AMD Reveals Plans to Move Beyond the Core Race

J. Dzhugashvili writes "The Tech Report has caught wind of AMD's plans for processors over the coming years. Intel may be counting on cramming 'tens to hundreds' of cores in future CPUs, but AMD thinks the core race is just a repeat of the megahertz race that took place a few years ago. Instead, AMD is counting on Accelerated Processing Units, chips that mix and match general-purpose CPU cores with dedicated application processors for graphics and other tasks. In the meantime, AMD is cooking up some new desktop and mobile processors that it hopes will give Intel a run for its money."

Comments

Same old.

[sam991]

Intel pushes the 'more power! faster!' philosophy while AMD just redesigns the architecture and it takes Intel a few years to catch up. Not much has changed since 2000.

Re:Same old.

[TrisexualPuppy]

Intel pushes the 'more power! faster!' philosophy while AMD just redesigns the architecture and it takes Intel a few years to catch up. Not much has changed since 2000.

Correct. Intel still has the lion's share of the market, and they want to keep it that way. It's interesting how they "cheat" and lock two dies together and call it dual-core or quad-core just to come out with the technology "first" to keep the investors happy.

AMD is smaller obviously, so it has fewer resources...but with those Alpha scientists, they're going to keep going strongly. It's just a matter of time with business directives like this before AMD takes over. They've been having some really cool ideas...and a few more over a few years, the innovators may win. And no, I'm not an AMD fanboi, but I have talked to some architects from IBM and Intel, and they do concur.

Re:Same old.

[Martin Blank]

To be fair, the Merom architecture is a true dual-core architecture. The "quad-core" chip that Intel announced recently is simply glued together, and AMD's recent split design isn't terribly much better (though the two cores are linked by a dedicated datalink). AMD has a true quad-core design being prepped for next year, and Intel may have to follow suit, especially if AMD is able to show a decisive performance edge.

Re:Same old.

[SuluSulu]

AMD is smaller obviously, so it has fewer resources...but with those Alpha scientists, they're going to keep going strongly. It's just a matter of time with business directives like this before AMD takes over. They've been having some really cool ideas...and a few more over a few years, the innovators may win. And no, I'm not an AMD fanboi, but I have talked to some architects from IBM and Intel, and they do concur.

Doesn't that assume that Intel doesn't change their strategy? It seems to me that Intel has adopted a strategy that is designed more for marketing than for innovation. This seems to have worked VERY well for them, but it doesn't mean that they couldn't change their strategy in the future. It's interesting to note that whenever AMD pulls ahead in terms of processor performance Intel catches up and vise versa AMD catches up to Intel.

As for whether or not AMD will beat Intel, I favor Intel's marketing strategy not because it's what I would buy but because it's what will be easier to sell to the masses and that's where the lion's share of the money is.

Re:Same old.

[teg]

Correct. Intel still has the lion's share of the market, and they want to keep it that way. It's interesting how they "cheat" and lock two dies together and call it dual-core or quad-core just to come out with the technology "first" to keep the investors happy.

Cheat? The result is 4 cores in one socket. Things like "they cheated!", how many nm the process is etc is really irrelevant. What matters is the end result, like performance, power usage, memory bandwidth. That AMD can't do it yet and had gotten slow and docile by a couple of years success and being the top performer will hopefully be fixed soon, but as of now, they're badly lagging.

Re:Same old.

[Sloppy]

Heh. Yeah, "cheat" or not, they delivered a product that works. It's inelegant when you look at it, but winning with inelegance is practically Intel's historical identity. Look at 8086 vs 68000. ;-)

And copying that, is how AMD managed to threaten Intel in the first place. The very idea of "extending" x86 to x86-64 is disgusting, but there were plenty of 64-bit architectures out there, and guess which one finally managed to unseat the 386.

Integrated graphics..

[mr_stinky_britches]

That sounds great and all, and the AI that the article mentions really does sound interesting...but I am not clear on how a processing unit for extremely specialized tasks is going to translate into significant performance gains? Is the current generation of CPU not optimized for mathematic operations? This seems the most direct way to get the best all around performance, to me. Also, isn't it kind of sucky to make the processor only good at a few good things rather than fast in a diverse range of applications?

If anyone can give me any insight here...please speak up.

