Slashdot Mirror
Nvidia Working on a CPU+GPU Combo
Max Romantschuk writes "Nvidia is apparently working on an x86 CPU with integrated graphics. The target market seems to be OEMs, but what other prospects could a solution like this have? Given recent development with projects like Folding@Home's GPU client you can't help but wonder about the possibilities of a CPU with an integrated GPU. Things like video encoding and decoding, audio processing and other applications could benefit a lot from a low latency CPU+GPU combo. What if you could put multiple chips like these in one machine? With AMD+ATI and Intel's own integrated graphics, will basic GPU functionality be integrated in all CPU's eventually? Will dedicated graphics cards become a niche product for enthusiasts and pros, like audio cards already largely have?" The article is from the Inquirer, so a dash of salt might make this more palatable.
What I don't understand is that I thought GPUs were made to offload a lot of graphics computations from the CPU. So why are we merging them again? Isn't a GPU supposed to be an auxillary CPU only for graphics? I'm so confused.
What I'm not confused about is the sentence from the above article: Oh, I've worked with my fair share of DAAMIT engineers. They're the ones that go, "Yeah, it's pretty good but
My work here is dung.
Because they're never wrong and never sensationalize a story for a few clicks.
Perhaps something like this could be used in a general-purpose computer. Many technical hurdles will have to be overcome, but it may be possible after several decades of research.
AMD and Intel have their own fabs that are at the leading edge of semiconductor technology. I highly doubt that nVidia will open up a fab for their chips. But who knows, IBM may produce their chips for them.
I think the better option would be to have a graphics chip fit into a Socket 939 on a dual socket motherboard, with an AMD chip. It will have a high-speed link through hyper-transport, and would act just like a co-processor. I'm no chip designer, so I have no idea what the pros/cons of this are, or if it's even possible.
How much heat do integrated graphics solutions put out?
I can't imagine it is that much
(since they mostly suck)
[Fuck Beta]
o0t!
With this integration, does that mean a standard for 3-d? No more Nvidia/ATI drivers. The OSDEV guys would love this if it came to that. But how would this integration work? A co-processor space like MIPS? If so, does that mean that graphics calculations have somewhat been moved back to the CPU? And what about the actual workings itself, I'm guessing the actual registers would still be memory mapped in someway (or I/O ports for x86, whatever).
I'm thinking way too much. It did alleviate boredom for about a minute though...
Unless Nvidia is partnering with Intel to release this chip, I think they're getting too far out of their confort zone to be successful. Sure, a dual or even quad core chip with half of the cores handling graphics would be great, but can Nvidia deliver? I doubt it ... look how many companies have gone down the tubes after spending millions/billions after trying to make an x86 compatible chip, let alone trying to integrate top end graphics as well.
Nvidia is a fantastic graphics card company - they should continue to innovate focus on what they're good at rather than try to play follow the leader in an arena they know nothing about.
Crack - Free with every butt and set of boobs
There seems to be a cycle of integrating and decoupling things.
We had separated math co-processors, that later were integrated in the CPU.
Then the separated GPU, which will soon be integrated back too.
factor 966971: 966971
GPUs are going the way of the math co-processor. I think it's inevitable.
"False hope is why we'll never run out of natural resources!" - Lewis Black
It's quite clear that the AMD-ATI merger was to aquire the IP and expertise necessary to integrate gpu core(s) on the same die as cpu core(s). Nvidia does not have to actually market a design, but rather patent some key concepts, and this could provide revenue or protection.
I would very much doubt that they could compete with AMD and Intel who have already patented many x86 cpu concepts.
It's a shame that Intel has decided not to buy nvidia, and go it alone with it's own design staff.
Why doesn't google buy Nvidia?
Okay, I admit, I haven't RTFA yet, but if GPUs do get folded back into CPUs, I think we need to thank MS.
... Seriously. Think for a minute.
:)
No.
The major driving force right now in GPU development and purchase are games.
The major factor that they have to contend with is DirectX.
As of DirectXv10. A card either IS, or IS NOT compliant. None of this "We are 67.3% compliant".
