ATI Stream SDK V2.0 | OpenCL Full CPU & GPU Support | Introduction

[64NOMIS]

The ATI Stream Team is proud to make available the fourth beta release of ATI Stream SDK v2.0 that provides the first complete OpenCL™ development platform. This release is certified fully compliant with OpenCL 1.0 by the Khronos Group and is supported on a wide range of AMD GPUs as well as any x86 multi-core CPU supporting SSE3. AMD offers the market both high-performance CPU and GPU technology, and as such we are delivering on this unique ability to provide an OpenCL platform that enables developers to create applications that run the way they were meant to be run, on all the available processors in the system! The beta is available for immediate download as part of our ATI Stream SDK beta program and we encourage you to take a look. For an introduction, see Simon Solotko on the impact of open, paralell computing.
Dealing With Reality | The Introduction | ATI Stream Technology and OpenCL | Part 1 by Simon Solotko

And for anyone who has writen in C or C++, I thought there was a gap between some of the deep dive technical material and the glossy material so I interviewed AMD's Ben Sander and we explain how OpenCL works and some of the basic development methodologies.
Dealing With Reality | The Introduction | ATI Stream Technology and OpenCL | Part 2 by Simon Solotko

Edit - Good discussion point that explains how some of this works:

After looking at the documentation it seems that I was correct. You shouldn't have to recompile for other architectures because OpenCL enumerates and utilizes resources (GPU/CPU/Cell/Etc) at runtime. A compiled OpenCL Kernel should be able to run on either company's hardware. There is a way to include what they call Native Kernels in your program. They are hardware specific - but they also aren't standard and being able to execute native kernels apparently isn't even a requirement in the standard.

Brace yourselves. The Kernel is not compiled until just prior to execution (hardware-specific native Kernels seem unlikely in practice since the whole idea is to be hardware independent but the HPC guys might do this) when it is directed to be built/compiled by the application. The compilation step is hardware dependent; if the application is executing on a platform with an x86 processor and an ATI GPU the Kernel will have either path available for compilation, potentially even determined at runtime! So if the GPU is busy, your applications could complile the Kernel for the CPU, execute it, and at a later time, compile it for the GPU, and execute it! You could hold an iner-application race, compiling separate instances for both, one for x86 and one for the GPU, and running them concurrently! I describe some of this in the interview: http://links.amd.com/openinterview

[netkas]

So, opencl support only for radeon gpus with compute shaders, its 4 series and higher.

[zalbard]

But something made in OpenCL should still work with Geforce, right?

[flippin_waffles]

You know, I just now realised the huge advantage AMD has here. When AMD says they are the only ones with both CPU and GPU, it's more than just marketing talk. Once DirectX11 and OpenCL release, AMD's platform releases with both, this should be the platform of choice for developers. Now that we have this announcement, all that is left is the Win7 launch I guess.

[Final8ty]

"ATI Radeon HD 5700 Series" = ati2mtag_Evergreen, PCI\VEN_1002&DEV_68B8
"ATI Radeon HD 5700 Series " = ati2mtag_Evergreen, PCI\VEN_1002&DEV_68B9
"ATI Radeon HD 5700 Series " = ati2mtag_Evergreen, PCI\VEN_1002&DEV_68BE
"ATI Radeon HD 5800 Series" = ati2mtag_Evergreen, PCI\VEN_1002&DEV_6898
"ATI Radeon HD 5800 Series " = ati2mtag_Evergreen, PCI\VEN_1002&DEV_6899
"ATI Radeon HD 5900 Series" = ati2mtag_Evergreen, PCI\VEN_1002&DEV_689C
"ATI Radeon HD 5900 Series " = ati2mtag_Evergreen, PCI\VEN_1002&DEV_689D

inf's

[NBF]

Its very nice to be able to test CPU and GPU performance and check which one runs better for the specific code.

[Manicdan]

"ATI Radeon HD 5900 Series" = ati2mtag_Evergreen, PCI\VEN_1002&DEV_689C
"ATI Radeon HD 5900 Series " = ati2mtag_Evergreen, PCI\VEN_1002&DEV_689D

is there a 5900 being planned?

[64NOMIS]

Here is the link for the Stream SDK V2.0 - lots of info and the download.

http://developer.amd.com/GPU/ATISTRE...s/default.aspx

[SKYMTL]

But something made in OpenCL should still work with Geforce, right?

