It is an important question.

All the current nvidia and ati graphics cards require proprietary, closed-source drivers.
If the GPU is to be integrated into the CPU, either they will have to keep the new ISA a secret or we will finally start getting access to the information required to really write Free graphics drivers.

Re:One unanswered question?

[Anonymous Coward]

Pardon good sir, but how often are you upgrading your numerical co-processor? I mean you could just as well have written "If I want to upgrade the FPU, now I have an additional cost because I have to upgrade the CPU as well and vice/versa. So, from economical perspective, is this the best way to go", if this had been in the early 486 days..

Re:One unanswered question?

[Anonymous Coward]

No, the real hting lingering in mind is: Why? If I want to upgrade the GPU, now I have an additional cost because I have to upgrade the CPU as well and vice/versa. So, from economical perspective, is this the best way to go?

RTFA you dumbass. This is most likely for laptops. How often do you upgrade your laptop GPU?

Re:One unanswered question?

[Anonymous Coward]

I think this could also be good for desktops. I have seen quite a few programs use the CPU for graphics operations, with this intigrated GPU, they would probably be able to make use of it, speeding up the avarage desktop draw times (maybe now we can have animated desktops without real performence trouble?), also maybe even video players, and other minor graphics tasks.

The grandparent might well be asking "how am i supposed to upgrade my onboard video? I think its rather unecenomical to replace the whole motherboard", obviously, this is just a CPU extention, nothing more, you can still talk to the same hardware, meaning you can still use your video cards, only now, certian well written apps can take advantage of another graphics processor for small, low latency requiring operations.

Re:One unanswered question?

[nomel]

Does it run linux? Well, with the processing power of some of the GPU cards used for things like SETI@Home, I imagine that's not all that far fetched. Could use the GPU as a second processor perhaps! :D

Re:One unanswered question?

[Anonymous Coward]

ATI is owned by AMD.

Re:One unanswered question?

[Evil Shabazz]

You pay for your graphics drivers?

Re:One unanswered question?

[jawtheshark]

Don't laugh... Drivers weren't always free as in beer, and perhaps sometimes they still aren't (but I wouldn't know). We payed the Iomega Drivers for OS/2 to run our Bernouilli Box. It's a long time ago, but paying drivers did exist. (Perhaps it was just shipping, I don't know... I was a teenager, and it was my dads computer)

compatibility nightmare

[Anonymous Coward]

Ok, add the GPU part .. call the ISA

x86-64-G

Now what happens when they need to add physics ..

x86-64-G-P

Now what happens when they need to add extra multimedia capability..

x86-64-G-P-M

Ok, maybe we need more graphics .. the old ISA is obsolete

x86-64-2G-P-M

etc.

So it'll be a nightmare when you buy some software to make sure you have the right processor. Furthermore there'll be a silly AMD vs. Intel war on instruction sets all over again.

Well .. I just hope it ends up being cheap for the consumer (though I doubt it .. more like forced upgrade).

Re:compatibility nightmare

[Wesley Felter]

That mess already exists: MMX, 3DNow!, SSE, SSE2, SSE3, SSSE3, SSE4... Yet software just works.

Re:compatibility nightmare

[dadragon]

That's because Microsoft Word doesn't use MMX, 3DNow!, SSE, SSE2, SSE3, SSE4...

The thing is that most general purpose computers execute the SIMD instructions somewhere between rarely and never.

Re:compatibility nightmare

[Wesley Felter]

And any new "fusion" instructions will work the same way, so it's not a problem.

Could ISA in this case mean . . .

[Anonymous Coward]

Could ISA in this case mean Instruction Set Architecture?

Just wondering what this latest TLA (Three Letter Acronym) means.

Re:Could ISA in this case mean . . .

[Grishnakh]

Whoosh!!!

Yes, ISA means Instruction Set Architecture. It also means Industry Standard Architecture, in reference to the s--l--o--w 8- and 16-bit bus in the early 8088 and 286 PCs, which was used for quite a long time to connect to video cards until it was replaced by faster buses such as MCA, EISA, VLB, and finally PCI.

This is what's known as a joke.

Re:Could ISA in this case mean . . .

[Sj0]

Correction: This is what is known as a shitty joke.

For future reference, a good joke would subtly poke fun at the situation or some fundamental truism or commonality among people, not just point out that the word does, in fact, have two different meanings.

Re:Could ISA in this case mean . . .

[Anonymous Coward]

Actually it's a good joke, because it's pointing out that the article summary should have mentioned which ISA it was talking about, for the benefit of those people who didn't know what ISA meant in a CPU context.

Re:Could ISA in this case mean . . .

[Anonymous Coward]

Replaced? There are millions of industrial PCs which still use ISA and do not look forward to its replacement any time soon.

Re:Could ISA in this case mean . . .

[Grishnakh]

I hope they're not still making them then. An embedded version of PCI has been available for ages, and ISA is a horrible bus which really shouldn't be used any more.

Advantages?

[Tainek]

A lot of people seem to be having issues working out why AMD is doing this

people are forgetting that they are not always the target market for computers (this isnt aimed at you if you upgrade one more than the other)

for example, what is easyer for your computer illiterate father to do, change one slot component, or install a graphics card , and a cpu.

it also allows for even smaller form computers

i will concede, that these gains are pretty small though, i cant see it being worth it

Re:Advantages?

[fatty ding dong]

people are forgetting that they are not always the target market for computers

Until this was posted, I couldn't figure out the "why" to this problem. The "why" is indeed, pre-built home systems.

Think about your average Joe that doesn't know a USB from USPS. He's not going to concern himself with more statistics than he has to when it comes to buying a computer. If he can get his CPU and his graphics rolled into one component at a lower cost than having them separate, he will without a thought. It doesn't even come down to ease of repair since most average users will have someone else repairing their system. It will simply come down to cost.

I've had the same home built box for 4 years, I just upgrade as needed (and sometimes as wanted). Average pre-built buyer doesn't care for upgrading. As soon as it breaks: fix it. If it costs too much to fix it: replace it. The word "upgrade" isn't in an average computer purchaser's vocabulary. If the Fusion can last even a year and a half on average, that would be adequate given that most pre-built buyers are conditioned to buy a new box every 2 years give or take. For average computer buyers, it will work as long as AMD can keep the cost down. I will agree however, its definitely not practical for those of us who buy new video cards almost as often as we buy milk.

Re:Advantages?

[Barny]

Not to mention, mobile phones, house phones, hdtvs, blu-ray/hd-dvd players... just about anything that needs a bit of cpu grunt and must have a display output would benifit from integrateing a good cpu and graphics.

Imagine getting a nice, soft edged, menu effect enabled interface on your dvd player instead of the |> || |>|> symbols flashing every time we push a button?

Re:Advantages?

[fermion]

I also think people forget that chips are often separate because of technical limitations. For that matter we used to have individual transistors instead of integrated circuits.

In the near past, things like the FPU, the GPU, and even the PMMU had to be fabricated separately and then connected with expensive and inefficient busses. This made the computer slower, and, as you mentioned, bigger. However, offloading work to these ancillary chips, even though there were inefficiencies, made the overall computer much faster, and made things like a GUI usable. Recall that computers that depends on the CPU for graphics require a much heftier CPU.

Finally, there is the issue of reliability. One huge issue in the early computers were the sheer number of components that had to work together. For example, it took 8-16 slots to achieve the memory that now takes 2-4 slots. We may complain about lack of expandability, but the reliability to price ratio is phenomenal. I recall reading that in manufacturing the cheap computers, Dell and HP will leave off features for a 50 cent savings. Sound small but it is big for them.

Ultimately, complaining about this is like complaining that the computer does not a separate floating point unit. Why should we not be able to upgrade the floating point capability of our computer without replacing the entire CPU. Sounds like a conspiracy to me.

