Hardware AMD Betting Everything on OpenCL

AMD discusses how to make GPGPU computing mainstream, as well as the potential for ARM CPUs.

http://www.bit-tech.net/hardware/cpus/2011/05/30/amd-betting-everything-on-opencl/1

 

It depends largely on the software-engineers to write software that isn't x86-based/dependant, but seeing that most of the software today didn't even get transitioned to x64 I'd say that a change to GPGPU and OpenCL is wishful thinking for the next 10 years to come.

Hardware-capabilities have increased tremendously and they improve so fast as of today that software can't keep up with that speed. This is due to software being developed on current hardware available and software being developed for the majority of systems currently around.

So we need atleast to wait another five years, before the old hardware not capable of OpenCL, x64 etc isn't used anymore by the majority. Then we'll see new software being developed for those new hardware.

 

Thx for an excellent article! Give us the other side of the story now guys, please do a similar interview with representatives from nvidia!

 

Good interview.

I don't think this will work out for them... although I hope it will.

 

Oh dear!

Intel kick's AMD's arse.
ARM give's intel a good kicking in the mobile space.
AMD says "ARM who?"

 

Inb4 consoles all use old hardware

 

I think you've probably read a bit too much into that myself. As stated in the article, Terry Makedon is AMD's manager of Fusion software marketing - I'd say it's simply not his job to be intimately knowledgeable of other CPU architectures.

I'd be willing to bet that most people who live outside of the Southern Hemisphere couldn't name Australia's Prime Minister or tell you how he's performing*, that doesn't mean they've never heard of Australia...


*He's a she actually, (Julia Gillard).

 

from what I've seen in just plain dictionary attacks.. an optimized ati card using opencl does kick cudas ass- think it's because they have more shaders than nvidia.. the thing is cuda has been courted and so you see things like folding totally dominated by nvidia

my opinion anyways.. I've seen like triple the performance, so maybe amd has something here- once optimized might put an end to cuda.. intel is going to have a fun time with 22nm parts

 

I don't really see how AMD can hope to gain an individual advantage by pushing an "open" standard.

Intel will have DX11 capable Ivybridge by Q2 2012 and so directly compete in gpgpu applications against them. Of course how well they perform comparatively will be debatable.

Even Nvidia without a cpu will benefit with their desktop version of Optimus...is it Sync? I for one would quite happily run a sandybridge for lower power graphical tasks and have a large discrete Nvidia gpu basically turned off in the background just waiting to deliver its grunt for a gpgpu task....of course with the additional option of having more considerable grunt than any APU.

And further down the line Nvidia Denver project products will be gpgpu monsters which will completely blow away AMD's and Intels hybrid cpu/gpu offerings in terms of gpgpu.

AMD have a small window of opportunity over the next 9 months in the lower powered sector but sadly for them the software environment is not there now to support it.

 

"Obviously, our products are going to be 32nm when we launch Llano, and we're going to be on an equal-footing from a technology perspective, but I think GlobalFoundries is working on a new technology process and making it better every day."

Not for nothing, but since everything I have seen from AMD comparing their best vs Intel's Core i5 Sandy Bridge, ignoring Ivy Bridge's future existence, it sounds like they are hoping that GlobalFoundries' tech is close enough to Intel's beastly i7s that they dont have to worry.

 

i'm an amd fan but anyone else notice that they explicitly ignored the question regarding "what if opencl doesn't take off - will the cpu portion of the APUs be sufficient?". that doesn't make me feel too comfortable.

as said in this interview, hardware has to be optimized for opengcl. like sure you could run some really old nvidia card on an opencl program made by amd and it will run better than just using a cpu by itself, but the point is that nvidia card won't have the optimizations. also, intel is being a huge contributor to opencl so its not like either company is losing anything - i'm sure both amd and intel realize that being proprietary about both software and hardware is going to backfire. cuda didn't get popular just because it was proprietary, it got unpopular because nvidia didn't even let you run another type of gpu. amd doesn't really give a crap if you're using someone else's product with theirs. i'm sure the way they see it is "give anyone the option to use our products and that will increase our chances of getting customers. as long as someone is using us for something, we're happy".

 

Personally I'm surprised people are not seeing the ties between the old school matchup between Microsoft - IBM. Bill Gates saw there was greater value in the software side of things than the selling of the hardware. IBM was wanting to OS basically to be there so people could use the hardware, which is what they were pushing while Gates saw the software as being the key to selling the hardware. Yes, Intel/AMD will not be making any royalties off of the software as its an open standard, however if their hardware supports it best it will sell better than nVidia's. Ignoring the infighting that occurred with the OpenGL group, OpenGL was a superior API over DirectX for the same reasons as OpenCL is superior to CUDA - it works across all OSes.

 

1)Well from this its obvious that AMD APU is weaker clock for clock then Sandybridge (with out OpenCL).
2)AMD better get OpenCL to work better then CUDA did for net books ION 1 is still very buggy. Flash plays and sometime just stutters as it decided midway through a video to us the CPU and not the GPU. If AMD get this OpenCL to work we still do not know if it will Beat or even match Sandy bridge as Intel will obviously also use the 3000 GPU on the sandy bridge chips to do the same.
3)Nvidia does CUDA well not only because of software but their Hardware design as well which is much more complex per Core then ATI. It will be very interesting to see how this pans out.

From the article it does sound like the the overall fastest system will be Intel/Nvida and sadly not AMD/AMD.

 

True, but remember, this is ARM architecture. It might be a freaking monster...without directly competing against AMD/Intel. I was under the impression Nvidia was going for HPC applications with Denver, and that Tegra would continue to be their consumer-level hardware. If that's the case, it won't matter to the average citizen - it will only be running on hardware they'll never see. There could be some trickle-down knock-on to the consumer space, but not direct competition.

As to what AMD has going on now, I hope it pays off. It's been a sadly long time since they've been toe-to-toe with Intel.

 

Isnt CUDA based of OpenCL.. and Nvidia talk up CUDA's ability to handle OpenCL applications.

 

How many times will AMD throw their Marketing BS to get you all talking about it but for them to disappear when the product is out and turns out to be a turd burger again? or then moan that all you reviewers are wrong and that their product is great anyway?

Give us independent benchtests or STFU AMD!

ps. How was buying ATi a gamble, when it's the only division turning a profit for the last 3 years? Making your losses seem not a great!

 

how many times are you going to moan and complain about every amd article ever released on this website regardless of what the topic is about? it was an interview about amd joining an OPEN SOURCE project that even intel and nvidia can take advantage of. the only comment amd made about their own products being superior is due to a question bit tech specifically asked, and its pretty obvious that a company is not going to say "o ya our product is very unlikely to compete with what is already out there" whether it is true or not. seriously, stop whining. this article isn't about competitive performance, it's about improving overall performance on any platform, but more specifically amd platforms.

 

I'm not quite so sure it has much to do with AMD gaining, rather AMD preventing themselves from falling behind. If AMD come out with their own proprietary solution when CUDA already has a comparatively large share in the GPGPU market, why would someone choose one over the other? However, if AMD support an open standard then they have greater opportunity to claw back market share.

I think that open standards have greater footing these days, as other OSs become more widespread and accepted and companies seek to reduce costs. This will undoubtedly play into OpenCL's hands this time around after the eventual slowing-to-a-crawl that has been OpenGL support after DX9-11.

 

Indeed. In fact it seems out of place to be discussing Intel offerings in the cpu department other than because they are both in the same game ultimately.

 

APUs - Armoured Personnel Units