This provides a known target that can be reached. I wouldn't be surprised if the DirectX10 (video) featureset becomes synonymous with 'VGA Graphics' given enough time.
Yeah, sure, MS will come out with DX11, and those CPUs won't be compatible, but so what?, If you upgrade your CPU and GPU regularly anyway to maintain the 'killer rig', why not just upgrade them together?
This space for rent. All reasonable inquiries will be entertained at proprietors discretion.
The article is from the Inquirer, so a dash of salt might make this more palatable.
A dash?..hell! better use the whole damn shaker
A while ago -- and maybe it was in the Slashdot discussion about ATI, I'm not sure -- somebody described a cycle in computer design, where various components are built-in monolithically, then broken out as separate components, and then swallowed back up into monolithic designs again.
Graphics chips seem to have done this cycle at least once; perhaps now we're just looking at the next stage in the cycle? We've had graphics as a separate component from the processor for a while, perhaps the next stage in the cycle is for them to combine together into a G/CPU, to take advantage of the design gains in general-purpose processors.
Then at some point down the road, the GPU (or more likely, various GPU-like functional units) might get separated back out onto their own silicon, as more application-specific processors become advantageous once again.
"Ladies and gentlemen, my killbot features Lotus Notes and a machine gun. It is the finest available."
With nVidia making CPUs, and of course Intel and AMD/ATI (DAAMIT) making CPUs, how could nVidia expect to grab any market share? No offence to the nVidian engineers, but their product would have to be miles above the Intel/DAAMIT offerings in order to make most people even consider a system with an nVidia CPU. I think they would be better off if they attempted to enter into a contract with Intel for their CPU/GPU combo ideas, maybe Intel could get a few nice server chipsets out of the deal? :)
Why is everyone getting excited about this? Now we're going to have a CPU that's only partially documented, and we lose even moreso to closed-source blobs.
This isn't a good thing unless they also release documentation for it!
I prefer my articles with a dash of accuracy.
Haven't they already???
The vast majority of machines (at least, from my experience, which could not be broad enough) seem to be shipping with integrated graphics on the motherboard. Certainly, my last 3 computers have had this.
Granted, I buy on the trailing edge since I don't need gamer performance, but I kind of thought most consumer machines were no longer using a separate graphics card.
Anyone have any meaningful statistics as opposed to my purely anecdotal observations?
Lost at C:>. Found at C.
I've been expecting this for a while, ever since the transistor count of the GPU passed that of the CPU. Actually, I thought it would happen sooner. It's certainly time. Putting more transistors into a single CPU doesn't help any more, which is why we now have "multicore" machines. So it's time to put more of the computer into a single part.
NVidia already makes the nForce line, the "everything but the CPU" part, with graphics, Ethernet, disk interface, etc. If they stick a CPU in there, they have a whole computer.
Chip designers can license x86 implementations; they don't have to be redesigned from scratch. This isn't going to be a tough job for NVidia.
What we're headed for is the one-chip "value PC", the one that sits on every corporate desk. That's where the best price/performance is.
The Inquirer is more of a rumor site than a news site. They have a pretty good track record for their rumors, but they don't have people on record backing this one up.
What NVidia eventually does may not bear much resemblance to the story.
http://www.antsbitemylegs.com/mountain.jpg
From someone who has mainly been involved in computers as a semi-competent user:
Why are the multiprocessor units suddenly so popular, relative to why e.g. the Voodoo graphics cards failed? I remember them being ridiculed and ending up in the performance backwaters through their 2-4-8(-16) multiprocessor cards, but it seems that there are engineering reasons why multiple processors are now suddenly coming into favour, or?
