ATI Stream SDK 2.0 beta 2 lacks OpenCL GPU support

Author: Tim Smalley
Published: 10th August 2009
Comments (10) Stumble
ATI Stream SDK 2.0 beta 2 lacks OpenCL GPU support

AMD's ATI Stream SDK v2.0 beta 2 doesn't support GPU acceleration for OpenCL and instead only supports CPUs at this time.

Despite releasing an updated version of its ATI Stream SDK last week, AMD still doesn't support GPU acceleration for OpenCL applications. The current release, version 2.0 beta 2, only supports x86 CPUs with support for SSE3 (or later) and GPU support isn't included.

Unfortunately, we wrongly reported that the SDK included support for GPU acceleration and, upon speaking to AMD, we have managed to get a better understanding of the company's plans.

AMD said that this release was "purposefully targeted at the CPU and the SDK portion of CAL (Compute Abstraction Layer) was not included on purpose."

A company spokesperson was keen to stress that the current version of Stream SDK 2.0 is a beta and doesn't fully reflect the final product. He went on to say that the CAL portion of the Stream SDK will be added back in, complete with GPU accelerated OpenCL support, "in the very near future," but stopped short of giving a more specific timeframe.

The spokesperson recommended that developers who wish to continue using CAL in the meantime should use ATI Stream SDK v1.4, which includes support for Brook+ but not OpenCL.

Considering the focus AMD has placed on its support for OpenCL and other 'open standards', the company isn't yet delivering on all of its promises. Naturally, support is forthcoming as AMD has made clear and the OpenCL drum has been banging since before the Radeon HD 4890's launch, but it's not ready for prime time yet.

With that said, it's very early days for OpenCL at the moment and although Nvidia has made OpenCL drivers available to developers for all GPUs based on its CUDA architecture (G80 and beyond), they're still some time away from making it into the publicly released drivers.

Discuss in the forums.
Quote KoenVdd 10th August 2009, 16:05
Told ya. Though I do doubt that OpenCL will make things faster. If AMD doesn't manage to make its own efficient vendor specific API an open API won't help.
Quote HourBeforeDawn 10th August 2009, 16:11
now I thought OpenCL was more so for the physics sides of things at least thats what people always talked about, I guess I will have to read up more on OpenCL
Quote Tim S 10th August 2009, 16:19
Quote:
Originally Posted by HourBeforeDawn
now I thought OpenCL was more so for the physics sides of things at least thats what people always talked about, I guess I will have to read up more on OpenCL

It's for anything that isn't graphics processing and physics is one task that OpenCL can be used for. Havok is adopting OpenCL and I'm hoping Nvidia will do the same for PhysX. It makes sense for consumers and hopefully common sense prevails. (GPU accelerated) OpenCL can also be used to accelerate video transcoding, video decoding, encryption/decryption and anything that is heavily threaded - the benefit is that if the application is written correctly, it should spread tasks across all available processors that support OpenCL (i.e. CPU and GPU).

KoenVdd, developers have had good success speeding things up with Nvidia's C for CUDA API, so I've got high hopes for OpenCL.
Quote rembo666 10th August 2009, 16:57
What OpenCL really brings to the table is standardization. Once a developer can write the same code that can work on ANY GPU and/or CPU, that will open the doors to using the power of current GPUs EVERYWHERE. Lack of a standard API is what's really holding back the progress right now.
Quote technogiant 10th August 2009, 17:08
I don't know....I'm not an Nvidia fanboy and my current graphics cards is an Ati 3870x2....but Ati just don't seem to have the push that Nvidia does.....more games are optomised for Nvidia by way of their TWIMTBP....their drivers are often better optomized.....Ati had DX10.1 but never pushed home the advantage and now after apparently holding back on GPGPU awaiting a common API (or that was their excuse for doing nothing) while Nvidia pushed ahead with CUDA they now are not ready for it????


I'm kind of glad that someone has found a way to run SLI on the X38 chipset.
Quote KoenVdd 10th August 2009, 17:20
Quote:
Originally Posted by Tim S
*snip*
KoenVdd, developers have had good success speeding things up with Nvidia's C for CUDA API, so I've got high hopes for OpenCL.

I'm basing myself on that nVidia is almost twice as fast in f@h because they expose some high speed shared memory to be shared between threads which allows to skip so double computations, in CUDA, among other things. Now unless the f@h developers are lazy for not trying to make use of everything AMD offers or AMD has not yet chosen to expose all of their hardware features (dispite all my love for AMD, I do get the feeling they have a habbit of implementing features without making them usable), AMD is probably not going to magically catch up to nV with OpenCL. Don't expect f@h to bring out a OpenCL client anytime soon. But who knows, maybe OpenCL will make it easier to write efficient code. OpenCL will mainly make it possible to write vendor independant gpgpu, things may start going faster because optimizations can easily be shared and possibly more effort is going to put in gpgpu now you don't have to choose between vendor API's. But it won't expose any magic features that will make things faster for either party.
Quote Goty 10th August 2009, 19:53
Quote:
Originally Posted by technogiant
I don't know....I'm not an Nvidia fanboy and my current graphics cards is an Ati 3870x2....but Ati just don't seem to have the push that Nvidia does.....more games are optomised for Nvidia by way of their TWIMTBP....their drivers are often better optomized.....Ati had DX10.1 but never pushed home the advantage and now after apparently holding back on GPGPU awaiting a common API (or that was their excuse for doing nothing) while Nvidia pushed ahead with CUDA they now are not ready for it????

NVIDIA had more influence because they could toss money at developers/programmers. That's strictly what the TWIMTBP program is.
Quote Elton 10th August 2009, 21:00
Quote:
Originally Posted by Goty
Quote:
Originally Posted by technogiant
I don't know....I'm not an Nvidia fanboy and my current graphics cards is an Ati 3870x2....but Ati just don't seem to have the push that Nvidia does.....more games are optomised for Nvidia by way of their TWIMTBP....their drivers are often better optomized.....Ati had DX10.1 but never pushed home the advantage and now after apparently holding back on GPGPU awaiting a common API (or that was their excuse for doing nothing) while Nvidia pushed ahead with CUDA they now are not ready for it????

NVIDIA had more influence because they could toss money at developers/programmers. That's strictly what the TWIMTBP program is.

Pretty much, DX10.1 was halted due to Nvidia's complaint to Microsoft that their cards couldn't support it.
Quote HourBeforeDawn 10th August 2009, 22:58
Quote:
Originally Posted by Elton
Pretty much, DX10.1 was halted due to Nvidia's complaint to Microsoft that their cards couldn't support it.

well more so to the gaming industry then to MS, thus the patch for Assassins Creed that downgraded it from 10.1 to 10 ~_~
Quote aussiebear 13th October 2009, 20:54
Beta 4 release just came out! It now supports Radeon HD 4xxx and 5xxx series GPUs!
=> http://developer.amd.com/GPU/ATISTREAMSDKBETAPROGRAM/Pages/default.aspx
Quote:
What’s New in v2.0-beta4

* First beta release of ATI Stream SDK with OpenCL™ GPU support.
* ATI Stream SDK v2.0 OpenCL™ is certified OpenCL™ 1.0 conformant by Khronos.
* Added Microsoft® Windows® 7 support.
* Added native Microsoft® Windows® 64-bit support.
* Float comparisons in kernels no longer produce a runtime error.
* Various other issues from previous v2.0 beta releases have been resolved.
Log in

You are not logged in, please login with your forum account below. If you don't already have an account please register to start contributing.