Showing my age

[WidescreenFreak]

I guess I'm showing my age. As soon as I saw "ISA" I immediately thought, "Why the HELL are they thinking about bringing this back?

:(

Re:Showing my age

[supaneko]

*LOL* You are not alone. I was beginning to wonder the same thing...!

Re:Showing my age

[njchick]

It's not your age. It's just a problem of the current TLA namespace. Another reason to switch to XTLA (extended three letter acronyms).

Re:Showing my age

[david.given]

Another reason to switch to XTLA (extended three letter acronyms).

Nah, expanded TLAs are far more flexible. They allow you to have more than one expansion at the same time from different manufacturers and don't require you to upgrade to a new version of English to use them...

Re:Showing my age

[Thundersnatch]

Rumor has it the Chinese and South Koreans are well on their way to deploying TLAv6 nation-wide. TLAv6 enables enough acronyms to assign one to every atom in the universe. As a result, there exists the possibilty of a "fractured" acronym space in the near future, one in which American and European acronym users cannot excahnge information with Asian acronym users. The US and Europe are destined to lose acronym-industry leadership if they do not act swiftly!

Hm, an ISA extension for 3D graphics...

[wirelessbuzzers]

... maybe they should call it 3DNow or something?

Re:Hm, an ISA extension for 3D graphics...

[Lord Kano]

Or maybe they could creat a set of multimedia extensions.

They could call it MMX.

LK

What happened...

[dduardo]

What happened to the RISC philosophy? Keep the hardware simple and let the compiler do the work.

No, lets create 1000 more instructions for graphics, 1000 for physics and 1000 more just for the heck of it.

Re:What happened...

[MadEE]

That philosophy pretty much went out the window when we started seeing consumer 3D video accelerators hitting the market.

Re:What happened...

[shofutex]

x86 never was RISC. The last bastion of RISC processing in consumer PCs was the PowerPC.

Re:What happened...

[forkazoo]

What happened to the RISC philosophy?

We decided we wanted cheap, fast hardware, and we decided the philosophy made more sense at the software level.

Re:What happened...

[cyngus]

BZZZZZZ! WRONG! Processors from Intel and AMD both execute in a RISC-like fashion in their core. They crack x86 instructions into smaller "micro-ops" (what Intel calls them) and then execute these. What the programmer sees may be CISC, but what happens inside is RISC.

Re:What happened...

[Guy Harris]

What the programmer sees may be CISC, but what happens inside is RISC.

The article is discussing what the programmer sees, i.e. the instruction set architecture.

Re:What happened...

[Nahor]

What happened to the RISC philosophy?

It's a bit more complicated than that. The x86 at its very core is RISC: it converts all the CISC instructions into RISC like micro-instructions.

But then, I read somewhere that Intel is now starting to do the opposite again with its "Wide Dynamic Execution": it combines several micro-instructions into one macro-instruction and also combines CISC instructions into even bigger ones.

Re:What happened...

[keeboo]

>> What happened to the RISC philosophy?
> We decided we wanted cheap, fast hardware, and we decided the philosophy made more sense at the software level.

Uh... More likely you folks have decided you want to run DOS and Windows.
Since both were (are) locked to the x86 ISA, it gave this decrepit architecture a reason to live.

Re:What happened...

[Grishnakh]

Because doing graphics with a general-purpose CPU is a stupid idea. You'll always get better performance out of special-purpose hardware than general-purpose CPUs. The only reason to use these is for flexibility. When your requirements demand performance impossible with only software, you go to specialized hardware.

Which would you prefer, a super-powerful CPU that does all the graphics and physics, but operates at 20 GHz and consumes 3000 watts, or a low-power system with separate CPU, GPU, and PPU (even if they're just separate cores on a single die) which consumes less than 100 W overall?

Re:What happened...

[Sj0]

That decrepit arcitecture is the fasteest consumer hardware platform in existance.

Re:What happened...

[Anonymous Coward]

A major non-DOS, non-Windows OS just switched from a supposed RISK architecture to the x86/x64 architecture. I say supposed RISK, because there hasn't been a commercially successful RISK design in a long time, only slightly less CISCy designs. RISK suffers from the inherent flaw that the required addressing information does not scale down with instruction word size. That wasn't a problem when memory was fast enough that it was possible to feed the CPU new instructions when it needed them. Current architectures have to go to great lengths to keep the CPU busy. There's no memory bandwidth left to waste on huge addressing information for tiny instruction units.

Re:What happened...

[keeboo]

That decrepit arcitecture is the fasteest consumer hardware platform in existance.

I agree.
If Volkswagen had a massive consumer base to justify producing an old-style Beetle model capable of reaching 400km/h, you bet they would so.

But you see, despiste such an interesting market for x86 processors, how many companies are able to invest massive amounts of money in order to make a x86 speedy processor?
Compare this to the development money spent in UltraSparc T1. Geez, I bet those chinese guys spent a nickel to develop that MIPS-based Godson processor.
My point is: we could have much better and cheaper processors weren't we tied to the x86 architecture.

Re:What happened...

[EvanED]

RISK

It's RISC! It's not a reduced instruction set komputer!

Re:What happened...

[Dunbal]

an old-style Beetle model capable of reaching 400km/h

      Suddenly I want one of those...

Re:What happened...

[Sloppy]

What happened to the RISC philosophy?

People bought a metric shitload of x86 boxes instead, and anyone trying to sell RISC (or anything else that wasn't x86) faded into obscurity (at least in terms of personal computers).

Re:What happened...

[Anonymous Coward]

i think its because were reaching the limit in clockspeeds, cant afford to waste cycles retrieving adding and storing and retrieving and adding and storing, when it can all be executed at once u know.

the manufacturing process is what really killed the POWER processor, i think the last iteration was still being made @ 170nm. imagine a 32 core POWER4 on a single die, that doesnt seem to infeasible. + it would run linux.

Re:What happened...

[Anonymous Coward]

That's what you get from trying to run so many RISK games at the same time.

So... is it my turn yet?

Re:What happened...

[forkazoo]

Uh... More likely you folks have decided you want to run DOS and Windows.
Since both were (are) locked to the x86 ISA, it gave this decrepit architecture a reason to live.

Personally, I have no particular ties to x86 or DOS or Windows. I wrote my previous post from an Ubuntu box, and I am writing this one from a PPC box. But, all that "decrepitness" that makes x86 unclean is actually pretty damned useful. The wacky instruction encoding is horrible to look at, but also means that you generally see better code density on x86 than you do on a more pure RISC architecture. RISC came at a time when instruction decoders were a really significant part of a CPU. Now, with increased transistor budgets, on a high performance CPU the decoder is a non-issue. Making the decoder simpler wouldn't get you any benefit, and it would reduce the effectiveness of your instruction bandwidth and instruction caches.

I'm no x86 evangelist. My main personal server is an Alpha, I love my MIPS hardware and I even have a VAX. But, x86 hardware can't be beat for cheap speed. Not with anything currently out, anyhow. If somebody comes out with something that is elegant, cheap, and beats x86 for my typical workloads, I'll jump on board in a heart beat. :)

Re:What happened...

[Anonymous Coward]

" Keep the hardware simple and let the compiler do the work. "

Hah! It was a nice idea back in the 1980s when it looked like compiler technology was going to advance. Since it hasn't, though -- compiling optimized general purpose code for even simple instruction sets is a tough problem -- RISC has died a well-deserved death in the marketplace.

Itanium had a pretty simple instruction set, but half of the reason it failed was because nobody -- not even Intel -- could write a decent compiler for it. (The other half of the reason it failed is because it was a very, very slow part for the price point Intel wanted to sell it at).

Re:What happened...

[tolonuga]

It never worked as well as CISC, even though lots and lots of people
tried to get it to work as well. So currently: CISC won. Until the
next guy comes aroung with an idea for a much better RISC processor
(and maybe fails like all of them before).

Re:What happened...