This is happening because all the big vendors know that once consumers really start using stuff with the cell processor they won't need all the other stuff for most purposes. Not all, but or most normal consumer desktops. It's sneaking in everywhere because it was a good idea and it works. You have multiple cores that go do different stuff and talk to each other really really fast, makes building a computer a lot simpler and in theory can be a lot smaller. I know which one I would pick say at a low end, a 300 buck computer with a normal even duocore processor and a cheap amount of ram and cheap vid and sound, or something for the same price with the cell processor where different cores did didfferent stuff and 4 times the ram. No contest. There will always be a market for higher end and specialised cards and chips, but once you can whip out one thing that fits in one slot and takes the place of 3/4ths of the current mobo you will be making some cash on that badboy. Pretty soon you'll be able to just plop your cellphone (which will act as your main unit) down on your desk in front of a wireless enabled screen, keyboard and mouse and do basically most stuff you do today with a normal big machine, and it is all because of the multiple core chips, they will make it possible and cheap.
When I saw this headline I immediately thought of this article, an interview with Jen-Hsun Huang (CEO: nVidia) by Wired dated July '02. In it, the intention of overthrowing Intel is made quite clear, and ironically enough they even mention the speculation from a time when it was rumored that nVidia and AMD would merge.
It's actually a very good article for those interested in nVidia's history and Huang's mentality. Paul Otellini ought to be afraid. Very afraid.
It already is.
:wq
"what other prospects could a solution like this have?"
Duh. Gaming consoles. Add memory, a sound controller, and some sort of storage, and you're in business.
Oh, you're not stuck, you're just unable to let go of the onion rings.
"back in the day" when my 80387 (7?..coprocessor) was sitting to the side with it's own instructions to complete, the commands never had to traverse up and down a much slower ISA bus. why not apply the same idea to upgrading your CPU and GPU separately through a slot/socket design? they can still share memory (not necessarily cache--sorry), but the agp/pcie bus is removed. communication between a single die g/cpu will still have some sort of bus, albeit a very tiny bus with tiny pathways; why not just make the pathways a bit bigger and build them into the mobo?
There's certain advantages to having (A) the GPU functionality integrated in the CPU (cost, certain performance aspects, others) and some to having it (B) in a separate GPU (more easily upgraded, more real estate, less heat problems)...
...so finally the innovations for design (A) are reaching a critical mass... all those innovations that recently couldn't be pounced upon because of the separation of GPU/CPU are now making (A) looking pretty damn nice again.
Every once in a while an unrelated tech innovation comes around that benefits (A) or (B) in some indirect fashion. It could be faster bus speeds, more sophisticated GPU instruction sets, etc etc etc. Doesn't really matter what they are, but they happen all the time and each may benefit (A) or (B) more than the other.
So currently we are at a time when (B) is the standard design- GPU and CPU remain separate. Whenever an innovation has come along that benefits (B) more it has been integrated into the latest NVIDIA or ATI design. Whenever something that benefits (A) has come along it has been ignored in the last few years... this probably includes things such as the balooning costs of the GPUs, difficulty in getting GPUs into now popular laptop form factors, texture latency, etc. etc.
This CPU/GPU cycle happens every few years... We've been in an (A) cycle many times before- Remember what MMX was originally for? It was because we didn't have GPUs and couldn't do efficient block operations for video at the time... Remember the IBM PCjr? That was arguably another (A) cycle because IBM wanted to save money on circuitry/memory for the video subsystem, which is arguably just a primitive form of a GPU.
This GPU->CPU oscillation has been, and will continue to be, going on forever.
Well, the thing about a high-end CPU is that it's something like 80% custom logic, where a GPU is much more "standard cell" design. So the fact that NVIDIA is good at GPUs with lots of transistors doesn't mean that it will be easy for them to build a CPU. It will be very difficult to build something competitive with Intel and AMD. But if anyone out there right now has a shot at it, it's NVIDIA. Licensing of the x86 architecture is going to be a sticky issue.
Something that's interesting about this, if true, is that Intel might be the one playing catch-up. AMD will have ATI graphics, NVIDIA will have NVIDIA graphics, and Intel will have Intel graphics, which have always been pretty horrible.
