OpenCL SDK overview and hardware interoperability

0
votes

I am a little bit confused of the overall situation when it comes to OpenCL development so I'll just state my current understanding and questions as a list. Please correct me if I'm wrong.

  1. I know there are SDKs ("Platforms") by Intel, AMD (and I guess there is also OpenCL support in the Nvidia SDK?)

  2. Are there SDKs by other vendors?

  3. Will the SDK of one vendor support the devices of another? e.g. Nvidia devices with AMD sdk? I am able to run programs on my Intel CPU using AMD SDK.

  4. Is it the way to go to install multiple platforms at once and handle them at runtime?

Thank you!

3
openclgpugpgpu
Question 1 isn't a question. For questions 3 and 4, see stackoverflow.com/questions/14834403/… - Thomas

3 Answers

3
votes

1) Intel SDK offers OpenCL 1.2 support for CPU and GPU devices (for instance i7, HD4000, Xeon Phi, etc.), AMD SDK offers OpenCL 1.2 support for CPU and GPU devices (for instance Opteron, HD5xxx, HD6xxx, HD7xxx, etc.), NVIDIA SDK offers OpenCL 1.1 support only for GPUs (for instance GTX 2xx, GTX 4xx, GTX 5xx, Tesla, etc.).

2) Yes, from Apple, ARM and more.

3) Yes, thanks to OpenCL ICD, you can compile an executable on NVIDIA SDK and run on a AMD hardware. However you have to be careful with OpenCL versions, some problem can arise from compiling with an OpenCL 1.2 SDK and running on a 1.1. platform.

4) Yes, you can query OpenCL at runtime for the list of available platforms and use one or all of them. I uses everyday Intel OpenCL CPU device side by side with multiple AMD OpenCL GPU devices.

2
votes

This is a duplicate of How to use 2 OpenCL runtimes, which was answered. Short answer: The SDK just gets you the headers. The installed drivers provide the platforms and devices. You can use any that are installed.

0
votes

The Intel OpenCL SDK supports OpenCL 1.2 for the CPU but only OpenCL 1.1 for Intel HD. Quoting their webpage "OpenCL 1.2 on CPU mixed with OpenCL 1.1 on Intel® HD Graphics". However, the drivers for Intel HD don't seem to be very good. It crashes on most of code that runs on everything else.

The Nvidia SDK does not include the C++ bindings file cl.hpp. You can copy it form another SDK or get it from the Khronos Group OpenCL site.

I had problems running OpenCL on non AMD devices with the AMD SDK so I use the Intel SDK instead and it works on AMD, Intel, and Nvidia. However, the 1.1 C++ bindings and the 1.2 C++ bindings are not 100% compatible. If you want your code to work on both Nvidia and AMD GPUS only use OpenCL 1.1 and this define: #define CL_USE_DEPRECATED_OPENCL_1_1_APIS