9

is there any way I can test the CUDA samples and codes from a computer with no NVIDIA graphic card? I am using Windows and the latest version of CUDA.

lezebulon
  • 7,607
  • 11
  • 42
  • 73

1 Answers1

7

There are several possibilities:

  1. Use older version of CUDA, which has built-in emulator (2.3 has it for sure). Emulator is far from good, and you won't have features from latest CUDA releases.
  2. Use OpenCL, it can run on CPUs (though not with nVidia SDK, you will have to install either AMD or Intel OpenCL implementation (AMD works fine on Intel CPUs, btw)). In my experience, OpenCL is usually slightly slower than CUDA.
  3. There is windows branch of Ocelot emulator: http://code.google.com/p/gpuocelot/. I haven't tried it, though.

However, I would recommend buying some CUDA-capable card. 8xxx or 9xxx series is ok and really cheap. Emulation would allow you to get some basic skills of GPGPU programming, but is useless when you write some real-world application since it doesn't allow you to debug and tune performance.

aland
  • 4,829
  • 2
  • 24
  • 42
  • What are the downsides of using 2.3? I almost only need CUDA in order to do linear algebra operations. If I still can do it using the 2.3 version then I'll do this. – lezebulon Sep 06 '11 at 16:33
  • 1
    cuBLAS (BLAS library for CUDA) shipped with 2.3 toolkit implements very few functions (see [2.3 manual](http://developer.download.nvidia.com/compute/cuda/2_3/toolkit/docs/CUBLAS_Library_2.3.pdf) vs [4.0 manual](http://developer.download.nvidia.com/compute/DevZone/docs/html/CUDALibraries/doc/CUBLAS_Library.pdf)). Also, I haven't found any information whether [CULA](http://www.culatools.com/) runs on an emulator or not. Besides, 2.3 might support only old versions of gcc/icc, and has some bugs. However, all basic features of CUDA are present in 2.3. – aland Sep 06 '11 at 16:59
  • ok thanks! Also does it mean that I can compile & run the code made by version 2.3 on the 4.0 version without having to change anything? – lezebulon Sep 06 '11 at 17:21
  • 1
    Most likely yes. However, if your kernel is longer than ~10 lines of code, moving from emulator to real GPU will likely cause much more troubles, than moving from 2.3 to 4.0. – aland Sep 06 '11 at 17:47
  • Sorry I wasnt clear I was refering to cuBLAS only. For instance if I write a few functions relying on the older cuBLAS version, am I assured that it will run the same on 1) version 4.0 and 2) the real GPU? Thanks – lezebulon Sep 06 '11 at 17:54
  • _Almost surely_, but there are bugs found in cuBLAS occasionally, so no guarantees. – aland Sep 06 '11 at 18:31