[Sique]

But Volkswagen has a 400 km/h Beetle... Ok. They call it differently.

Re:What happened...

[Andrewm1986]

Power PC Apple Macs faded into obscurity?

Re:What happened...

[Anonymous Coward]

EPIC wasnt a "pretty simple instruction set". It was a complicated, black magic boondoggle, built out of simplistic instructions. The bundling and binding of the simplistic instructions into execution words was the trick no one could do. ... I think there's a fair argument for calling it super-CISC; as well as an abject failure and a shining example of why you should feed your design teams only high quality drugs.

Re:What happened...

[Sj0]

I've always gotten the feeling that people who disparaged x86 would have us leapfrogging from one platform to the next every few years, chasing the "Next really cool thing".

That's really nice in theory, but in practice, switching processor architectures every few years is a pain in the ass, as shown by the Apple folks.

Re:What happened...

[ball-lightning]

Yes.

Not aimed at you..

[Vellmont]

Either I change video cards more often than CPU's or CPU's more than graphics cards, but in either case I seldom want to upgrade both at the same time.

I'm really guessing that anyone looking for high-performance 3d acceleration isn't going to be the target for this product. Video cards get a lot of high-performance by using insanely fast memory. My guess is this design would use the system memory just like integrated graphics controllers do now.

I'll venture that this GPU/CPU integration is really aimed at the low end market to cheaply increase graphics performance for Vista. The integrated graphics chips that exist now are really just 2d chips, and have little or none of the acceleration that Vista wants for all its eye-candy.

It's also possible that you could use both the Fusion processor and your graphics accelerated card at the same time (though I kind of doubt any game or graphics-subsystem is going to support that).

Re:Not aimed at you..

[mrchaotica]

It's also possible that you could use both the Fusion processor and your graphics accelerated card at the same time (though I kind of doubt any game or graphics-subsystem is going to support that).

The integrated processor would be good for any other parallel computations you happen to want to do, though, such as physics.

Re:ISA?

[MadEE]

ISA = Instruction Set Architecture

Re:ISA?

[lagfest]

I don't think it means, what you think it means.

That depends on how it's done.

[jd]

Current CPUs are 64-bit. Current GPUs go as high as 256-bit. Dunno about you, but if someone offered me a full 256-bit multi-threaded multi-core CPU, I, sure as hell, won't be updating it for a few years. (Yeah, yeah, I doubt that's what AMD are planning, but it would be truly cool if they did. Or hot. Or is that the heatsink?)

Re:That depends on how it's done.

[TheRaven64]

When a CPU is described as '64-bit' it means one (and usually both) of the following things:

1. It supports 64-bit pointers, allowing it to address 2^64 bytes of memory (typically less in reality, since no one is likely to need the full 64 bits of address space for a few years).
2. It has 64-bit registers, so it can operate on 64-bit values (integers, usually, since every chip has supported 64-bit floating point values for years) directly, rather than splitting them into two 32-bit values.

Most GPUs only operate on 32-bit floating point values, although the newer ones support double-precision (64-bit) floating point values. GPU instructions typically operate on vectors; fixed-size groups of individual values, and these may be 128 or 256-bits wide, but you wouldn't call the Pentium 3 or the G4 a 128-bit chip just because it supports 128-bit vectors.

Re:That depends on how it's done.

[Dunbal]

Current CPUs are 64-bit. Current GPUs go as high as 256-bit.

      And most people are stuck with a 32 bit OS. Talk about irony.

So?

[Anonymous Coward]

So what? does it reduce my energy bills? does it put food on my table?
What are the long term goals here - Bigger! Better! Faster!? Is that it? Is that what we're all about? Let's get some relevancy in please.

Re:ISA?

[Anonymous Coward]

ISA means Instruction Set Architecture or if you prefer the definition of the x86 'assembly' language.

Tried and true methods, young grasshopper!

[Bananatree3]

Perhaps AMD should start from scratch rather than try to cobble together something based on decades-old technology.

NASA, one of the most technologically advanced space programs in the world, has decided to use a "decades-old technology" for the new 21st century push for Mars rather than build a new, from scratch system. Tried and true methods must be the best, then. If this technique of using "old school" technologies works for NASA, it will work just as well for AMD!

Re:Tried and true methods, young grasshopper!

[Tainek]

NASA is using old technology because it is tried and tested, and thus Safe

even if a new system would work better, one flaw can cost millions in money, millenia of combined work time, or even worse, human lives

AMD Could afford a few flaws slipping through, and just make a few revisions for the 1.5, NASA cant

Good for portable gaming?

[Josh Lindenmuth]

This seems like the perfect solution for driving desktop capabilities to smaller devices (such as a PDA or portable game system). I realize they'd need to work out heat dissipation, but a low wattage fusion device seems like a viable solution for enabling quality 3d graphics on the next gen PSP or similar.

Re:Good for portable gaming?

[MadEE]

I guess it falls down to what you consider portable. Laptops, definitely high performance ones. However PDAs I really doubt. The X86 architecture has very little bang for the buck in terms of power dissipation to performance and are very rarely used for PDA applications. I don't see that changing any time soon.

Re:Good for portable gaming?

[Dunbal]

realize they'd need to work out heat dissipation,

      The new PS4. It's not just a game console, it will also make your cappuccino!

Re:ISA?

[daemon_mf]

I'm not sure if you're kidding here, but I'm pretty sure ISA here means Instruction Set Architecture and not Industry Standard Architecture

Might be aimed for you - eventually.

[ErikTheRed]

I'll venture that this GPU/CPU integration is really aimed at the low end market to cheaply increase graphics performance for Vista. The integrated graphics chips that exist now are really just 2d chips, and have little or none of the acceleration that Vista wants for all its eye-candy.

There's also been some speculation that GPUs will go multi-core / multi-socket because their architectures are inherently more amenable to that and also because the current crop of ~700M transistor GPUs are friggin expensive to manufacture. So it's entirely possible that with a reasonably high-speed interconnect like HT3 available you'll have a multi-socket motherboard and for the second socket you can plug in a 4 general purpose cores, or maybe 2 general purpose cores plus 32 or 64 graphics cores.

Re:Might be aimed for you - eventually.

[SirTalon42]

GPUs have been multicore for years.

Will Intel play ball?

[Nimey]

Realistically Intel will have to implement these instructions on their processors for any programs to use them. Macs aside, who's going to write a program that only a tiny (but growing as old PCs are replaced) percentage of the market can run?

Re:Will Intel play ball?

[MadEE]

My guess is that a lot of this will be abstracted by the video driver so developers will not have to (largely) worry about it. Though it's not unheard of for developers to use special features of CPUs such as MMX, 3DNow! etc I don't see this as much different from the developer standpoint.

Re:Will Intel play ball?

[Lord Kano]

Let's call it driver magic. If your OS sees a video device and the underlying APIs can access that device, does it really matter where it's located?

LK

Re:ISA?

[whimmel]

ISA means Instruction Set Architecture or if you prefer the definition of the x86 'assembly' language.

People keep saying that. It is "Industry Standard Architecture"

Nuclear power

[erice]

I'm sorry. I can't help it.

Every time I see an article about AMD's "Fusion", I think
"Everyone knows that the power consumption of modern cpu's has gotten out of hand. Still, you gotta give AMD credit for having the guts to propose the obvious solution: An on chip fusion reactor"

Re:Nuclear power

[Anonymous Coward]

Awesome! With a nuclear wessel I can tell the power company to fuck off! Next thing you know my PC will be taking whales to the future!

Re:Nuclear power

[Yetihehe]

Intel will be the first to have fusion reactor on a chip. But AMD will do better, colder and they will probably name it ColdFusion.

Re:Nuclear power

[rising_hope]

And I keep thinking "who needs 5 blades anyway?"

