5

In the install documentation http://www.aleagpu.com/release/3_0_2/doc/installation.html it says to download the latest CUDA Drivers from https://developer.nvidia.com/cuda-downloads.

I have done so and am not able to get some of the samples to run.

The Latest CUDA Toolkit is v8.0, however when running some of the samples I get an error not being able to find curand64_75.dll looks like it is the v7.5 dll.

I have downloaded and installed just the v7.5 Toolkit, now I get a "exception [cuRAND Error] LAUNCH_FAILURE" error. I assume it has somthing to do with running the v7.5 Toolkit while using v8.0 drivers.

How do I get Alea GPU to reference curand64_80.dll?

Or do I have to uninstall all the V8.0 stuff and just have v7.5 running for now?

SiHa
  • 7,830
  • 13
  • 34
  • 43
Mike U
  • 2,901
  • 10
  • 32
  • 44

2 Answers2

4

AleaGPU is compatible with CUDA driver 8.0. There are some wrapping over CUDA libraries, such as cuRAND and cuBLAS. These libraries are not part of CUDA driver, they are part of CUDA toolkit. By default, AleaGPU search for these CUDA libraries in 7.5 version, but you can configure it: http://www.aleagpu.com/release/3_0_2/doc/faq.html#config_cuda_version .

Xiang Zhang
  • 2,831
  • 20
  • 40
1

AleaGPU is compatible with CUDA Toolkit 8.0. I have it running on my machines.

Have you tried running a simpler sample? http://www.aleagpu.com/release/3_0_2/doc/gallery.html

The reason I ask this is: http://www.aleagpu.com/release/3_0_2/doc/advanced_features_csharp.html

Alea GPU cannot expose the cuRAND device level interface because these routines have to be directly compiled into the kernel function.

Without seeing any code it's a bit of guess.

redb
  • 512
  • 8
  • 22
  • 1
    AleaGPU works in LLVM IR level, and the cuRAND device level interface is provided as C library, that's the reason why you cannot use cuRAND device level APIs. To use device level random number generator, we need implement the algorithm directly from C# or F#. – Xiang Zhang Dec 29 '16 at 08:25