AMD's Graphics Core Next Preview: AMD's New GPU, Architected For Compute
by Ryan Smith on December 21, 2011 9:38 PM ESTWe’ve just returned from sunny Bellevue, Washington, where AMD held their first Fusion Developer Summit (AFDS). As with other technical conferences of this nature such as NVIDIA’s GTC and Intel’s IDF, AFDS is a chance for AMD to reach out to developers to prepare them for future products and to receive feedback in turn. While AMD can make powerful hardware it’s ultimately the software that runs on it that drives sales, so it’s important for them to reach out to developers to ensure that such software is being made.
AFDS 2011 served as a focal point for several different things going on at AMD. At its broadest, it was a launch event for Llano, AMD’s first mainstream Fusion APU that launched at the start of the week. AMD has invested the future of the company into APUs, and not just for graphical purposes but for compute purposes too. So Llano is a big deal for the company even though it’s only a taste of what’s to come.
The second purpose of course was to provide sessions for developers to learn more about how to utilize AMD’s GPUs for compute and graphics tasks. Microsoft, Acceleware, Adobe, academic researchers, and others were on hand to provide talks on how they’re using GPUs in current and future projects.
The final purpose – and what is going to be most interesting to most outside observers – was to prepare developers for what’s coming down the pipe. AMD has big plans for the future and it’s important to get developers involved as soon as is reasonably possible so that they’re ready to use AMD’s future technologies when they launch. Over the next few days we’ll talk about a couple of different things AMD is working on, and today we’ll start with the first and most exciting project: AMD Graphics Core Next.
Graphics Core Next (GCN) is the architectural basis for AMD’s future GPUs, both for discrete products and for GPUs integrated with CPUs as part of AMD’s APU products. AMD will be instituting a major overhaul of its traditional GPU architecture for future generation products in order to meet the direction of the market and where they want to go with their GPUs in the future.
While graphics performance and features have been and will continue to be important aspects of a GPU’s design, AMD and the rest of the market have been moving towards further exploiting the compute capabilities of GPUs, which in the right circumstances are capable of being utilized as massive parallel processors that can complete a number of tasks in the fraction of the time as a highly generalized CPU. Since the introduction of shader-capable GPUs in 2002, GPUs have slowly evolved to become more generalized so that their resources can be used for more than just graphics. AMD’s most recent shift was with their VLIW4 architecture with Cayman late last year; now they’re looking to make their biggest leap yet with GCN.
GCN at its core is the basis of a GPU that performs well at both graphical and computing tasks. AMD has stretched their traditional VLIW architecture as far as they reasonably can for computing purposes, and as more developers get on board for GPU computing a clean break is needed in order to build a better performing GPU to meet their needs. This is in essence AMD’s Fermi: a new architecture and a radical overhaul to make a GPU that is as monstrous at computing as it is at graphics. And this is the story of the architecture that AMD will be building to make it happen.
Finally, it should be noted that the theme of AFDS 2011 was heterogeneous computing, as it has become AMD’s focus to get developers to develop heterogeneous applications that effectively utilize both AMD’s CPUs and AMD’s GPUs. Ostensibly AFDS is a conference about GPU computing, but AMD’s true strength is not their CPU side or their GPU side, it’s the combination of the two. Bulldozer will be the first half of AMD’s future APUs, while GCN will be the other half.
83 Comments
View All Comments
DoctorPizza - Monday, June 20, 2011 - link
I can't understand that at all.The next architecture will have 16-wide SIMD. How does that fit computational problems better than a 16-wide MIMD VLIW architecture? VLIW can act as if it were SIMD if necessary (simply make each instruction within the word the same, varying only the operands), so how on earth can SIMD be better? SIMD is strictly less general and less flexible than VLIW. This makes it applicable to a narrower set of problems--if you have problems that aren't 16-wide, then you're wasting those additional ALUs, and there's nothing you can do with them, ever. MIMD can't always use them, but there the restriction is unbreakable dependencies, not an inability to encode instructions.
And while VLIW heritage is indeed statically scheduled, nothing about VLIW mandates static scheduling. The next generation Itanium will use dynamic scheduling, for example.
This whole article reads like AMD has offered a rationale for its architectural change, and the author has accepted that rationale without ever stopping to consider if it makes sense.
DoctorPizza - Monday, June 20, 2011 - link
(FYI: the *real* reason to go for SIMD instead of VLIW is simply that VLIW takes up more die area. AMD has decided that the problems people are working on have enough data- and thread-level parallelism that it's not worth having extra decode logic to enable extraction of more instruction-level parallelism.The result is a design that's actually *worse* for general-purpose computation--for non-vector computations, it'll only ever use one of those sixteen ALUs, whereas the previous design could in principle use them all--but better for embarrassingly parallel workloads.
Why the article couldn't say this is anybody's guess.)