Integration is the key to cost reduction, performance improvement and power efficiency.
e t/MPC8641DDLCRFS.pdf
L2 cache used to be external. Then they integrated it when technology and performance allowed. L3 cache then became external while L2 was integrated; now you can buy processors with all this inside. Put the memory controller inside the CPU and you no longer need to spread out high (er than CPU core voltage) IO lines with nasty length requirements between Northbridge and CPU, and can clock the bus faster. Put the ethernet and so on inside the Northbridge and you no longer need discrete chips and buses for them, and can run them faster with tighter integration to a DMA controller and embedded RAM.
Integrate the graphics hardware into the CPU and you can have most of the high-bandwidth devices on the fastest possible bus.
Take Freescale's nearly-done 8641D Power Architecture processor. It's 2 G4s, 4 gigabit ethernet, USB2, PCI Express and RapidIO, DMA, interrupt and memory controller, I2C, serial. This chip is priced LOWER than equivalently specced Core Duo 2 combinations (CPU, i975MCH/ICH combination), and the performance.. is about the same. However board implementation will be much easier, and lower power. All you need for a system is to add your peripherals; a SATA chip, perhaps. I can't think of anything else that is missing besides graphics.
http://www.freescale.com/files/32bit/doc/fact_she
Eventually SATA will go in there, you can bet on that. Then graphics. Then one chip per board is a possibility. You thought NanoITX was small..?
yes, indeeed, but is a sad state of the world. I hope OpenGL return again and force the Neverwinter Nights to develop Nwn 3 with a OpenGL path again :I
-Woof woof woof!
If you could preload half the code you run on the CPU over to a CPU/GPU chip, and cut down buss use by >50% by utilizing a 'smart' GPU chip, this should enhance overall system performance by tons in graphic intensive applications. Not to mention that it simplifies simpler system needs (say embedded or wireless) for smaller systems that require less space but provide required functionality with high graphics performance. Just my thought
Support NYCountryLawyer RIAA vs People
A proprietary CPU that you can't use with your free operating system.
Considering the recent security problems with their proprietary piece of shit driver I'd rather stay away from anything manufactured by NVIDIA.
What if you could put multiple chips like these in one machine?
;)
They'd probably be obsolete in three months, as opposed to one month
When the posters fear their moderators, there is tyranny; when the moderators fears the posters, there is liberty.
Thinner thin clients perhaps?
Maybe it is also a move to compete with the Cell Broadband Engine, where the SPU's can act similar to a GPU.
Wow, a systerm on a chip, never heard of those before...
Nvidia doesn't really have a choice in this matter. They need to at least explore the option of creating a CPU or be relegated to niche status.
It's pretty clear that both Intel and AMD are intent on swallowing up the lower 3/4 (hand-waving guess) of the GPU market over the next few years. And I believe that ATI will still be fighting it out at the top end over that remaining 25%.
That would leave Nvidea as a niche player in the uber high end, making GPUs strictly for graphics professionals and gamers with too much money. You can survive as a niche player in technology... but it's a tenuous life. Just ask Cray, SGI, Transmeta, Sun, Cisco, and hundreds of others*. Only the latter two of resemble themselves in their primes -- but even then, they've lost influence.
* Whatever happened to Gravis and the Gravis Ultrasound?
The problem with gluing the GPU and CPU together is that it'll be a humongous chip, with low yields and therefore very expensive -- more expensive than a comparable CPU and GPU separately.
This integration is nothing new. "Cyrix did it!"
First, processors now have integrated memory controllers. Now this.
Seems like Cyrix was way ahead of its time.
and do you really want to take away system ram for video ram?
with vista high ram use 128+ just for video whould make it even worse.
What folding@home has shown us is that specialized hardware can be put to other novel uses, in some cases yielding huge performance gains. In this case, using graphics circuitry for the protein folding computations. All Nvidia proposes is to reduce latency by moving the GPU on-chip (and perhaps have the added benefit of locking you into a GPU of Nvidia's manufacture...).
However, there are many algorithms that are not optimally solved by throwing EITHER a CPU or a GPU at them. A more interesting idea would be in the form of on-chip programmable hardware a la FPGAs, in conjunction with a nice low-latency on-ship setup. Then, developers could push more interesting and advanced circuits onto CPUs, such as crypto engines or crackers. This would allow for readily incorporating new high performance codecs for crypto, graphics, audio, protein-folding, etc.
