AMD and ARM join forces at last
Fusion 11: Not what you think, much much more.
Jun 22, 2011 in Chips, Finance, Graphics, Memory, Microprocessors, Mobile, Opinion, Rumors
Anyone watching the industry knows AMD (NYSE:AMD) and ARM (NASDAQ:ARMH) are up to something together, it is more obvious than two teen-agers giggling when they glance at each other. What that something is is not what anyone expects though.
At AFDS/Fusion 11, both AMD and ARM said it, and said it explicitly, but no one seems to have noticed. That is a pity because what they are doing is really cool. No, I would go as far as to say it is technically brilliant and also the only thing that they could have done. What is this wondrous thing? Take a look at these slides from AFDS, the first four are from Phil Rogers’ Keynote, the last two are from Jem Davies’ later in the same day.
If you read these slides together and still don’t get it, hold still, I have a hammer^h^h^h^h^h^heducation rod that I am glad to help you over the head with. Just cover travel expenses, and this form of consulting and gene pool upgrading is free. Act now. While I am waiting for the checks to clear, I can explain things in a slightly less, umm, impact-ful, way.
Look at the fist slide, the one about FSA and Open Platforms. Think about that, open platforms abound now, you could say that AMD’s current Hypertransport system is one of those, and others have similar plans. ARM itself is very open, as long as you are a licensee.
The ‘Fusion’ concept is not about platforms in the sense of things stuck on the motherboard, it is about the exact opposite, things stuck on a die with a bit of careful sub-micron welding. An open platform in this case would mean that the interfaces ON the chip itself, not TO the chip, are open. This is key.
The second bullet, ISA agnostic, and the third, or at least the hardware partners part, are the other relevant pieces. Why? Think about it, why would AMD need to be hardware agnostic? They have one set of hardware interfaces, and they control both sides, right? Why would hardware partners care about FSA if AMD meant off-chip? There is this little thing called PCIe, and it works more than well enough, especially with the 3.0 spec additions. On chip, they would be in bed with AMD enough to make designing for whatever spec AMD presented rather trivial.
Should there be any little bits to paper over in early hardware, that is what the next slide, and the next technology, FSAIL, is for. With luck, it won’t be needed in a generation or two, but the first spin or three of some hardware may need it. Think of it like a BIOS patch for future Fusion architectures.
On the next slide, AMD says the finalizer can re-order memory. Why would you need this? AMD CPUs are all bound by the x86 spec for memory ordering, broken, backward and painful as that may be. The ATI GPUs may use a different scheme, but that is an easily fixable problem, and AMD has explicitly said that this will change in Southern Islands, coming in a few months. For an all ATI design, this requirement is silly.
If ARM uses a different model, they will be consistent with their IP partners too, so this is a pointless spec, right? Unless you want to to mix components made for the two architectures, then it would be necessary. But mixing and matching components for both sides can’t happen, they are two different ISAs, and…. oh, first slide, bullet 2. Hmmmm……
But that would not be necessary if you were making a chip, you could just put a few gates in to flip the memory order with very low latency, right? Unless you wanted to make the whole scheme to be software agnostic, but software tends to be for a platform, if not a specific chip, right? The only reason for this would be to run low level software on multiple disparate platforms, something that is madness. Oh, bullet point 1, C++ and higher level language compatibility. Gotcha.
Then there is the final one of Phil Roger’s slides, the future directions one. It simply screams “we are going to open up our chip”. How much more direct do you need?
If that isn’t enough, look at the first one of the ARM slides, the one about memory model consistency. If you have an IP block for ARM CPUs that need very low level and granular access to memory, particularly in a low latency but consistent manner, you can design it easily. ARM will be more than happy to document their scheme to any partner that needs it, it isn’t top secret in any way.
Now why would ARM need to worry about such things if all their partners have one target, and only one target, the ARM core itself, to write to? Their GPUs also don’t need to talk to multiple memory models either. The only reason you would need this is if you wanted to mix and match, on the same chip, IP blocks from two or more different companies with disparate memory models. But something like that would be madness, the memory models between ARM and x86 are….. hey, wait a minute……
And then there is the last point on the last slide, standards. To quote the slide, “We need standard(s) – And a very small number of them.” In this case, two is probably too many, one is just about right. Silicon takes a while to make, much less standardize, so if there were only one standard in the making, the current and near future chips wouldn’t work, and no one would write software for them until they had a huge installed base.