Thanks

- I post interesting things or short articles I write here

Re:Integrated graphics..

[SQL Error]

Is the current generation of CPU not optimized for mathematic operations?

What do want to run on a computer that isn't "mathematic operations"?

More specifically:

Are current CPUs optimised for physics simulations? No.
For image processing? No.
For data compression? No.
For encryption? No.

These are all areas where custom cores can provide enormous performance benefits (both in absolute terms, and in terms of performance per watt) over current CPUs, which are general purpose.

Re:Integrated graphics..

[CAIMLAS]

Imagine a processor with special circuitry routines which will speed up the operation of the following by a significant percent:
- database servers
- web servers
- CAD and 3d programs (rendering)

Basically, it's not much different than MMX or any other extension to a processor. The programmers can still code for the x86 (or whatever) architecture and the same operating system, but then shortcut those instructions when the additional instructions are found to be available. Or maybe they can work it transparently so programmers don't have to do anything additional - it'll optimize on the fly (provided they can figure out how to do that). Overall, I think the software headache will be worth it to companies, as they will be able to have substantial gains in performance in the hardware department, cutting cost while gaining performance. What datacenter wouldn't love to use half as many machines to provide access to the same amount of information; what animator wouldn't love to have their workstation be able to render things at twice the speed?

Re:Integrated graphics..

[mo]

The thing is, there's a huge number of applications that do the same basic computations over and over again.
Just as the floating point coprocessor became the FPU section of the processor, it makes sense to give future processors the ability to do the common operations that are now done by graphics cards.
Things like matrix multiplications (which is actually will be a single processor operation in SSE3) are used all over the place in graphics, sound, and well, virtually anything that eats up CPU power these days. Doing this stuff serially in a traditional general-purpose CPU takes forever, but it's blazing fast if you do it in parallel specially designed hardware.

You might think that having hardware that just does matrix-multiplications limits your processor to only certain domains, but it makes sense to have a dektop process that's fast at desktop tasks: (play games, rip/encode video/audio, run skype, raytrace and use photoshop)? There's still going to be the server-class processors that are good at general-purpose non-mathy things like serving databases, but it just doesn't make sense to use a Xeon in a desktop when an AMD/ATI integrated chip will do all the stuff you want faster and cheaper.

Re:Integrated graphics..

[megaditto]

What I'd like to see is a couple of those field-programmable thingie cores that can reconfigure their circuits to a specialized calculation a program is doing... Wishful thinking but still...

Re:Integrated graphics..

[Deflatamouse!]

Remember, a large percentage of the time the entire system - including CPUs, chipsets, memory, disks, etc., are just pushing data around without performing any calculations. We could all gain from better performance of these operations as well.

Re:Integrated graphics..

[jcasper]

XtremeData (http://www.xtremedatainc.com/) has a board with a FPGA that plugs into an Opteron 940 socket. Not exactly what you asked for, but a step in that direction.

Re:Integrated graphics..

[GeffDE]

Physics simulations and image processing can be (and are) done on GPUs. Same for any hardcore math stuff, like Folding Proteins. The problem with the AMD approach is that there are only so many (and I don't think it is many, but I really don't know, so if you do, please let me know) different kinds of operations. Like I said, the physics simulations and image processing are the same type of problem and also conveniently tackled very proficiently by graphics hardware.

Re:Integrated graphics..

[stigmato]

Like I said, the physics simulations and image processing are the same type of problem and also conveniently tackled very proficiently by graphics hardware. Perhaps thats why AMD merged with ATI? Just a thought.

Re:Integrated graphics..

[blahplusplus]

"What I'd like to see is a couple of those field-programmable thingie cores that can reconfigure their circuits to a specialized calculation a program is doing... Wishful thinking but still..."

This is what human minds do but CPU's are far from this goal, not to mention the nightmare of managing it as complexity increases.

Re:Integrated graphics..

[kestasjk]

As a database guy I really don't think processors would make a bit of difference to database speed in the vast majority of cases. Database design is usually what's at fault when you're shown a slow database, followed closely by query design, followed by memory, followed by hard disks, followed by processors. The same sort of thing applies to web servers; the bottleneck is never the processor.