A guy from Intel recently presented at a seminar at my university. He is working with a group that is pushing for a CPU architecture that looks kind of like a GPU, when you look at it at a very high level (and perhaps your eyes squinted just a bit).
The unofficial title of his talk was 'the war going on inside your PC'. He argued that the design of future CPUs and GPUs will eventually converge, with future architectures being comprised of a sea of small and efficient but tightly interconnected processors (no superscalar), and that it is basically a race to see who will get there first - the CPU manufacturers or the GPU manufacturers.
One of his main points was that with increased compiler effort, potentially many computational workloads can be made to run on the tiled architecture of simple processors, much in the way that the process of graphic rendering has been able to be shifted into the type of workload that can leverage the 'tiles of simple processors' found in a graphics card today, even though the nature of graphic rendering was originally better suited for execution in a typical CPU, where control dependent loads run efficiently. When the workload cannot be mapped to the 'tiles of simple processors' architecture, just slap a superscalar processor in the corner of your die (like nvidia seems to be doing) to take care of those small corner cases.
So, we will likely be seeing a lot more of this in the future. Especially now that AMD and ATI are together.
(More details on the abstract of the presentation I mentioned can be found here)
Presumably you'd still need separate graphics memory. We all know how well integrated memory works now.
do you really want to take away system ram for video ram?
If using larger chips means I can get 2GB combined RAM for the price of 1GB system RAM and 256MB video RAM? Absolutely.
Why are the multiprocessor units suddenly so popular, relative to why e.g. the Voodoo graphics cards failed? I remember them being ridiculed and ending up in the performance backwaters through their 2-4-8(-16) multiprocessor cards, but it seems that there are engineering reasons why multiple processors are now suddenly coming into favour, or?
multiple processors (CPU, GPU or otherwise) are a way to add more 'cycles' based on current technology. This has the advantage of being able to get more out of your current designs and manufacturing technology, but comes at the cost of increased complexity in both the supporting hardware, and in software.
Getting a single core implementation faster is always the more efficient way to add processing capacity, but it is very impractical beyond a certain point due to power and heat considerations (where that point is exactly depends on the state of technology at any given moment but in the end is limited by the physical size of molecules, at least for as far as current technology goes)
So, multiple processors is not directly better from an engineering point of view, rather, it is a solution to overcome the speed limits of current technology, provided you can deal with the extra complexity (moving much of the hardware complexity into the chip itself like AMD and Intel are doing now removes the burden from systemboard designers, but the complexity itself is still there, esp. on the software side of the picture).
With regards to 3dfx, it seems to me that:
1. They failed to manage the additional complexity
2. As their competition showed, limits of technology at that time were much higher then what 3dfx managed, which indicates there were problems with either their design or manufacturing technology, or more likely, with both.
it seems that there are engineering reasons why multiple processors are now suddenly coming into favour
Traditionaly, when you move to a more compact production process, your parts are closer together, so it takes less time for the electric signals to move through them (propogate), so you can get faster clock speeds without really changing the design much. When Intel reached the 90nm process (or maybe the one before -- 130nm?), they were startled to discover that that effect just didn't work anymore. The chips were smaller, and thus cheaper to make, but they didn't work any faster than the old ones. AMD apparently ran into the same thing when they reached that size.
With their easy gains in clock speeds at a dead end, they went looking for something else to improve, and more CPUs was it.
Mostly because it's a different point in the life cycle. Many of the easy performance gains in CPU's have now been taken - it wasn't so long ago that a 100MHz clock increase was sensational, but now it's expected. It's getting tough to eke out big performance wins in each generation now, so it's easier to move sideways to multiple cores. Parallelism is the future (witness the supercomputers).
