High resolution picture available HERE
On my system, I have now 3 different OpenCL implementations or platforms: NVIDIA’s OpenCL, Intel’s OpenCL and AMD’s OpenCL. And I must say that all OpenCL platforms work fine together. To confirm this last statement, I launched at the same time all GPU Caps Viewer OpenCL demos on different OpenCL platforms and different OpenCL compute devices (GPU or CPU).
I currently have an ASUS HD 6950 (primary card) and a Zotac GT 240 (secondary card) on my system. Thanks to the new feature of GPU Caps Viewer 1.10.0, you can select an OpenCL platform and for this platform, you can select which compute device must be used for OpenCL kernel execution:
Then I launched the Julia4D on the first GPU compute device of AMD platform, the mesh deformer on the first CPU compute device of AMD platform, the Post FX on the first CPU compute device of Intel platform and the 1M Particles demo on the first GPU compute device of NVIDIA platform:
Really cool, isn’t it?
OpenCL offers many interesting possibilities and I think I will spend some time shortly to play with this open computing language…
16 thoughts on “(Test) OpenCL Support Gets Better and Better”
One doubt you said u have opened all the openCL at the same time .. How did you run Intel CPU openCL and AMD CPU openCL at the same time ??
I hope nvidia releases openCL 1.1 WHQL drivers soon ..
Yes, I ran AMD and Intel OpenCL CPU at the same time. For one demo, I selected AMD OpenCL CPU, and for another demo I selected Intel OpenCL CPU…
Maybe it’s not clear on the screenshots, but all four OpenCL demos were running at the same time on different OpenCL devices / platforms.
Well i can understand Crossfire and SLI but how did you have two processor (intel and AMD ) in one system 🙁 … Well i think i understood wrongly or i have gone crazy 🙂
Okay I see your problem. AMD’s OpenCL CPU code path works for both AMD and Intel CPUs. It’s not limited to AMD CPUs only. Same thing for Intel’s OpenCL CPU code path.
@Arun, it is not AMD CPU OpenCL, it is AMD GPU OpenCL. Since ATi branding is ditched in favour of AMD, so newer old ATi gpu`s are actually AMD GPUs.
ATI Stream ? WTF ??? 😀
There’s no such thing as ATI Stream – it’s called AMD APP now.
Please ‘fix’ this in your next version of the GPU CAPS Viewer 🙂
@Ahmad: Arun takled about CPU OpenCL, because AMD provides both CPU and GPU OpenCL code paths.
@OpenXL: yep, AMD APP is the new name, but AMD did not change the name in the OpenCL driver yet.
@jeGx .. Any update on Intel OpenCL performance post …
Also, i wanted to know how is the performance of Intel OpenCL SDK on AMD processor and ATI stream performance in Intel processor .. is there any difference ?
@Arun: I only noticed that AMD CPU OpenCL is slightly faster than Intel CPU CL in Julia4D. In other CL demos, same performance (on a quad core X9650). And for a test on an AMD processor, I can’t do it: I don’t have an AMD CPU…
@Jegx Thanks for the info …
I will try it at home as i have AMD phenom Quad core processor and let you know if there is any difference .
@jegx .. I was not able to install Intel OpenCL SDK on my AMD machine … it is giving an error .. unsupported hardware 🙁 …Is this same with ATI Stream installation on Intel Hardware with out AMD GPU ?
@Arun – you weren’t able to install intel sdk since your CPU doesn’t have SSE4.1 and THAT is exactly what is required. APP SDK should install without problem on compatible CPU platform even without AMD GPU.
@Promilus Ok i understand now … But i don’t think AMD supports SSE 4.1 as i can see AMD supporting SSE4a..
Is there any AMD processor that can support SSE4.1 ?
I think Intel OpenCL SDK will not work in AMD machines 🙁 nice ..
Arun – no, current Intel OpenCL SDK won’t work on any AMD CPU while AMD SDK works with even P4 CPUs. Intel SDK works on E7xxx, E8xxx, Q8xxx, Q9xxx and all Core i3, i5, i7 CPUs. It might work with AMD Zambezi (bulldozer uarch) if it has SSE4.x support. SSE4A from Phenom (K10/STARS) is not compatible with intel extension thus doesn’t work with their SDK. I guess Intel SDK is optimized for Nehalem and Sandy Bridge architecture and might be faster than AMD SDK with same CPUs in some apps.
Why run CL on CPU if got CL GPU and it’s much faster, i was getting 10x the speed of it on 4670 GPU than just CPU. If using both than the sky is the limit, games will look like a dream in future if using CL on geometry but textures can’t be dynamically changed or be shaped, well maybe some day…
how are you able to have 3 CL devices on the AMD platform in this system? I mean, do you have Intel CPU, AMD GPU and NVIDIA GPU all in your AMD platform? How? In my case, where I have AMD CPU, AMD GPU and NVIDIA GPU, my AMD platform lists only the 2 AMD devices.
Comments are closed.