No one would be crazy enough to preemptively write code for that kind of future model until it arrived. The whole concept is DOA. To make it work, and to get people interested, you would need something to paper it all over, and bridge the gap. It would have to be a low latency shim to flip bits here and there, like for memory ordering. It would have to be almost a virtual ISA for parallel programs where the software is ‘finalized’ to an ISA by a JIT compiler or ‘Finalizer’. Madness I tell you, it will never hap… hap…. Second slide, top bullet you say?
In summary, while neither side has explicitly said this directly, it is painfully clear that AMD and ARM are going to make a common on-chip interface and interconnection. You can take a Bobcat chip, yank the GPU, and slap a Mali or Imagination block in. Want a quad-core A15 that runs ATI demos? A shoelace tip controller that bootstraps a 6970? How about a Bulldozer or Trinity that uses that custom accelerator block that Google or Facebook is probably working on? Crypto blocks that are common to ARM and AMD? No problem. Get the picture now?
There is one other problem here, ARM cores and IP are the easy part. They are synthesizable with a small number of custom arrays and easily portable across process technologies. You can make them just about anywhere, and you can make them with a lot of different tools. You can’t do this with AMD CPUs, they are just the opposite, Right?
To make this whole crazy scheme workable, you would need an AMD core that is synthesizable with a small number of custom arrays, and it needs to be easily portable across process technologies. This is a non-negotiable point, it must not just be in place, it needs to be in place long long before you expect to talk to customers and partners, much less expect silicon on the market. Long long before means you need to start designing the core with these goals in mind years before you see any fruit from the project.
Anyone remember the very last talk at Hot Chips 22? Anyone remember slide 3, reproduced below?
Zing! Score!
If the above explanations aren’t clear enough, there are three choices open to you. First, you can go watch the keynotes, click here and sign up. If that doesn’t do it, you can pay me lots of money to come out and explain it to you, I do consult for a number of large, and small, companies. Barring that, for a lot less money, I can bring the SemiAccurate Rod of Reeducation, affectionately called ‘Thud’ for reasons I can’t figure out, to your location. Call now, operators are standing by. This limited time offer will close when the first inter-mixable silicon hits.S|A
Note: If there is anyone who thinks Dirk Meyer didn’t have a consumer electronics strategy, or that it wasn’t in place long ago, well, you were wrong.
50 Responses to “AMD and ARM join forces at last”
Trackbacks/Pingbacks
-
-
Jun 24, 2011
[...] [...]
-
-
Jun 24, 2011
[...] [...]
-
-
Jun 23, 2011
[...] AMD and ARM join forces at last – SemiAccurate Charlie Demerjian makes a compelling case… [...]
Excellent work Charlie. I know I’m late to the comments, I found this by accident but from now on I will be following your writing. I’m not sure I get everything you wrote yet so I’ll take another look or two later.
I am curious of two things. How does this fit in with Amd’s vision for integrating the CPU and gpuFrom what I understand, they plan on integrating parts of the gpu into the CPU, and it won’t look anything like llano?
And with windows 8 supporting arm, could we see a possible fusion chip with an Arm core for additional power savings?
Been traveling and just saw this – BRILLIANT !
With regard to “Thud” could you bring it here and demo it to a couple of our people… :-)
Jon
IIRC AMD GPUs are (or were to be) manufactured at both TSMC and GloFlo… two different process techs.
Lessons learned there could be applied to CPU manufacture.
AMD cannot transfer x86 IP under it’s x86 license though (correct me if I’m wrong)? Thus AMD could not adopt ARM’s design licensing model for it’s CPUs. Nothing stops AMD licensing it’s GPU IP or designs. Nor licensing ARM CPU designs for low power (say sub-watt) ‘Fusion’ products.
I worry because AMD has a touch of the Novell disease: produce some really really good tech every few years and then spend the next few years navel gazing.