I don't know the 3dfx history all that well, but I'd *guess* that their cards were getting hard to dig more performance from, and that they went two ways - one to work on a new core, one for multiprocessor. They could at least keep some market share if they got something out quickly. Unfortunately, the new core probably took too long, and the MP wasn't enough ; there were still new things to try for single GPU's, and they ended up with an expensive, slow card.
Well, yeah, but you've never heard of one that can run Windows Vista before, I'll bet. :)
System RAM is SLOW compared to GPU RAM. PCIe actually allows very high speed access to system RAM, but the RAM itself is too slow for GPUs. That's one of the reasons their RAM amounts are so small, they use higher speed and thus more expensive RAM. Also because of the speed you end up dealing with cooling and signal issues which makes it impractical (or perhaps impossible) to simply stick it in addon slots to allow for upgrades.
Even fast as it is, it's still slower than the GPU would really like.
What you've suggested is already done by low end accelerators like the Intel GMA 950. Works ok, but as I said, slow.
Unless you are willing to start dropping serious amounts of cash on system RAM, we'll be needing to stick with dedicated video RAM here for some time.
1) Processors are wicked fast at floating point these days. Have a look at the benchmarks a modern chip using SSE2 can do some time. Integer doesn't inherently mean faster, and chips these days have badass FPUs.
2) For many things, it DOES make a difference. You might ask why do we need more than 24-bit (or 32-bit if you consider the alpha channel) integer colour? After all, it's enough to look highly realistic. Yes well that's fine for a final image, but you don't want to do the computation like that. Why? Rounding errors. You find that with iterative things like shaders doing them integer adds up to nasty errors which equals nasty colours and jaggies. There's a reason why pro software does it as 128-bit FP (32-bits per colour channel) and why cards are now going that way as well.
3) In modern games, everything is handled in the GPU anyhow. The CPU sends over the the data and the GPU does all the transform, lighting, texturing and rasterizing. The CPU really is responsible for very little. With vertex shaders the GPU even handles a good deal of the animation these days. The reason is that not only is it more functional but it's waaaaay faster. You can spend all the time you like trying to make a nice optimised integer T&L path in the CPU, the GPU will blow it away. You actually find that some older games run slower than new ones because they rely on the CPU to do the initial rendering phases like T&L before handing it off, whereas newer games let the GPU handle it and thus run faster even though having higher detail.
Actually only the size of the frame buffer is multiplied, i.e. a scene at 1024x768x32 bit would take 3 MB without anti-aliasing, and at 1600x1200x32 bit 7.3 MB. With 4xFSAA those scenes would take 12 MB and 29.2 MB respectively. The rest of the memory is used for texture & vertex data, and other miscellaneous stuff like shaders. Those don't grow in size with the display resolution.
CPUs are very important for good games, and only getting more so. You think MS paid for a 3 core PPC CPU in the 360 for fun? If a low power cheap part would do the trick, they'd have jumped on it.
Even after you've offloaded all the graphics, there's still a ton to be done. AI, physics, and sound to name just three. Fire up a copy of Oblivion some time and you'll be amazed how hard it hits the CPU. It even makes use of dual cores, if they are available. Yes, a prime factor in game speed is having a GPU that can keep up with the shiny graphics, but equally important is a CPU that can keep up with the game data. If you stuck a 7900GTX on a P4 1.6GHz the game would look beautiful, but run slow (if it ran at all) because the GPU would be sitting waiting for the CPU to get it stuff to render.
For cool, latest-gen games, you need both a good GPU and CPU.
Intel has tried to integrate specialized I/O on their chips more than once, a nd so have other companies. Then they end up with something neither best in either and become commercial failures.
Yes the SYSTEMS Tom used to test have normal speed ram for systems. Duh. The graphics cards, however, have much faster RAM. For example my system at home has DDR2-667 RAM. That's spec'd to run at 333MHz which is 667MHz is DDR RAM speak. My graphics card, a 7800GT, on the other hand has RAM clocked at 600MHz, or 1200MHz in RAM speak.
Not a small difference, really. My system RAM is rated to somewhere around 10GB/second max bandwidth (it gets like 6 in actuality). The graphics card? 54GB/sec.