*grin*

Some things never change

[psycho8me]

The cycle of moving from a central processor to specialized ones and back has been seen before. Could this be a sign of this type of change? From the jargon file: http://www.outpost9.com/reference/jargon/jargon_18 .html#TAG411

Re:ISA?

[MadEE]

People keep saying that. It is "Industry Standard Architecture"

That is the ISA bus. Not a CPU ISA that is being talked about here.

CTM? Eh?

[smcdow]

Close to reference to metal ( eg, metal& )?

What kind of initiative is that?

easy upgrade to path to seperate GPU and CPUs ...

[Anonymous Coward]

After they get all the mechanics figured out using a single die multi-core CPU/GPU chip they can quite easily split the CPU and GPU to seperate physical chips that are easy for us to upgrade at whatever pace.... Basically this could be there way of fleshing out Torrenza make profit sooner....

Re:ISA?

[drakaan]

It does mean that, if you're talking about the old ISA slots that we've just established that we're *not*, in fact, talking about.

If, however, you're taking about an extension allowing the issuance of commands to a processor (which we are), then the acronym stands for Instruction Set Architecture.

Crossfire?

[Debug0x2a]

Multiple GPUs can coprocess via Crossfire or SLI, so would the CPU be able to suppliment an additional video card instead of replace it possibly?

Doesn't this imply open source compatibility?

[macemoneta]

If the video uses a documented instruction set, doesn't this imply that AMD/ATI CPU/GPU chips will be open source compatible? Shouldn't that be all the information needed (from the GPU perspective) to create a 3D hardware accelerated driver?

Re:Doesn't this imply open source compatibility?

[Wesley Felter]

Even if the ISA is documented, writing an optimizing shader compiler is not easy. But my impression is that there are still some 3D-specific, fixed-function parts in the GPU that aren't triggered by instructions and thus aren't being documented, so the driver problem remains.

Bad Idea

[swalters1]

On CPU graphics is always a bad idea. Inevitably you end up with something less than capable, and the only real advantage is for severly low cost systems to cost $49 less. Not a smooth move AMD, especially when you look at Nvidia's CUDA system that are taking the complete opposite approach. GPU as a CPU instead of GPU on a CPU.

Re:Bad Idea

[Anonymous Coward]

It's failed twice before so it's impossible. God I love your logic.

it died back in the 90's

[RelliK]

I guess I should clarify. RISC "philosophy" lives on, but practicality has long been dead. Modern CPUs have RISC microcode with a x86 -> RISC translator in front. The translator adds a bit of overhead and uses up some silicone, but on the other hand CISC instructions are smaller, so you can fit more of them in a given amount of L1/L2 cache.

Re:it died back in the 90's

[serviscope_minor]

I guess I should clarify. RISC "philosophy" lives on, but practicality has long been dead. Modern CPUs have RISC microcode with a x86 -> RISC translator in front. The translator adds a bit of overhead and uses up some silicone, but on the other hand CISC instructions are smaller, so you can fit more of them in a given amount of L1/L2 cache.

You are just plain wrong on many counts.

RISC outsells CISC my a massive margin. Just look at the presence of PowerPC, MIPS and (the biggest of all) ARM in the embedded market. Comparing number of cores sold anually, RISC is the winner by a very wide margin.

Dead is a funny description for something which sells far more units.

Secondly, Intel keep the micro-ops in the L1 instruction cache (known as trace cache), so the CISC instruction set doesn't help there (it does for the L2 cache).

And thirdly, its silicon. Rubber processors don't exist yet.

Re:it died back in the 90's

[Anonymous Coward]

Also PIC and Atmel microcontrollers, there are HUGE numbers of those in the wild.

Re:it died back in the 90's

[Anonymous Coward]

ARM is the winner by far. Beyond that, Microchip's PIC is popular for simple tasks and hobbyists. Although I enjoy Atmel and their AVRs they are more expensive and that is a significant limitation as it tends to scale worse because of it.

At the university I attend the majority of microcontroller work is done with PICs as we get them for free as samples. Our alumnus tells us of the joys of PIC for microcontroller work and ARM for DSP tasks.

Re:it died back in the 90's

[Anonymous Coward]

Secondly, Intel keep the micro-ops in the L1 instruction cache (known as trace cache), so the CISC instruction set doesn't help there (it does for the L2 cache).

That's the case only with the P4. In Core 2 Duo, it's back to a normal L1 cache.

Re:it died back in the 90's

[insignificant1]

The translator has artificially large breasts, eh?

A super-FPU

[thue]

As described by Ars Technica, the new NVIDIA G80 generation of GPUs are actually collections of general stream processors, a type of FPU. The GPU functionality is then programmed in software. The article from Ars Technica points out that "These threads could do anything from graphics and physics calculations to medical imaging or data visualization.". I assume the ATI GPU is moving in the same direction.

So what AMD is adding to x86-64 is probably not just a GPU, but a new powerful general purpose massively parallel FPU.

Re:A super-FPU

[kramulous]

So, the medical imaging dataset I'm currently trying to visualise, 137GB - 4096^3 of shorts, will fit into texture memory of this thing? .... cool :)

Re:A super-FPU

[rbanffy]

AMD is making the x86 more Cell-like.

Re:A super-FPU

[Anonymous Coward]

Yes. I've pointed this out every time that Fusion has been mentioned here: a GPU is parallel vector processor. The resources available for rendering games can just as easily be used to accelerate scientific applications, and integrating it into one die will reduce the power and cost requirements. Since the GPUs are already becoming more general-purpose for more sophisticated shader programs, it makes a lot of sense to utilize those same resources for other applications without depending on incompatible shader architectures or PCI-Express add-on cards. It also gives AMD something to do with future die space besides creating 32-core processors that will be largely underutilized by software. People should think of this as AMD taking SSE out back and and just say, "to hell with the amateur hour, we're going to have some monster fp power." The end result is that you'll also be able to have superawesome graphics in games, as well as efficient scientific simulations.

Re:A super-FPU

[lagfest]

Then again, 137GB doesn't really fit anywhere else, so what's your point?

Re:A super-FPU

[kramulous]

There are numerous ways to handle a 2^37 byte dataset for real-time interaction. It was just an example of why the GPU is becoming obsolete (for video purposes). Low memory calculations, however, is another issue.

But the pretend to know all, once sentence statement/question/flamebait lacks a point. But what's the point in pointing it out?

SLI

[Anonymous Coward]

Will you be able to have a SLI like setup with a pci-e video card and the cpu-gpu?

The Cycle Continues

[RAMMS EIN]

This is the old integration-separation cycle continuing its course.

When I programmed PC graphics cards, they were basically dumb devices, just converting whatever was in memory to the signals needed to drive the monitor. The CPU did all the drawing work.

Then came the 2D accelerators, and later the 3D accelerators. The CPU sent them instructions about what to draw, and they did the drawing.

Now, we're seeing a move back to putting graphics inside the CPU.

The reason we're seeing these waves is, of course, that both approaches have advantages. Separating things leads to more modular and flexible solutions, and offers better performance, because special-purpose hardware can be faster than general-purpose hardware, and having two chips in parallel obviously allows more work to be done in the same amount of time. Integrating things, on the other hand, leads to lower power consumption and lower latencies, and avoids duplication of work (often, components that start out as special-purpose chips evolve into general purpose processors or something close).

Floating Point Ops

[Anonymous Coward]

Am I the only one who thinks that the fusion of CPU+GPU won't be used for the "integrated graphic stuff" but just for speeding-up f.p. operations?

The things that pass for news

[Anonymous Coward]

Adding extensions to the ISA seems to be the only real newsworthy or contested issue, physically migrating the GPU is something that was bound to happen sooner or later anyways (remember AGP?, or the days when adapters hung off of the south bridge?). PCIe is already moving more in this direction too, so taking it a step further and placing it behind the on-chip crossbar switch just makes sense. Incidentally, this sort of thing has been going on in the embedded space for years, including related ISA extensions (some of the XScales for instance, which place the CPU/GPU and a separately pipeliend MMX2 block behind the crossbar switch).