Quantumboredom - Tuesday, June 21, 2011 - link
I don't understand your argument. They have moved from 16-wide SIMD where each instruction is a 4-operation VLIW (where there are quite a few restrictions on what that VLIW instruction can actually be) to _four_ 16-wide SIMDs where each instruction is scalar. The new architecture is in every way more general and more suited to a wide range of computational problems while retaining the same power. It does presumably cost more (in terms of area/transistors), but hopefully it will be worth it.DoctorPizza - Tuesday, June 21, 2011 - link
Where does it say that Cayman SPs are ganged into groups of 16? It says they're grouped somehow, but never makes the claim that their groups are as wide as the new SIMD short vectors.Quantumboredom - Tuesday, June 21, 2011 - link
It is well-known that Cypress and Cayman both have arrays of 16 processing elements operating in SIMD mode, and they have to execute work-items from the same work-group over four cycles, leading to a wavefront size of 64. See for example the AMD APP OpenCL Programming Guide 1.3c section 1.2 where this is described. Specifically it says "All stream cores within a compute unit execute the same instruction sequence in lock-step".DoctorPizza - Tuesday, June 21, 2011 - link
"well-known"? I assure you, the vast majority of people have not read AMD's OpenCL Programming Guide.Nonetheless, the article still makes little sense.
A vector of 16 instruction-parallel processors is more versatile than a vector of 16 strictly SISD ones. In the worst case, with unbreakable data dependencies, the former degrades to the latter. In the best case, the former can do 4 (VLIW4) or 5 (VLIW5) times the work of the latter. The average case cited in the article was about 3.5 times.
If you only had one thread of work, the old architecture would tend to be better. For every 64 ALUs (one old VLIW vector or four new SIMD vectors), a single-threaded task would average usage of 56 out of 64 ALUs (3.5 per VLIW) on the old arch, but only 16 out of 64 on the new.
However, AMD is plainly counting on there being many, many potential threads. If you have abundant threads then you can guarantee that you can fill up the remaining 48 ALUs with different threads, whereas the 8 unused ALUs in the VLIW arch are off-limits.
This is a less general architecture, but as long as all your problems are massively parallel, creating all those extra threads shouldn't be a problem. AMD is sacrificing generality in favour of the embarrassingly parallel.
Quantumboredom - Tuesday, June 21, 2011 - link
I actually asked a similar question at the AMD Fusion Developer Summit.The minimum number of wavefronts (i.e., batches of 64 work-items) needed to keep a Cypress/Cayman CU fed is two, while GCN requires four wavefronts (so twice as many). However it is the case that quite often (for all of my programs actually) you really do need four wavefronts per CU on Cypress/Cayman to effectively hide the global memory latency. The guy I was talking to at AMD seemed to thnik that in practice the number of work-items needed would stay about the same between Cayman and GCN for most applications.
I've asked this question on the AMD developer forums as well, but I don't know how many answers will be given about GCN there.
DoctorPizza - Tuesday, June 21, 2011 - link
I certainly wouldn't be surprised to hear that typical GPGPU workloads could inundate the GPU with threads and so provide more than enough wavefronts. The GPGPU workloads are pretty much all of the embarrassingly parallel kind, so creating more threads should tend to be pretty trivial.So your experience certainly makes sense with what I'd expect.
It's not that I think this is necessarily a bad change for the applications that people use GPGPU processing for.
It's more that I'm disputing the implication that this somehow makes the GPU more general and easier to take advantage of; to my mind it's doing the exact opposite of that.
Or to put it another way: virtually every single program has a reasonable amount of instruction level parallelism. Data-/thread-level parallelism is much rarer. We're losing the former to improve the latter.
For problems amenable to massive thread-/data-level parallelism the result should be substantially more ALUs available to process on. But for problems with only limited data-/thread-level parallelism, it's a step backwards.
name99 - Thursday, December 22, 2011 - link
"The next architecture will have 16-wide SIMD. How does that fit computational problems better than a 16-wide MIMD VLIW architecture? VLIW can act as if it were SIMD if necessary (simply make each instruction within the word the same, varying only the operands), so how on earth can SIMD be better? SIMD is strictly less general and less flexible than VLIW."A VLIW system has to have instruction decoders and routers for every instruction, and thus for every data item that is processed.
A SIMD system only has to have one instruction decoder and router for every 16 data items that are processed. If your computations consist primarily of doing the same thing to multiple data items this is a win. (More processing for less power and less silicon.) If your computations do NOT consist primarily of doing the same thing to multiple data items, it's a loss.
Or, to put it differently, is it worth investing silicon in moving instructions around with great facility, or is it better to invest silicon in moving data around with great facility? Seymour Crane thought (for the problems he cared about) the answer was data. I'd like to think AMD know enough about what they are doing that they have the numbers in hand, and have calculated that, once again for them the answer is data.
MySchizoBuddy - Tuesday, June 21, 2011 - link
where is the information about the toolkit to take advantage of this hardware?