Video cards have fast RAM subsystems. They use fast, expensive chips and they have controllers designed for blazing fast (and exclusive) access. You can't just throw normal, slow, system RAM at it and expect it to perform the same.
They don't even need to license it. Today, GPUs are more complex than standard x86 CPUs. So if nVidia can handle GPUs, they can handle CPUs. They have the tools, the know-how, and the engineers. Developing their own x86 CPU shouldn't be too hard for them.
Don't need good graphics. An Intel GMA950 does just fine for desktop use. Hell, it actually can play older games, and is enough to handle AeroGlass in Vista.
The problem is the GP seems to think that nVidia would have a viable market combining their excellent graphics chips with shitty processors. Nope, not so much. If you want a good graphics card, you pretty much by definition also want a good processor. There are actually plenty of people who can use a good processor and have no need for a good video card, but the reverse is very rarely true. Games are the big use, but any other app I can think of, like HFSS or Solidworks, where good 3D card is a benefit, you also need a good processor.
So ya, nVidia could try the low end market, but there's some real stiff competition there in the form of Intel and their GMA series. That's what the cheap computers now feature, a cheap Intel cheap and their cheap accelerator. However those that think they could combine a good GPU with a cheap CPU as a gaming system are off base.
The one-chip PC is a falicy... if the integration of all blocks produced higher performance, then you can charge more for that integration. The slower blocks (ATA, USB, PCI) don't perform significantly better when integrated, thus consumers don't value integration (board manufactures will, but only by a fixed amount).
If you integrated the gpu/ethernet/sdram/cpu, then all the high-speed blocks can talk with each other at die level speeds. It's clear as day that performance will significantly increase.
The southbridge IO (ATA USB PCI) requires many pins, and moving them to a single chip doesn't save enough money at board level to justify the added expense at die level. The power cost of running a bus to the southbridge is tiny compared to the total system budget.
But the combined RAM is slower then video ram.
GPUs are larger, but not more complex. It's a fairly homogenous array of execution units - massive SIMD, in other words.
It doesn't take anywhere *near* as much knowhow to produce a GPU as it takes to produce a CPU - and it's an entirely different sort of knowhow, as well.
nVidia might have a chance at producing a (special-purpose, GPU-like) supercomputer chip; they'd have serious trouble producing an x86/x86_64 CPU competitive with what Intel and AMD make. Of course, they know that themselves, so they're not doing that.
Are we just waiting for patents to expire to see more FPGAs in general purpose systems?
Imagine a GPU that plugs into the existing Xeon (Core 2 versions) socket. Now the four-core CPU in the other socket can talk to it over the memory bus. Put a separate bus on the "far side" for the GPU's local memory and/or framebuffer.
Historically, yes, but RAM's been getting faster lately, and I expect that will continue for the next couple of years before these chips are expected to come out.
Once upon a time, there were mainframes, and computing was good. Only it was only good for large universities, governments, and a handful of corporations.
Then, there came minicomputers. Fewer components meant smaller machine with smaller prices. Banks, small colleges, and even some high schools got into the game. My high school had an AS/400.
Then came the micros. These are what we call PCs now. They started as hobbyist toys. Then, the chips got more powerful and more memory was put in them, and they started to be more powerful than the minis of a few years before.
There were math coprocessors, but those got absorbed into the CPU. Then came accelerated 2d video, and later accelerated 3d video. Sound processing appeared on expansion cards, and then on the system board.
Then came bigger mainframes, faster minis, and high-end workstations. Micro users wanted more power. Micros were made with more instructions, some that did things with more bits at a time. Then the internal scheduling was made so that multiple instructions could be in different stages of being processed simultaneously. It was called pipelining, and the users saw that it was good.
Mainframes grew again, these things called servers that had a brain like a micro but a body like a pizza box but with lots of IO strength and more mouths and ears sprung up in the Data Center, which had been just the closet next to the coffe pot. People saw that servers were good, because the processing went on in the Data Center instead of on their Micros. And the servers sometimes joined together into great hordes called Clusters that could challenge any mainframe.