The only interesting thing from this article seems to be that they're using a separate crossbar switch which HyperTransport hangs off of. This makes little sense, as if the crossbar switch was HyperTransport to begin with, they'd already have this model. There are already other non-x86 implementations already using HyperTransport as an on-chip crossbar switch _today_. With the ability to break out HT lanes from the package it's also possible to do fun things like interfacing a bank of RAM in another room with a cable and treat it as on-chip RAM (something done when doing the Linux bring-up on POWER5, while the external memory controller hadn't been completed yet).

It's nice to see the x86 space moving towards something more sensible and scalable. Perhaps Intel will finally wake up and get a clue.

Why call it a GPU?

[pclminion]

The only point I can see in doing this is for more general use of the GPU than just rendering graphics. Graphics pipelines are pretty damn serial and the latency caused by putting the GPU on the north bridge is what, microseconds? Less? I don't see how this makes even the smallest difference in gameplay.

However, if you're using the GPU for something else like massive DFTs or physics simulations or any of the other cool stuff people are coming up with these days, where memory access patterns are more random, then I can see the benefit of reducing latency.

So why bother calling it a GPU? It's actually a fairly feature-rich complement to the traditional CPU, and putting it directly on the die seems only to reinforce that notion. Call it something else.

Re:Why call it a GPU?

[Anonymous Coward]

Hmm, you're right. It's becoming pretty generic in nature, so let's call it a Generic Processing Unit, or GPU for short.

Re:Why call it a GPU?

[pclminion]

That's silly. A GPU is less generic than the CPU itself. If anything should be called a "generic processor" it should be the CPU. How about we call it an SCU, or Super-Cool Unit?

AMD's Fusion

[smoker2]

Now with 5 cores ! (and a seperate core for those tricky areas)

Re:AMD's Fusion

[LordKronos]

Now with 5 cores ! (and a seperate core for those tricky areas)

Well, if my experience with the free Gillette Fusion I got in the mail has taught me anything, it's that the separate core will be a 386SX.

In short:

[msimm]

If you're a hard-core enthusiast, your upgrades may cost more.

stopacronymrecycling

[Anonymous Coward]

now THAT's a tag!

Fusion and CUDA lead the way.

[ravyne]

I've been following GPGPU stuff for awhile now, casually at first but much more closely now with the AMD/ATI merger and the release of nVidea's G80 architecture. Both of these represent the first big steps toward GPGPU technology (buzzword: stream computing) becoming reality.

The initial approach I suspect from the Fusion effort will basically be an R600-based, entry-level GPU tacked onto the CPU die. I'd imagine that this would have 4-8 quads (GPU 4-wide SIMD functional unit) as standard. This would mostly be targetted at the IGP market for laptops and small and/or cheap desktops. Its likely that CTM will enable this additional horsepower to be used for general clculations, but its primary purpose will be to replace other IGP solutions.

A little further out I see the new functional units being woven into the fabric of the CPU itself. This model likens closely to having many 128-bit-wide extended SSE units, likely to have automatic scheduling of SIMD tasks (eg - tell the CPU to multiply 2 large float arrays and the CPU balances the workload across the functional units automatically.) A software driver will be able to utilize these units as a GPU, but the focus is now much more on computation. It functions as a GPU for low-end users, and suppliments high-end users and gamers with discreete video cards by taking on additional calculations such as physics. Physics will benefit being on the system bus (even though PCIe x16 is relatively fast) because the latancy will be lower, and because the structures typically used to perform physics calculations reside in system memory.

Even further out I see computers very much becoming small grid computers unto themselves, though software will take a long time to catch up to what the hardware will be capable of. I see nVidea's CUDA initiative as the first step in this direction - Provide a "sea of processors" view to the machine and allow tight integration into standard code withought placing the burden of balancing the workload onto the programmer (which nVidea's CUDA C compiler attempts to do.) nVidea's G80 architecture goes one further by migrating away from the vector-based architecture in favor of a scalar one - rather than 32 4-wide vector ALUs, they provide 128 scalar ALUs. Threading takes care of translating those n-wide calls into n seperate scalar calls. Most scientific code does not lend itself well to the vector model, though over the years it has been shoe-horned into vector-centric algorithms because it was neccesary to get addequate performance. Even graphics shaders are becoming less and less vector-centric, as nVidea research shows, because many effects (or portions there-of) are better suited to scalar code.

Eventually, I think this model will grow such that the CPU will be replaced by, to coin a phrase, something called a CCU (Central Coordination Unit) who's only real responibility is to route instructions to the correct execution units. Execution units will vary by type and number from system to system depending on what chips/boards you've plugged into your CCU expansion bus. The CCU will accept both scalar and broad-stroke (vector) instructions such as "multiply the elements of this array by that array and store the results in this other array" which will be broken down into individual elements and assigned to available execution units.

All of this IMHO of course.

Re:Fusion and CUDA lead the way.

[Anonymous Coward]

You can buy an 8-core microcontroller for $13, so you're pretty much right. I have one and use one core as a "gpu" of sorts.

Re:Fusion and CUDA lead the way.

[ravyne]

Heh. I'm familiar with the Propeller Chip. I was one of the Demo Coders for the Propeller based Hydra game console by Andre LaMothe. Some of the other demo coder's wrote GPU routines that used 4-6 of the pchip's cogs in unison ;)

Re:Fusion and CUDA lead the way.

[Anonymous Coward]

Even further out I see computers very much becoming small grid computers unto themselves,

Yeah, these Fusion chips will make the second-generation OLPC kick serious ass !

I like your solution

[TubeSteak]

It neatly sidesteps the fact that high-end GPUs are massive compared to a CPU core

Intel Core Duo CPU
ATI X1800 GPU

BUT, you'd also have to squeeze all the other microchips that are on a high-end graphics card board... I don't know if you'd be able to squish all that into a CPU sized area. And if you can't, you're just changing the form factor & moving the graphics card onto a faster bus.

Anyone have a better idea how you can put quality graphics into a CPU?

Re:I like your solution

[Lorkki]

You must be very good - I certainly couldn't make a judgement on die size or complexity by looking at pictures of the outer packages.

Sarcasm aside, AMD seems to have a knack for integrating things, so I wouldn't put it past them to think of a way to strip some needless bus logic off of the GPU, marry it with a memory controller and put it inside one package ready to be stuck into a socket. The rest of the electronics pretty much involve getting data in and out of the frame buffer and to a format more suitable for display hardware, and that part could be integrated to the motherboard.

Sharing the memory bus with the CPU still doesn't sound like paving the way for extremely high-end graphics, though.

Re:I like your solution

[mabinogi]

Why don't you ask AMD, as they've apparently already considered it, or they wouldn't be talking about putting both the CPU and the GPU in the same package.

Without knowing anything about it, it would seem that if CPU+GPU in the same package is possible, then CPU + GPU in two separate CPU sized packages would be possible.

Re:I like your solution

[mrchaotica]

Thank you -- that's exactly what I would have said, if you hadn't done so first.

Re:I like your solution

[Anonymous Coward]

There is a better comparison of sizes here.

Re:I like your solution

[HonoredMule]

Having this "Pin compatible GPU" doesn't mean no video card. Rather, it means the video card just goes back to being merely a display output device. Obviously, graphics ram would disappear altogether (as it would replace main memory), and the other display chips, such as the RAMDACs, would remain on the video card. But that beast of a GPU would get a nice socket, HSF, and all the ram it wants, plus CPUs would get GPU-class ram accessing performance. To top it off, your next "video" upgrade would be the cost of the GPU...sans ram, output processing, PCB, and cooling solution costs. Also, your next ram upgrade would benefit your WHOLE system, be it size or performance, and such an interchangeable system could use its extra slots for graphics, physics, hardware acceleration of secure sockets, or whatever non-core function your machine most needs. The "CPU" slot becomes the next standard of extension, partially replacing PCIe, and "onboard video" becomes about as horrible a thing as onboard ethernet.

