Amd gpus to support cuda?

[monstercameron]

Or am I seeing things?
http://www.anandtech.com/show/9792/...e-announced-c-and-cuda-compilers-for-amd-gpus

 

[ViRGE]

Probably better to say "AMD GPUs to support CUDA for high performance computing (i.e. servers)".

The AT article is very explicit that this isn't a mean to run compiled CUDA code, which is what games and consumer applications would be.

 

[monstercameron]

I'm surprised that amd is making another compiler since they abandoned open64. So now they are supporting gcc, llvm/clang and the new hcc. Seems like a lot of work.

 

[Dribble]

It's more an admittal that just trying to support OpenCL is getting AMD no-where in the HPC market as OpenCL is always behind the CUDA curve. Hence they are adding their own libraries and extra support to bring them up to CUDA like levels, once they support the same feature set it's possible to write something to convert from one to the other.

However this is all just power points right now. The amount of effort to write the libraries/debuggers/etc to do all their own HPC libraries properly is huge, and tbh would seem to be well beyond what AMD can realistically hope to achieve. Hence I don't see how they are going to get anywhere alone?

I mean if you are using CUDA, which has much better support (both from nvidia but in 3rd party libraries), nvidia has the better hardware, the proven track record, etc why on earth would you buy AMD cards and then attempt to use conversion software to convert all your working CUDA source to run on some pretty untested AMD libraries? You'd just buy nvidia cards and be done with it - HPC programming is hard enough without adding a whole pile more complexity.

 

[Erenhardt]

Nope, amd gpus will not support cuda. AMD wants to provide tools to easily port from cuda to their own compiler HCC based on c++.

They seem to concentrate on way too many things at once with such slim budget. They are really pushing innovation on all fronts. The big question is how long will they have steam to push it.

Loosing a legal battle with nv could be the end of amd

 

[Azix]

Loosing a legal battle with nv could be the end of amd

you mean nvidia suing over the CUDA thing? CUDA apps are made by developers, not really sure how allowing them to convert their apps could give nvidia reason to sue. They could try

 

[Headfoot]

Its more like "AMD GPUs to support compiled code from a source-to-source translator" which is good but not the same

 

[Headfoot]

I can guarantee if some dudes in a forum are thinking that there might be liability, AMD has already gotten a full clearance opinion from very smart, very well connected, very expensive lawyers and they would not have proceeded if they couldn't have taken the risk. This is a multi billion dollar enterprise we're talking about here after all...

 

[SPBHM]

Or am I seeing things?
http://www.anandtech.com/show/9792/...e-announced-c-and-cuda-compilers-for-amd-gpus

"To be clear here, HIP is not a means for AMD GPUs to run compiled CUDA programs. CUDA is and remains an NVIDIA technology. But HIP is the means for source-to-source translation, so that developers will have a far easier time targeting"

 

[monstercameron]

"To be clear here, HIP is not a means for AMD GPUs to run compiled CUDA programs. CUDA is and remains an NVIDIA technology. But HIP is the means for source-to-source translation, so that developers will have a far easier time targeting"

I read the release, that question was more about the politics than the technicalities.

 

[railven]

I read the release, that question was more about the politics than the technicalities.

Simple politics, they need to make their product more attractive. Nvidia has a huge chunk of that ecosystem, and AMD "was" cut off until now.

 

[ViRGE]

I read the release, that question was more about the politics than the technicalities.

Keep in mind that this is a tech forum. Technicalities are what we live for.

you are technically correct -- the best kind of correct.

 

[xpea]

AMD seems to be really desperate :\

Like Erenhardt said, no way they can execute their big software plan with such low R&D budget. They are chasing too many targets. It's all powerpoint announcement :thumbsdown:
Especially when you know that current AMD workforce is a bunch of fanboy veterans and lot of over promoted interns without enough experience for the job they have to do...

 

[NTMBK]

I'll believe it when they deliver something useful.

 

[dacostafilipe]

Especially when you know that current AMD workforce is a bunch of fanboy veterans and lot of over promoted interns without enough experience for the job they have to do...

:thumbsdown:

You have no idea. Just look at what's happening with AMDGPU. Great work done there!

This is too far away and too limited to a small aspect of CUDA (only runtime API, not driver API) but it could be great for them if this works.

 

[xpea]

:thumbsdown:

You have no idea. Just look at what's happening with AMDGPU. Great work done there!

Sorry ?
Where's AMD in the TOP500 ?
Where's OpenCL going on when his biggest promoter officially gives up for CUDA ?
Hardware is only one small part of the equation. Without software, support, ecosystem, and promotion it's useless...

 

[dacostafilipe]

Sorry ?
Where's AMD in the TOP500 ?
Where's OpenCL going on when his biggest promoter officially gives up for CUDA ?
Hardware is only one small part of the equation. Without software, support, ecosystem, and promotion it's useless...

This nothing to do with the software team. It's a management problem.

 

[Genx87]

This makes more sense than most of the stuff coming out of AMD HQ. This is a high margin market.

 

[ThatBuzzkiller]

Sorry ?
Where's AMD in the TOP500 ?
Where's OpenCL going on when his biggest promoter officially gives up for CUDA ?
Hardware is only one small part of the equation. Without software, support, ecosystem, and promotion it's useless...

AMD isn't even giving up OpenCL for CUDA, their attempting attrition to Nvidia's software ecosystem ...

 

[VirtualLarry]

Or am I seeing things?
http://www.anandtech.com/show/9792/...e-announced-c-and-cuda-compilers-for-amd-gpus

But more importantly than that, the headless Linux driver will be implementing an HSA extended environment, which will bring with it many of the advantages of the Heterogeneous System Architecture to AMD’s FirePro discrete GPUs. This environment, which AMD is calling HSA+, builds off of the formal HSA standard by adding extensions specifically to support HSA with discrete GPUs. The extensions themselves are going to be non-standard – the HSA Foundation has been focused on truly integrated devices ala APUs, and it doesn’t sound like these extensions will be accepted upstream into mainstream HSA any time soon – but AMD will be releasing the extensions as an open source project in the future.

Quite the task, seeing as their current "Linux AMD Proprietary Drivers" WONT even let you run BOINC's GPU compute on Linux (Ubuntu-derivatives, like Mint).

http://forums.anandtech.com/showthread.php?t=2454773

If they can't even get that working, then how the F are they going to get headless compute servers and HSA+ and all of that stuff working?

I find this potentially hilarious. War of the powerpoints, they're as bad as GlobalFoundries.