Then there were vector instructions added to the PUs. And some PUs could do a vector instruction and regular instructions both at once. And the vector unit stole some of the glory of the graphics accelerator.
People put two CPUs (or PUs, since they're not really central at that point) into one system, then four or eight. Then someone put two logical PUs on one chip.
Someone figured out that lots of the processing for graphics was still done on the CPU -- and more so now with vector instructions. The graphics accelerators grew their own vector units, and become very wise in the ways of the vector and in pipelining. More graphics work was placed with the accelerators than ever before.
A Major Designer of PUs determined that what was good and Intel-ligent for math processors in the beginnings of the micro was still good for other types of processing on the micro. The path had widened again, and two of these wider PUs per chip was becoming the norm. The die had shrunk yet again, meaning that now a stronger PU was now smaller than ever. Even more vector units could be fit on a chip, despite the fact that the chip held two whole PUs. In Embedded Land, in the valley of Console, a strange company called So Knee had put one PU on a chip with a massive vector unit, and called it Sell. A Major Designer determined that this was good, and noticed that it looked something like a PU and something like a graphics accelerator.
Now, graphics accelerators had become known as GPUs, because they were doing their own fair share of Processing. They had started to team up as well, sometimes with more than one GPU inside one Micro. Not since the Friendly days of The Great Chicken Head had so many chips gone into one machine. Most micros had at least two PUs, even if they were on one chip. They had one or more GPUs, even though the "G" was questionable since they could do other kinds of Processing, too. They had sound, often from a DSP. They had Networking, which is those extra mouths and ears of which the Servers have in such abundance.
And so, the die was shrinking again, and A Major Designer was swallowing a maker of GPUs. The socket for a different type of processor, which had been so Intel-ligent so long before, started to look like a good place for a GPU. And with the die shrinking, making more room on a chip, the thought of a CPU and a GPU together started look
Many years ago, when the 3d video card revolution was taking place under the helm of companies like 3dfx, Intel announced that eventually video cards would become obsolete. I forget the exact time frame but I believe they said by 2006 to 2008 cpus would be handling the majority of video processing.
Now, this statement was made before the cpu manufacturers started hitting the brick wall, and 3d video technology was still in relative infancy. So, to be fair, we should probably tack on another 5 years onto that estimate. Once we do so, the market seems to be heading in the very direction Intel predicted between Cell and GPU/CPU integration, we aren't too far off from this becoming common place at all.
You are who you are, let no one tell you different. But, never close your mind to a new point of view.
Why doesn't one of the game console makers take their favorite processor, and add 8 GPU's to it. Maybe join forces with a large cpu maker, like IBM. They could give it one of those silly names, like "cell processor" and make a game console based on it. If it could run Linux, and folding@home, that would really be a bonus.
Everyone on slashdot would surely want one of these, even if it were a little overpriced.
Who would win this election: Andrew Weiner vs Andrew Weiner's weiner.
News is just rumours that have come true.
I was just having some crazy thoughts, how about this; AMD liceses HT, cHT and various x86 technologies to nVidia to help them get competetive in the x86 market. AMD/ATi and nVidia manufacture all of their discrete graphics chips on the cHT platform effectively locking intel out of the CPU + discrete graphics market segment. Intel is confined to their own integrated platform for low in Desktop and and Laptop configurations. AMD/ATi and nVidia reap the spoils of an enlarged market now that intel is locked out. AMD becomes the new intel, nVidia becomes the new AMD (in terms of the current x86 landscape)
If they were to do it I'd suspect that this also has something to do with that talk of using GPUs for AI processing. The main problem they are having at the moment is that graphics cards are designed to send data back to the CPU which would be needing for any AI processing which wasn't simply graphical in nature. If the CPU and GPU were on the same piece of silicon (or more likely completely integrated together into a single or dual core microprocessor), information could easily be passed back and forth to allow for not only very fast graphics, but also AI. The question is though whether nVidia would be able to make a cpu which could compete with offereings from Intel and AMD.