If AMD produces fully synthesizable GPU and CPU tech, will it then engage in an internal debate or war about whether it is a high-TDP video-card/CPU company, or a mid-TDP Fusion company, or a company capable of addressing the ‘processing market’ from cell phones to HPC clusters? …and waste it’s advantage. Or will it pursue the advantage the tech gives it?
I think it will come down to quality of management, because the engineering talent seems to be present already.
Maybe hire a Pat Gelsinger?
Pat turned AMD down.
John Taylor, Director for Client-Systems at AMD denied officially that AMD intents to develop ARM-based Fusion-Chips! However, there might be a different kind of cooperation: What if ARM provides the CPU-core and AMD the graphics as well as the interface to stick all together? This could also make sense if we consider that AMD is already selling IP to IBM for fusion like Cell-Chips for game computers. So perhaps we should not expect Tegra-look-alikes from AMD but from other companies who combine the cores from AMD and ARM for tablets, smartphones and other gadgets.
That’s pretty much it.
AMD is cashing in on the GPU expertise they have by selling other ARM companies GPU related services.
The brand recognition AMD has in game graphics helps the ARM guys too. Nvidia probably won’t license their tech to competitors, so they are stuck with PowerVR or something else which doesn’t have the same cachet.
another superb delivery from charlie.
It is not even possible to utter “memory interface”, let alone “memory interface standards” without the rabid likes of RAMBUS drooling.
Maybe Dirk was looking down the road too far as far as the board was concerned. “We want tablet processors nao! Stop working on future stuff, and turn our limited resources to turning Bobcat into a tablet chip!”
??
You know, I kind of wish they didn’t dirk Dirk. That guy was good for the company.
“Note: If there is anyone who thinks Dirk Meyer didn’t have a consumer electronics strategy, or that it wasn’t in place long ago, well, you were wrong.”
So they why did Dirk get canned?
great article charlie!
charlie has said in the past that tegra was dead, but it looks like it’s going quite well? so i think the other ARM partners like TI and Qualcomm, will want Radeon graphics bad, so that they can use them to compete with nvidia. could be a lot of business :-).
i wonder how much amd can cut their GPU down for cell phones? bobcat got 80 shaders, maybe a cell phone would get 10? does that make any sense? no idea!
Well qualcomm already kinda sorta have amd/ati onboard, since amd sold their mobile chip division to them in 2009. The adreno GPU in snapdragon is descended from ati imageon. Not radeon, I know, but there is a link there.
right, but it probably sucks compared to tegra, won’t support any GPGPU standards, and there are still a bunch of other ARM makers that don’t have access to good graphic tech with GPGPU.
IT’s completely insane.
And I like it :)
Frankly Charlie this is a really great article.I have been a fan of yours since the days of the Inq.and this is up to your best.Please keep the keen insights and fine writing coming. It is indeed a pleasure to read.
Did Dirk lose his job on this?
So… Does this explain the strange statement from Apple about leaving Intel for the portables field?
He shoots! He scores!
Seriously, I’m not sure I get everything you are trying to relate Charlie (little chips which can run anything written for ARM/AMD/ATI/Mali as well as a completely common ISA?), so I’m going to reread this article a few more times. Good darn work (you too AMD/ARM) :)
Now for a series of articles on nVidia and Chipzilla panicking :D
AMD is trying to get their graphics into ARM cored products, and they will build the tools to make moving between platforms less painful.
I’m not sure if this is just for graphics, or it will be for both graphics and computations. It could end up being for both since the entire point of high-level programming languages is portability between hardware.
But is AMD’s GPUs low power enough for the devices that ARM SoCs target?
I had the impression that, for example, ImgTec GPUs have been designed from the ground up to be low power, while AMD’s goals for their GPUs have always been high performance (in competition with Nvidia).
a true hetrogeneous APU, should have the best of all worlds including a RISC core.
Also it will DWIM!
Questions
- When will I be able to get a “Fusion PC”
- I’ve used CUDA in the past. A main drawback is that you have to manage host and device memory. This is gone with Fusion, right? How will I be able to program this? (Framework, API, etc)
well, you can get a brazos pc, which is fusion, now :-). but i guess llano is still gonna be a couple weeks or who knows what?
i’m not sure, is the plan to use openCL or c++/java/.. to write GPGPU code? the latter, right? so then are they canning openCL?