The Wheel of Reincarnation

[Anonymous Coward]

rolls round and round, round and round

http://www.catb.org/jargon/html/W/wheel-of-reincar nation.html

it's not just you

[j1m 5n0w]

I am more concerned with being able to select the best CPU/GPU combo, and not being stuck with a great CPU and lousy video card or vice versa. And by "lousy video card", I don't really mean poor performance so much as a lack of decent drivers. This is something that can change after buying the hardware; under Linux, the quality of the Nvidia drivers (which I currently hold my nose and use; last time I tried to use an ATI driver, they hadn't yet ported it to the version of X.org that I wanted to use) varies from release to release. It would be nice to at least have the option of replacing the video card rather than having to replace the whole system if driver quality takes a turn for the worst.

Re:it's not just you

[SirTalon42]

Since they're planning on adding to the X86 ISA, the CPU/GPU combo really seems like the best possible way for Linux to get good/great support for GPUs (ATI's drivers suck and are closed source, OSS drivers are ok, but performance sucks, nvidia's drivers are pretty stable, and perform very well, but are closed source). I could be wrong, but I think that this may be the best possible thing for Linux when it comes to graphics.

Forget OpenGL/Direct3D...

[Svartalf]

Most people associate these with their fixed functionality paths and the coding for the same.

That'd be right for the older games or the older hardware.

It'd not be right for the new hardware or the new games...

The new GPUs use programmable vertex and fragment shaders and the fixed functionality paths go
through an emulation of those paths in GLSL or HLSL. There's not much left that
isn't merely a simplified computer like a DSP is for signal processing- this is merely one that
is designed for graphics and similar operations instead.

The new games use their own shaders, etc. which is why GLSL is such a big deal and a tool to migrate
HLSL over is as much of one.

Who can say for certain that this doesn't make sense? I'm not going to venture a yes or no- because
I can see where they could pull it clean off and I can see some where it could let them fall flat on
their face.

Re:Forget OpenGL/Direct3D...

[gripen40k]

Hmm, well I admit that I cannot entirely grasp what you were saying there, but what I can say for sure is that there shouldn't be any emulation. In that respect I mean that if the emulation stage was removed, or replaced by dedicated hardware, then the standard game made today would run much much faster. So if AMD and ATI made this new fusion to be more of a dedicated instruction handler that off-loaded a wide range of operations from the CPU, as opposed to a dedicated on-die GPU, they might have a better chance of pushing into the hardcore gaming sector.

What I mean by dedicated instruction handler is this: The CPU receives a series of instructions (maybe a few hundred) to make a 3D model. Instead of making the CPU have to figure out what the heck to do with it all, it just re-routes it to the on-die 'GPU'. This in turn crunches some numbers that it was built to crunch, then throws it at the actual GPU, which would be situated off-die and through some kind of PCIe x16 bus. Of course the info that the on-die GPU sends the off-die GPU will be specially formatted to run really really fast on that particular card (ATI/nVidia/anything else).

I'm just guessing here, but I imagine that that would not only appeal to the hardcore nerds out there with too much money, but would also make games run much much faster. Am I on to something here or do I have it all wrong?

Re:Forget OpenGL/Direct3D...

[Anonymous Coward]

You have it wrong.

We talking about PC's here, not actual deticated tech. Because of this, automatically passing something to the video card (or whatever you want to call it) is a BIG nono. Let the software do it, that is what it is there for, and the hardware is not there to do things behind your back. Granted, in a better world, this would be a greate approuch, but this is x86 we are talking about, the type of hardware that just "eh" works, and as such, your approuch dosent fit into design, and how would the CPU know the formating for the video card? It cant, and NEVER the newer ones, which is why you should leave passing data around to software.

Now if you want to on-die-GPU to be able to talk to a framebuffer, thats ok, as it is purely in memory, and wouldent be incompatible with x86, mearly strange (which is greate, x86 is a big pile of hacks built off the crappiest hack of a CPU anyone could come up with). I really dont think you want this GPU to assist other GPU's, why bother? It would be better to have two CPU's, the on-die-GPU for low latency tasks (such as desktop UI drawing, image/video rendering, music visulisations, etc), and a video card GPU for all your gaming needs, and anything that sucks in the on-die-GPU. Seeing how most games are written in opengl anyways (i dont windows, so i dont know any games that use this strange "direct x" crap), and wont make use of this new CPU/GPU stuff anytime soon. Desktop and video/image related libraries/applications however are almost always looking to push out a few extra performence somethings, maybe even X could use this new GPU to assist in rendering?

Re:Forget OpenGL/Direct3D...

[somersault]

In a way, using any drivers is 'emulation' as you write your application/OS to interface on a general level with drivers, and then the drivers translate your function calls into a format that the hardware can understand (no, I've never written a driver, so I'm maybe not using great terminology, or have details wrong). This is really the best method anyway, as then you can change your hardware without changing your core software - you just need to update the drivers. I would have thought that it doesn't matter whether your graphics processing is being done on the GPU or integrated into the CPU, as long as the drivers are there. If you're going to integrate all the graphics processing into the CPU then it would of course be more efficient to not have to go through a driver, but then you start limiting your options. The speed benefit you get from running code directly on the processor rather than passing it off to a separate piece of hardware would soon be wiped out by the fact that an external GPU can use the latest, fastest memory, or can be run in SLI mode etc.

I still like the idea of integrating all parts of the PC onto one chip though, and if games start using pure CPU instructions rather than passing stuff off to a GPU, then there should be a speed boost. As others have pointed out though, x86 is a bit crap (well I always believed that to be the case in my Amiga and Mac days, though x86 has won out just as VHS did). Eventually we'll end up with a decent processor that just emulates the x86. Or maybe there will just be more and more extensions to it, as in AMD64, and now graphics extensions. All graphics processing is just computation anyway, and again as others pointed out, everything will be, and already is, going the way of the Cell processor (ie just have lots of cores and assign differnt tasks to each core, with one or more cores for phsyics, AI, graphics, and so on).

Re:Forget OpenGL/Direct3D...

[gripen40k]

"Eventually we'll end up with a decent processor that just emulates the x86."

Ahh, the days of the transmetta... Unfortunately that never really caught on :(

Re:Forget OpenGL/Direct3D...

[somersault]

It may not have, but it will have to happen eventually, s'quantum. Wossname.

Anyway I used to think that emulation was a really bad idea for that kind of thing, but it worked for Apple/Amiga with PowerPC, then for Apple moving to Intel processors.. as long as the new processor is sufficiently faster than the original design, then it can emulate at acceptable speeds, though there are likely to be a few incompatabilities.

Is the real goal just bundling?

[tap]

It used to be that CPUs didn't come with floating point units. You had to buy a 287 or 387 to go with your 286 or 386, and they weren't cheap either. I think we paid $400 for a 20 MHz 387 back in the early 90s. Around the end of the 386s' use in desktops, competitors to Intel (Weitek, Cyrix, some others I think) had produced 387 compatible chips that were faster and cheaper than Intel's. For the 486, Intel decided to integrate the floating point unit, which made it pretty much impossible to buy someone else's chip. Sure there were technical merits to that, but I'm sure that fact that it killed any possible competition in the FPU market wasn't lost on Intel's execs.

Trying to bundle products is nothing new. A company that makes a whole package doesn't like it when parts of the package can be bought from other companies. Instead of just competing for the whole package (and the few companies who can provide that), they need to compete for each individual part, and every company that can make any one of those parts. If AMD puts the GPU in the CPU, then it's pretty hard for nvidia get OEM's to include their GPU. Nvidia will have to build a CPU that's as good as AMD's, and that's not going to happen any time soon.

Re:Is the real goal just bundling?