As for CAD, well I think that would be quite a waste. Remember that processor designers only have so many transistors to use, and they have to make the most of them. It would be a waste of die real estate, chip designer time, CAD software writer time, etc, just to get a slight performance boost. I sure wouldn't want to pay for CAD specific instructions when I don't use any CAD tools.

I think general purpose processors should leave as much as possible up to software; optimize the general purpose stuff as much as possible so that everything runs faster, and if a user needs some extra fast processing capability for a specific task they can get an extra processor for that purpose.
You can buy external graphics, crypto, and physics processors; if there was enough demand there would also be external database, web server, and CAD processors.

np-hard optimization board

[GrEp]

Nvidia, please make a board for solving small instances NP-complete problems. Mainly max-clique and graph coloring :)

Re:np-hard optimization board

[hritcu]

Mainly max-clique and graph coloring

Solving SAT in hardware would be enough, since you can reduce most NP-complete relatively easily to it, the SAT-solving algorithms are already highly optimized, and there were even previous attempts to build special purpose hardware for solving SAT.

Re:Free Enterprise

[Moridineas]

Like if there were hypothetical competitive operating systems like Mac OSX, Linux (and the competition therein--Ubuntu, Fedora, etc), FreeBSD, OpenBSD, Solaris, etc?

Re:CPUs and GPUs

[Bright Apollo]

This is expected news, if you step back objectively.

AMD loses and will continue to lose the manufacturing race with Intel. Intel will likely continue to develop smaller and smaller dies, and AMD could never hope to leapfrog them for lack of cash to do so. Of course, give Intel their due: they employ some pretty smart people as well.

Ultimately, making your CPU do more specialized tasking, or capable of programmatic specialized tasking (think FPGA) is the right kind of innovation for them. I would also look to see more RISC-based operations, and wouldn't be at all surprised if they went off in that direction in some way. If they do, IBM has something to worry about... ... which brings me to the POWER CPUs. Where I work, I can architect a solution in a variety of ways, and currently I choose to build p550s with POWER5s (later POWER6s) with all the nice dynamic partitioning and micro-partitioning that you cannot get (at that level) from anyone else. I wonder how comfortable IBM would be feeling if they saw AMD start to offer the same kinds of partitioning elements in their CPUs and architectures?

This is all good news for me.

-BA

hyper transport

[Joe The Dragon]

Intel would be better off if they where to start useing hyper transport Even having two cores on same die with linked by hyper transport to each other with one link to the chip set is better then 2 cores shearing the FSB.

What is the point of having 32 cores with only one link to the chip

Even with the new Xeon's there still only one link per cpu and the cpus need to use it to get to ram

Amd chips right now have up to 3 newer ones will have up to 5 links

Re:hyper transport

[aquaepulse]

This is supposed to be the plan for Intel. Why would they use an existing standard?

Re:hyper transport

[ZachPruckowski]

FSB 1333 (333 QDR) seems to be holding up well for 4 cores at the moment. It only really seems to be much of an issue in the 4+ socket world (which is admittedly lucrative). You are correct in the sense that Intel stands to gain from a move to a hypertransport-like system, but that only raises the counter-issue: If Intel has the performance lead now, even with the FSB issues, then AMD's 07-08 products have to hit it out of the park to beat Intel's chips without that handicap.

The real issue is feature size. AMD is hurt badly by being consistently behind on that. Intel's been at 65 nm for a while now, and AMD is only now releasing 65 nm parts. Intel will be at 45 nm in some lines by this time next year, while AMD is a year behind them. Feature size brings with it higher yields (more chips per wafer) once you work the kinks out, lower heat, and more transistors per chip. That's the game winner right there, unless one of them shoots themselves in the foot again, like Netburst.

Re:hyper transport

[Fred_A]

Intel will run CSI ? Cool Does that mean I'll finally get to know who rooted my friend's Windows boxes ?

Re:hyper transport

[mrand]

In general, lower voltage (presumably due to smaller geometry gates) does result in lower dynamic power in CMOS technologies:

Pd = f*C*(V)^2 [Power= Frequency x Capacitance x Voltage squared]

The problem with smaller geometry is that, starting in the 65nm to 90nm area, static power dissipation (heat independent of operation frequency, such as leakage current across a transistor) can start to rise noticeably. The question is if that leakage starts becoming a significant fraction of the overall power dissipation. In processors running at GHz, I'm guessing not. In FPGA's, it can be.