Yes. That's the whole point of having an open platform.

[zalbard]

Yes. That's the whole point of having an open platform.

Just making sure that it is the case for this SDK in particular, you never know lol.

[Chumbucket843]

But something made in OpenCL should still work with Geforce, right?

you still have to port it. its not like you compile it twice and you are done. the question is "will it run well on both gpu's". opencl is a lot like cuda but a bit more abstract so it can be run on more architectures.

[Solus Corvus]

About time ATI. Now let's see some OpenCL benchmark programs made.

you still have to port it. its not like you compile it twice and you are done. the question is "will it run well on both gpu's". opencl is a lot like cuda but a bit more abstract so it can be run on more architectures.

Are you sure about this? If so it's the worst news I have heard in a while. I thought the whole point of OpenCL was that you would get a single binary compatible with multiple GPU architectures. If you have to port the code to each GPU and compile a new binary for each that would pretty much ruin it, imo. Now I'm going to have to go do some research to make sure that you are wrong, lol.

[netkas]

Developers still can make opencl code nvidia-only or ati-only, for example, in SnowLeopard, some of demos from Apple and Nvidia runs on Radeons, some doesnt, because of errors.

or can be blocked by vendor-id, just like in batman aa.

[Solus Corvus]

I don't doubt that it has that capability. Any robust language/API should let you make exclusive cases like that, if you choose to do so.

My problem would be if there isn't the ability for binary compatibility across multiple GPU architectures/vendors. If it HAS to be like: ATI exe, NV exe, larrabee exe, CPU exe, etc. Then half the point of OpenCL just failed before it even started.

[netkas]

if opencl kernel code exists in source form in app- then it should be almost compatible, but if it exists in binary form for some specific target device, then, you know the answer.

[Solus Corvus]

After looking at the documentation it seems that I was correct. You shouldn't have to recompile for other architectures because OpenCL enumerates and utilizes resources (GPU/CPU/Cell/Etc) at runtime. A compiled OpenCL Kernel should be able to run on either company's hardware. There is a way to include what they call Native Kernels in your program. They are hardware specific - but they also aren't standard and being able to execute native kernels apparently isn't even a requirement in the standard.

Some information:
http://www.khronos.org/registry/cl/s...ncl-1.0.43.pdf
http://arstechnica.com/apple/reviews...-x-10-6.ars/15

[Eson]

Sigh... I just don't seem to have any talent in parallel/stream programing. Ive looked at Cuda, CTM, OpenCL and C-based hardware description languages such as Verilog, Mitrion C and Impulse C. Many claim they use ANSI C. But i just don't have any clue in any of their 'easiest' examples.

Why aren't there never any 'Hello World' examples? Is it because i don't know C++ that i never understand anything?
I do know a lot of C and Java.

[Solus Corvus]

"Hello World" isn't very parallel.

[RaZz!]

is there a 5900 being planned?

i'm wondering as well, as there wasn't a x900 before with the 3000 and 4000 series, only a 4890

[Serra]

Sigh... I just don't seem to have any talent in parallel/stream programing. Ive looked at Cuda, CTM, OpenCL and C-based hardware description languages such as Verilog, Mitrion C and Impulse C. Many claim they use ANSI C. But i just don't have any clue in any of their 'easiest' examples.

Why aren't there never any 'Hello World' examples? Is it because i don't know C++ that i never understand anything?
I do know a lot of C and Java.

Well... here's a hello world program for CUDA. Only issue is it's got to be 100+ lines long... not exactly what I would call intuitive.

I can get around basically in C++, but I don't think I understand more than a few lines of that. Wish I did though, I always thought it would be fun to know how to use CUDA.

[Eson]

OK now im crying. Did they make a multithreaded 'Hello World' or is all that just to get a 'Hello World' printed out on the screen?

[Solus Corvus]

That's not too bad. Most of that's just setup, the actual kernel isn't very big.

[EniGmA1987]

"Hello World" isn't very parallel.

Ah but wouldnt it be fun to see just how fast you could process that command with 8 cores or more

[Pontos]

No HD3000 series support?

Why is that? they support Compute Shaders, just like the HD4000 series cards...

[Solus Corvus]

Also from the links provided so far: http://developer.amd.com/gpu/ATIStre...ialOpenCL.aspx