[Prof.Phreak]

I can imagine the real goal being DRM. With everything on the CPU, making some instructions priviledged, they can force any program that wants to manipulate (decode?) video at a ``fast'' rate to call the OS to perform the decoding---allowing the OS to ensure the video has valid signatures before it proceeds.

Sure folks would still be able to use libraries that run on the CPU, but if those are inefficient/slow compared to the specialized instructions... then who knows.

Just being paranoid...

nforce? anyone?

[sam0vi]

To me this sounds like some nforce-Mega-integration thing. I dont like it because when i think about it i think about: Drivers?! Maybe for the PS$ or something

Aren't we forgetting something?

[Driving Vertigo]

I think people are forgetting the idea that the integrated "GPU" is going to be more like a programmable DSP than an actual graphics accelerator. We know the CPU is wonderful at computing integer math, but it's always been lacking in floating point. The modern GPU has become a floating point monster. The integration of this new element, including instructions to utilize it in native "stream processing" will likely cause a small leap forward in computing power. So, please, don't think of this as a replacement to your video card, put more like an evolution in the area of math-coprocessors.

Re:Aren't we forgetting something?

[jmv]

We know the CPU is wonderful at computing integer math, but it's always been lacking in floating point.

Actually, recent CPUs are already at least as fast doing floating point than integer. The difference is that the GPU has massively parallel capabilities.

Re:Aren't we forgetting something?

[Driving Vertigo]

Only when pointing to instructions like SSE, which in itself is specialized and specifically targetted at the desktop sector. I am simply pointing out that as far as general processing goes, integer is the bread and butter of the CPU. Go take a look at the Intel ARM processor and then come back to me.

Re:Aren't we forgetting something?

[jmv]

Have a look at the (pre-SSE) x87. The performance is still about the same as for integers (two pipelines, about one result per cycle each). Comparing to ARM isn't fair because it doesn't have an FPU. For newer CPUs that have an FPU, integer performance is about the same as float. I'm not sure if GPUs can do integer, but if they do, their integer implementation is probably also much faster than the CPU's integer arithmetic. So it's not about int vs float, it's all about general-purpose vs. specially optimized for (vector) number crunching.

Not necessarily

[r_jensen11]

Microsoft has a boatload of documentation, but that alone doesn't seem to make the EU happy enough

Why two versions?

[msobkow]

It would be the responsibility of the OpenGL rendering implementation to take advantage of a fused CPU/GPU core, much as the drivers took advantage of SSE implementations. The application should see no difference under OpenGL or DirectX unless it's using vendor-specific extensions.

What I find more interesting is the possibility of applying "graphics" operations to general data types, not just the presumed integral axis references. Imagine feeding currency into a bump-map transform to texture a global sales map, or temperature to texture a thermal relief map. Additional axis could be conveyed as color, and key references like "average" or "mean" could display as "sea level".

Date projections act the same as Z-buffer clipping/projections, except for the data type. Hardware accelerated time-based history data could define a "current" plane feeding the above visual displays, allowing you to literally scroll through years of data.

Algorithms used for motion detection in video processing, temporal cleaners, static removers, etc. could be very interesting for detecting and correcting data anomolies.

Data is data, regardless of whether the end result is pixels.

nah

[Anubis350]

Naw, not showing your age, just exposure. I'm 20 and it was my first thought too. Admittedly though, my Dad's been in the industry for over 30yrs. We have *tons* of old machines with ISA and tons of ISA cards around

I for one...

[kenglish]

welcome our CPU/GPU fusing overlords.

Vector Engines

[Kadin2048]

I don't think so. The application can be linked against a single graphics library. The GL just swaps some function pointers when special hardware is available.

It seems like this problem must have been one that's been solved before. How do people write code for processors that have vector engines, like IBM's AltiVec or its Intel equivalent?

Re:Vector Engines

[quanticle]

I don't know about IBM, but if you're compiling for x86, you have to specifically state what extensions (SSE, SIMD, etc.) you want to use. This is why most software is compiled for i586. That instruction set is guaranteed to be present in most every desktop CPU, so your application is guaranteed to run.

On the other end, you can use an optimizing compiler and specify which processor you need to use. A lot of games do this, and specify that you need a certain CPU (or better) to run.

The real question is...

[Schraegstrichpunkt]

... do they think we're going to run binary blobs on our computers in order to use the CPU?

Re:The real question is...

[maxwell demon]

... do they think we're going to run binary blobs on our computers in order to use the CPU?

You already do. Except that the binary blob (aka microcode) is embedded into the CPU. Which is somewhat inavoidable because without a working CPU there would be no way to read that binary blob into the CPU (although I could imagine to have a CPU which initially cannot do anything but read some binary blob from a specified memory address residing in the system bios flash, which then contains the actual implementation of the CPU instructions).

Hardware raytracing

[Anonymous Coward]

I would find it cool if AMD/ATI decided to go for the hardware raytracing route.
Some impressive results (http://graphics.cs.uni-sb.de/~woop/rpu/rpu.html) have been achieved with a simple FPGA at 66 MHz. Imagine a dedicated (or several) ASIC(s) at much greater speed.
Sure, you wouldn't expect as high level graphics as todays top-end cards immediately, but I think it is the way to go forward.

Wait for it, wait for it...

[athlon02]

What about this? ... one day as non-volatile memory based on carbon nanotubes becomes more prevalent and cheaper they'll slap that on the "CPU" die as well... Remember this company? Nantero. What you'll have in the mobile market is a CPU core, a GPU core, and probably a good 512MB non-volatile memory all in one die. Also, with stuff like the PPC chip PA Semi is working on, see here, we'll probably see pcie, sata, gigabit ethernet controllers built-in as well. So what do you have? ... quite the small system with low latency and power consumption. And for a bonus, you'll probably eventually have carbon-nanotube memory off-die as well in place of a hard drive. So those "yeah, that's fake" computers we see in movies that look like a medical tricorder with gigs of storage, will actually be real. Now, of course, this is years off, but I like the direction. I especially like the idea of 10's to 100's of gigs in a single carbon-nanotube based DIMM (or whatever is out then) or flash-drive replacement. Then we can finally shed hard drives, optical drives, and the like for purely non-mechanical devices (w/ the exception of course that tape-backups and such will likely still be present, but less so for the consumer market).

OK, time to stop dreaming and get back to work.

Instruction Set Architecture

[dunc78]

I believe it stands for Instruction Set Architecture, with x86 being an example of an ISA, not the old bus of which you are thinking.

No way!

[Ayanami Rei]

The point is not to add a "video card" to the CPU. This Fusion would in no way prevent you from needing a graphics card. What it would do is add a general purpose DSP-style unit similar to a set of processing nodes in the new NVidia and ATI GPU implementations, capable of doing many graphics functions (vertex, pixel ops, etc.). You would still need to mate it to an output card which would have some of the less embarrasingly parallel operations of 3d graphics, framebuffers, 2D scaling acceleration, and output drivers.

In this fashion, using multi-socket boards you could scale your effective 3D processing power with additional or more dense chips that can be balanced between 3D, physics simulation, and other workloads.

This would be changing the way we do graphics and game design for enthusiasts, professionals, and home users alike.

RISC sucks for performance.

[Ayanami Rei]

It may help you make cheap chips, but for performance, it sucks.
You can't feed an instruction cache fast enough to justify RISC in high performance computing. The richer the instruction set, the more branch prediction, the more out of order and speculative your execution, the better.

You totally don't get it.

[Ayanami Rei]

They're putting CUDA _in_ the CPU. It's a CPU, plus massively parallel stream processing. No PCI-e bus to traverse. Shared L3 cache. Scalable with however many cores that platform supports...

Math CoProcessor

[KingNaught]

Big deal so the graphics card is basicly becoming the Math Coprocessor of the new millenium.

Re:Am I the only one? Yes, you are.