      Marc

Amiga?

[vjl]

That sounds like the Amiga's way of doing things...over 20 years ago! I'm glad it's catching on, and I'm glad AMD is doing it; AMD usually gets things right, and makes their products a lot more affordable than Intel...

/vjl/

Dedicated processors for "other" tasks

[Gothmolly]

Sounds like OS-dependency and driver hell to me. Imagine if you had an MP3 decoding co-processor, an MPEG-2 encoding co-processor, an Excel co-processor, a GCC co-processor... getting it to all work seamlessly would make today's 200MB video card drivers pale in comparison. So you install WMP version 42 and you have to check "use dedicated MP3 coprocessor" in Tools->Options? The whole point of CPUs is that they are general purpose.

Re:Dedicated processors for "other" tasks

[Anonymous Coward]

Don't forget crypto, the hardware AES on a 1GHz VIA C3 runs circles around a A64 X2 4800+ doing the same in software, at something like 10 vs. 80W power consumption.

Free Reliant

[Anonymous Coward]

This is the type of innovation that usually comes when there is true competition in the market. Imagine how much better the OS market would be with similar competition.

This is the type of innovation that happens when you abandon a crew on Ceti Alpha V. Imagine how much better the OS market would be with similar KHHHAAAAAAANNNNNN!!!

Naturally... not all processes are

[mseidl]

infinitely parallel. Gaming on one hand, can be very much parallelized. With physics and an ever increasing amount of vertices to transform and AI to calculate, and in general crap to render.

A lot of other software is not. Such as: Office productivity, operating systems...(these can benefit, but ultimately they'll reach a limit).

The other question is, when you put hundreds of cores on a chip, how do you handle logistics of accessing cache? Or cache coherency?(not required) They it'll go up to 16 or so cores before they might run into some cache latency issues.

I think the other question is... how long till software catches up? We're at a point where hardware has been carrying software. Software is coded for the most part, pretty crapily(thanks to out of order cores). When are the software designers going to get with the program and leverage hardware more? I know hardware is very dynamic. But, now we're seeing hardware reach it's limit, and that multiple cores don't do anything unless some key multi-threaded apps are running.

PPC called and wants its AltiVec back?

[AHuxley]

Where will all the optimised code come from?
What will the cost be in making it all work 'just' for AMD?
How locked in would any code be?
Over the life of a project, will it be worth 'porting' code to AMD?

Re:PPC called and wants its AltiVec back?

[elhedran]

Where will all the optimised code come from?

Believe it or not, 100cores requires optimized code as well. programs don't magically become multi-threaded, a developer has to work out how to split the work up into 2/4/100 threads and not lose performance due to locking/thread communication.

What will the cost be in making it all work 'just' for AMD?

Probably about the same as making it work for a new graphics card

How locked in would any code be?

It sounds to me they are talking optimization. hence it would run on an intel, just slower.

Over the life of a project, will it be worth 'porting' code to AMD?

Ah, I'm not going to try and answer this one. But it is an excellent question and should be asked (and answered) by AMD.

Hybrid Graphics & the Cell roadmap.

[lightversusdark]

The article is a bit light on detail, there's a webcast of the presentation on AMD's Investor Relations site (needs a login (BugMeNot doesn't work) and it's WMP or Real only. And it's apparently four hours long.
The most interesting thing for me was the mention of "Hybrid Graphics":

According to AMD, notebooks with hybrid graphics will include both discrete and integrated graphics processors. When such notebooks are unplugged, their integrated graphics will kick in and disable the discrete GPU. As soon as the notebook is plugged back into a power source, the discrete GPU will be switched on again, apparently without the need to reboot. AMD says this technology will enable notebooks to provide the "best of both worlds" in terms of performance and battery life.

It also looks like they're also extending the Fusion concept along Cell-like lines, with additional cores for non CPU or GPU purposes.
Their road map through 2008 only talks about up to quad core, although I assume this means CPU cores (I'm not sure that I would accept a CPU+GPU on a single die branded as a 'dual-core' chip). I think the Cell has eight cores, but due to yield issues not all are enabled in a PS3, and they are not all functionally equivalent. I don't know if this is the case for the Cell-based IBM blades, though.
The roadmap basically looks like periodic refreshing of the product line reducing power consumption with each iteration, which is where I think Intel have got a head-start on AMD. However, if AMD can sort out the yield issues, and compilers and developers begin to take advantage of these "associate" cores in Cell and future AMD architectures, then maybe Intel will have turned out to have missed a trick, as they did with x86-64.