Nils-
With CUDA 4 (64-bit), this is no longer the case. There is one memory space for both.
Nils – AMD is pushing OPenCL pretty hard. The Stream SDK (or whatever AMD is calling it now) also supports ‘IL’ which is a rather assembler-like language but only for the GPU side.
Be more condescending .. but still a great article.
A solid piece of tech journalism, given that almost every move made by AMD in the last 2 years has been as good as they can make it with the ressources on hand. The AMD board should be ashamed of their actions in letting Dirk go…
AMD isn’t big enough to dictate an AMD-only “standard”. Like AMD64, they want and need the competition (Intel, Nvidia) to adopt it.
AMD64 won because it an evolutionary aproach to the problem of adding hardware 64-bit computing unlike Intel’s IA-64. Intel woke up when AMD had the specification done and could only adopt it.
Charlie, though it’s not been cited for a long time, I recall AMD was really proud of them achieving llano’s design as a mixed process where the GPU core is synthesizable, the CPU likely not. I’m waiting for the day of heterogenous CPU’s (AMD & ARM) and my laptop acting as a standby phone and music and video player using Arm IP, while consuming near zero resources and then when I need it, powering up the x86 high performance trinity (or future) cpu.
No surprises here. Rational extrapolation of current trends. Of course, it makes sense to build Fusion modularly, so that mix/match is not only possible, but commercially feasible.
A huge win-win for AMD and ARM’s licensees.
Maybe one chip company alone can’t beat Intel, but what if most of them collaborated?
it’s be about time.
With Tegra 2 doing well, Tegra 3 on the way with even higher projections, and Project Denver supposedly happening, it is clear that AMD has to do something with ARM as well. Being that Nvidia’s had this much time with their Tegra platform, it would be an obvious move for AMD to try and make their CPU’s and GPU’s easier to integrate and jumping on the ARM bandwagon is apparently how they’re doing it. The benefits are clear and it’s actually quite odd that they don’t have any products available yet.
The thing I don’t understand is why they seem to be carrying “we’re on the cutting edge” badges when others have crossed that bridge long ago.
Because unlike nVidia, AMD is focusing more on establishing the underlying infrastructure, not just a product. They want to be part of the team that sets the rules, not just someone who just sells a chip.
Well Said.
I’ll add that they have vision and the desire to take computing somewhere much further than we are right now.
FSA(IL) reminds me somewhat of Larrabee’s goals.
This market is getting crazier by the day.
Wonder what the chaps in Satan Clara are thinking bout this :D
Hmmm… Windows 8 on ARM with a Radeon graphics core anyone?
with 7xxx ati graphics :)
I guess you must be content with Kepler for a Looooooong time before a 7xxxx card embedded with it , more probably it would be a 8xxx vs Maxwell
‘Thud’ instead of ‘Mjolnir II’? shame, lol
Charlie, You are either nuts or brilliant. I guess time will tell, but a great article nonetheless. And if this happens, and it works (no guarantees on that), Intel would really hate it. Maybe that’s enough of a reason :-)
I second that. Brilliant read and good to see technology at least being a little bit more open. I hope your right charlie
Insightful and investigative.
The next step needs to be a compiler back-end for GCC. (This covers C, C++, Fortran, Pascal, and Ada), and finalizers for (at a minimum) x86-64 and Radeon video cards, say post 3000 series. Adding ARM, and tablet graphics chips is something for ARM to do, and the benefits accrue to everyone who participates.
Can nVidia join the crowd? Probably. IBM has some really good compiler and development tool guys, and some CPU architectures to target. AMD can tune their finalizer to do best on their CPUs, particularly with the SIMD floating-point instructions. Intel would then need to produce one tuned to use their SIMD units best, but that leaves Intel’s graphics at the starting gate.
Why? The current GPU in Sandy Bridge chips is not well designed for CUDA or OpenCL code. Worse, it doesn’t do double precision floating point at all.
LLVM is already setup to do what you suggest, and you wouldn’t have to mess around with the GCC codebase.
It works like this:
Source Code –> LLVM(Bytecode [At this point you could export the bytecode and use the rest of the LLVM pipeline as a JIT on the target ISA] –> Machine Code)