1

I recently found out that there is a way, using OpenJ9 with CUDA, to exploit GPU performance. But right now I am still facing problems. I do have the most recent Version of CUDA installed on my PC as well as the current version of OpenJ9. In addition I used the things written on this page to set up my environment:
https://www.ibm.com/support/knowledgecenter/SSYKE2_8.0.0/com.ibm.java.vm.80.doc/docs/gpu_enabling.html

Unfortunately I am unable to perform any processing on my GPU even though all PATH-vars are set correctly I think. Even though I switched on the logging process while running my programm the only feedback I received are these information:

[IBM GPU JIT]:  Dynamic linking error: Unable to locate NVML library nvml.dll
[IBM GPU JIT]:  Dynamic linking error: null pointer while loading shared library
[IBM GPU JIT]:  Device Number  0: name=GeForce GTX 1070, ComputeCapability=6.1

My program itself is quite simple, just using a Lambda in order to generate some workload to try out GPU-processing.

Maybe someone is/was facing a familiar problem and mastered it so he/she can provide some help. Thank you!

Abra
  • 19,142
  • 7
  • 29
  • 41
bensch
  • 11
  • 1

1 Answers1

0

I expect you just need to add the directory containing nvml.dll to your PATH.

The page you referenced is somewhat difficult to read, but it does mention the need to update PATH and, in particular, include the folder with nvml.dll.

You can find the NVML library in your NVIDIA drivers directory. The default location of this directory is C:\Program Files\NVIDIA Corporation\NVSMI.

If the NVIDIA software is installed in the default location on your system, the following should allow your test to run:

set PATH=%PATH%;C:\Program Files\NVIDIA Corporation\NVSMI

You can use the 'System Properties' control panel dialog to make that change permanent.

Keith W. Campbell
  • 151
  • 5