Re:Hybrid Graphics & the Cell roadmap.

[Watson Ladd]

They wouldn't need to work on compilers, and developers wouldn't need to rewrite code if they encouraged people to use BLAS and then optimize BLAS. I think that a lot of this multi-core stuff will end up being matrix and vector math units with some kind of MIMD based on GPU style masking branches. If they wrap it in a special-purpose API, they only end up hurting their benchmark scores.

Re:Hybrid Graphics & the Cell roadmap.

[acidrain]

The cell has 8 SPUs which are stripped down vector processors and one PPU with is an only mildly stripped down PPC core. On the PS3 one of the SPUs are disabled to increase yield.

The problem with the cell is the the SPUs are hell to program if you have a problem that doesn't fit nicely in the 256k ram that an SPU has. And most programming tasks these days don't. If the programming is essentially DSP work then you are good to go, but hopefully AMD learns from Sony's mistake, and still allow all cores random access to memory.

However L2 cache eats up a *lot* of die space, so who knows what nightmares we have coming.

Cue :

[CODiNE]

20 people asking "Why would anyone need this?"
50 people replying "I encode video"
45 people replying "Games"
10 replying "Babes of course"
1 karma whore incapable of making a decent top 10 list. :)

Re:Free Enterprise

[badboy_tw2002]

Of course, for the competition of the type that exists between Intel and AMD or AMD/Nvidia you need a common standard to compete with. If all apps ran on the same OS/GUI API then you'd have a true choice in operating systems (this one is more secure, this one faster, this one runs Word twice as fast and handles more DB load, etc). CPUs have x86, GPUs have DirectX/OpenGL, OSs need a standard application interface commmonly accepted by software developers. Otherwise you're comparing not just the OS but all the stuff that goes with it (skins, music players, etc etc etc)

Transputer

[temojen]

Compaq used to sell those; they're called transputers and came as a PCI card with 4 FPGAs, some RAM, and a PowerPC CPU.

Re:Free Enterprise

[Magic5Ball]

That would make NetBSD the most competetive operating system ever!

most of them idle most of the time?

[bcrowell]

The two big obstacles to getting better performance from parallelization are that (1) some problems aren't parallelizable, and (2) programmers, languages, and development tools are still stuck in the world of non-parallel programming. So from that point of view, this might make more sense than simply making a computer with a gazillion identical, general-purpose CPUs.

On the other hand, I'd imagine that most of these processors would sit idle most of the time. For instance, right now I'm typing this slashdot post. If I had a video card with a fancy GPU (which I don't), it would still be drawing current, but sitting idle 99.99% of the time, since displaying characters on the screen as the user types is something that could be done back in the days of 1 MHz CPUs. Suppose I have a special-purpose physics processor. It's also drawing current right now, but not doing anything useful. Ditto for the speech-recognition processor, the artificial intelligence processor, the crypto processor, ...

There are also a lot of applications that don't lend themselves to either multiple general-purpose processors or multiple special-purpose CPUs. One example that comes to mind is compiling.

On a server, you're probably either I/O bound, or you're running a bunch of CGI scripts simultaneously, in which case multiple general-purpose processors are what you need.

For almost all desktop applications except gaming, performance is a software issue, not a hardware issue. I was word-processing in 1982 with a TRS-80, and it wasn't any less responsive than Abiword on my current computer. Since I'm not into gaming, my priorities would be (1) to have a CPU that draws a low amount of power, and (2) to have Linux do a better job of cooperating with my hardware on power management. I would also like to have parallized versions of certain software, but that's going to take a lot of work. For example, the most common CPU-heavy thing I do is compiling long books in LaTeX; a massively parallel version of LaTeX would be very cool, but I'm not holding my breath.

Re:Free Enterprise

[Crouching Turbo]

It's called POSIX, and everyone supports it except Windows.