[mollog]

It might not be optimal for gamers, but then again, it might actually be optimal for gamers. Imagine changing your CPU/GPU but keeping your original motherboard and getting a big jump in performance.

Putting your petty concerns aside, I think that this CPU/GPU combo will revolutionize certain computing tasks. Currently grid computing is used to try to complete tasks that are computationally intensive. But the general purpose CPU is not as effective at much of this computation as is the GPU. GPUs are more than an order of magnitude faster than CPUs at certain iterative calculations. If the grid were populated with CPU/GPU multi-core systems, we'd be slamming research so fast it would produce an explosion of medical discoveries and development. With Alzheimers, cancer, and other age-related diseases looming, this sort of research will have a huge impact on both our physical as well as economic well being.

I, for one, welcome our new AMD CPU/GPU overlords. Let's hope ASUS is on the job and has a mobo ready when the time comes.

x86 CELL any one ?

[DrYak]

For me, the whole thing looks like AMD trying to develop a x86 answer to the CELL chip.

I think Fusion isn't intended to replace the graphics cards.
I think Fusion is specially intended to add more GPU-grade vector processing capabilities to the CPU. Just like the CELL, more power to calculate more complex things (physics, geometry, etc...) but, in the end, the texture shading is still done on a separate card. This looks specially obvious as the Fusion doesn't communicate with it's GPU part is if it were a separate GFX card. Instead the vector functions are just accessed with a additionnal intruction set, just like other coprocessing units on the die (SSE, MMX, 3DNow, FPU)

In fact this could be interesting to cut down bottlenecks.
- Such GPU-vectore-engine is near to ressources that it needs : Physics and later goemetry depends on data available to the main CPU. With a fast inerconnection between GPU and CPU on the same die, those things will run faster.
- After geometry processing, the scene data in the Fusion chip will look exactly like the scene will be once rendered by the GPU. Hidden surface removal and culling can be implemented at this step and thus less geometry data has to travel between the Fusion and the GPU on the GFX-board.

RAM in video cards needs to stay.

[sowth]

I would like to point out eliminating dedicated video ram is a bad idea. I remember my atari 130XE would be somewhere around 30% faster when you disabled video because the cpu and video shared the same ram. A chip driving anything like a CRT will require constant access to memory. If you share ram with the cpu, then you will slow your computer down, especially if you like high resolutions and refresh rates. According to my calculations, 1600x1200 @ 100Hz with 24 bpp needs 576 MB/sec. Up the res to 3200x2400 and you need 2.3 GB/sec. I don't use resolutions this high, but from what I've seen, plenty of people do. Even with today's faster ram, it would be a siginificant drain.

Re:RAM in video cards needs to stay.

[HonoredMule]

That's a rather short-sighted point. Everything, CPU especially, benefits from faster memory access just as much as does the CPU, and applying the current technology that is invested in dedicated GPU memory solutions across the whole system has a performance impact upper bound of O(c). I seriously doubt that GPU+Physics hungry processing will ever account for less than two thirds of overall memory I/O (barring large file transfers, network services, and other things that do not go alongside rendering/physics hungry operations).

It's a one time performance hit with an everlasting performance payoff. If you have a hard time seeing the potential payoff, consider the new nVidia processors, and the benefit they would receive from higher bandwidth inter-processor communication for that shiny new API that offloads more vector hungry work for non-rendering purposes. A GPU as an actual standalone specialized processor has radically higher potential than a system that was inherently designed to have data dumped on it and never seen again.

To gain further perspective, recall that we're cutting GPU performance by considerably less than one half. Even when all physical limitations have been reached, bandwidth can be doubled by doubling data bus width. Bandwidth is bound by a constant too, and it's a bigger one than the inverse of the one-time performance hit. Latency is not even an issue, as it would not even be affected, so long as each processor socket gets its own channel of data access, capable of operating simultaneously on mutually exclusive memory spaces (worst case:latency would then be doubled and bandwidth halved on threadlocked data thrashing between processors, which is still a far cry better than communicating across PCIe). Now as adding channels is effectively doubling data bus width, it would be more reasonable to expect one socket to have a dedicated memory channel, and another shared between remaining sockets...with the intention that the most important function (GPU for games or other rendering tasks) would get the dedicated socket. AMD's onboard northbridge could seem like a problem, but in fact, it's actually a minor detail of implementation. The "master processor" could easily be responsible for forwarding additional channels of memory I/O to other devices and enforcing data consistency...that's not exactly a new job for it. In fact, the system I'm describing is radically simpler and more direct than the one currently in use, provided multiprocessor capability is already present (which it is). This may seem more complicated, but only because I got somewhat wrapped up in expressing some of the inane details explicitly.

Bottom line: if we're willing to double the current GPU memory bus width, split into 2 channels (one dedicated to the GPPU) and use it (and GDDR-whatever-the-latest-version-is RAM) as the model for our main memory, the very first system of this design will already boast a performance increase and a powerful potential as platform for much more improvement yet. With current graphics cards, a lot of proprietary technology is bundled in one gelatinous package, with no interchangeable parts. It's time for that technology to branch into segments where individual components can shine on their own merit, and more companies can become a part of the graphics+specialized processing technological landscape. It's almost like an open source argument for hardware, that the open system is inherently capable of maximizing the use of human capacity for ingenuity.

In my opinion, it's a time for specialized (read, not general purpose OR necessarily x86 ISA based) secondary processors in general to shine, and reach a more commonplace status.

Re:RAM in video cards needs to stay.

[sowth]

What are you talking about? I was saying the amount of continuous memory access required for supplying the CRT with data it needs are huge. It has nothing to do with a GPU, but the chip reading from the framebuffer and creating the analog signals going out to the VGA connector. I suppose you could engineer the RAM to have dual output ports (the one for CPU/GPU and another for the framebuffer), but this would seem like a specialized task and make general purpose ram more expensive (and incompatible with systems which didn't use this scheme). Why not make this specialized change to a small portion of ram? Even 4096x3072 at 24bpp is 36MB of ram. Small compared to the 1GB or more which is standard on computers today.

Maybe I wasn't clear. I just think dedicated framebuffer RAM should not be eliminated. As for the GPU having freely available programming specs and/or being a separate entity to the graphics card, I'm all for it.

Re:RAM in video cards needs to stay.

[HonoredMule]

In that case, I apologize for the misunderstanding. Of course, the video output would still need a buffer just like every other I/O device. And that is, of course, minuscule compared to the amount of ram that goes on a video card when it is, as currently, so much more than an I/O device. Before GPUs went onboard, video cards were carrying about 8-16MB framebuffers for uncompressed data.

It is worth noting, however, that the concept you coined as "dual outputs" is essentially how most system-on-a-chip devices work. I/O devices are simply mapped to a specific address in main memory, and that's their buffer. I cannot vouch for their use in more performant devices, but I'm sure it's not unheard of. In this case, its quite undesirable. Video output should be just a normal I/O operation...just with higher bandwidth requirements than ethernet or firewire. In keeping the normal I/O device philosophy, the CPU can do software rendering to the video card's framebuffer without a dedicated GPU, or delegate that task to one. Thankfully, the cache size needed for the framebuffer is considerably less than 36MB even for triple buffering, because it uses good compression algorithms to store the frames. To reduce bandwidth requirements, it would be a good idea to standardize the compression (and keep it softcoded for upgradeability through firmware), allowing lower bandwidth requirement between the processor cluster and the video output. Uncompressed data throughput for 60fps at your mentioned resolution would be ~2GBps...more than even the best GPUs today can handle on top of the other memory accesses they require.

I'm also thinking about dual GPUs working in tandem for the same output, so a dedicated I/O buffer still makes the most sense there too (rendering alternate frames would be the most easily implemented method currently in use)...and I'm rambling again, for which I apologize. I know, I know...I should start a weblog so that everyone who cares can gather in one place and criticize me all at once. :P I have to stop now.