Intel may be early

[Eideewt]

I do have my doubts about Intel's "more cores than you can shake a stick at" approach. I can't see the use in more than a few full-speed cores. They all have to be able to get at instructions quickly or most will just spin their wheels so hundreds of cores are a big challenge in more than just a making them fit and operate together sense. How much can we parallelize before most of the cores are doing little to nothing because their caches are empty? For that matter, the average user doesn't usually utilize one CPU core fully. Even on Dual-core (including actual dual CPU) desktop machines both cores are rarely needed for a responsive computer.

Intel's standpoint seems to be that there's a world of data crunching lurking in all our computers (automated photo sorting, face recognition, and photo-realistic rendering), but none of these strike me as killer apps waiting to happen. All are things we could get used to and come to depend on, but I don't think any of them are being held back just because of our computing capacity, although photo-realistic rendering may be close. I'm pretty sure these aren't solved problems yet. Even if we were itching to do all this, one can only sort so many photos. It seems a bit wasteful to have all that power waiting around most of the time. Are we really nearly living in a world in which computing power is so plentiful that we can have that kind of ability even though we hardly ever use it?

On the other hand, AMD's approach seems to have more immediate application. Video/audio encoding and other parallel processes are things that many of us do do frequently. A couple hundred cores could be pressed into use for this, but that seems much less elegant than purpose-built hardware.

I don't know which approach will be best in the long-run. Probably both. It does seem to me that Intel is at best a few years to early to be hyping large numbers of cores.

Yes, if those were competitors.

[Somatic]

The thing is, I like to be able to play games on my home computer. Any old game I see at the store. I want to be able to play it. A home computer is half about entertainment. Windows has no competition in that area. They just don't.

I use Linux daily at work, but, I have no driving need to have a Linux box at home. I don't do that much worky stuff at home. I'm already burned out after doing it all day at work (and if I need to do more work, I can ssh in with Cygwin from home). And you really can't game on Linux. Yes, I'm sure some games are compatible with various Linuxes (the only one that comes to mind is Puzzle Pirates, and that's because it's written in Java), and I guess there's Windows emulation. But let's be honest. You're not gaming on Linux.

Mac, well. I can honestly never see myself owning one. If I was going to go another OS, it would be Linux for its flexibility. Apple just makes me nervous with all its proprietary stuff. I know lots of people own Macs and are happy with them. But the number of programs in general that are written for Mac is too tiny for me, games especially.

I have a box of junk in the corner with a bunch of games in it, and out of curiousity, I pulled them out while writing this, just to see which were compatible with Mac right out of the box... not with some lameass windows emulation that will run it at 1/10 speed, but actual Mac compatibility. Everquest (and about 7 expansions): Windows. Yeah, there was a Mac version a few years ago, where you get to play on your very own Mac server. Good luck with that. GTA: Vice City: Windows. Knights of the Old Republic: Windows. Dark Age of Camelot: Windows. There are plenty more in the closet. In fact, I'm pretty sure that the only one in my entire collection that is compatible with Mac right out of the box is World of Warcraft.

But I hate World of Warcraft.

Sorry, but, my home box is my entertainment. And Microsoft knows it too. They agressively pursue the gaming angle with developer tools. Now they've got XNA, with the goal being to write a game once and have it play on both xbox or windows with a minimum of development fuss.

They know where their dollars are coming from. They court the market, because they know geeks like me would flee if Linux could entertain me even 1/100th as much. So what I'm saying is no, no there is no Windows competition, not in my market.

Re:Yes, if those were competitors.

[Anonymous Coward]

What? In order for Linux to have games, there needs to be a market for Linux games. That requires Linux gamers. And for that, you need Linux games!

Developers will jump through all sorts of hoops to make games for popular platforms. The only metric which matters is market share.

Simply false

[DMiax]

Windows supports POSIX: look here.

In any case you have a point in that Microsoft does not really encourages programming for POSIX-compliant OSs, but just for Windows.

Transputer? Don't think so.

[dabadab]

The transputer was an invention of Inmos and they were interconnected little CPUs designed for parallelizable task. Transputers had an own language, Occam where for every block you had to specify if its instructions were to be executed in serial or parallel manner.
It was a rather fascinating system (especially for its time) but it has died on the market.
(Sorry for the off-topic ranting, but I programmed these during my studies and quite liked the concept)