Hi Dion, It seems impossible to make 2 parallel/standalone Veros run

With <a class="issue-link js-issue-link" data-error-text="Failed to load title" data-i

List of environment variables related to bohrium: <div class="snippet-clipboard-co

RuntimeError: Two parallel Veros+Bohrium runs on two GPUs on the same node about veros HOT 4 CLOSED

team-ocean commented on May 31, 2024

RuntimeError: Two parallel Veros+Bohrium runs on two GPUs on the same node

from veros.

Comments (4)

omegahm commented on May 31, 2024 1

With bh107/bohrium#526 having been merged to master on the Bohrium project, you should be able to choose the different GPUs with BH_OPENCL_DEVICE_NUMBER=0 and BH_OPENCL_DEVICE_NUMBER=1.

from veros.

dionhaefner commented on May 31, 2024

What kind of environment variables did you use?

You need to tell Bohrium which GPU you want to use; that should be possible via BH_OPENCL_PLATFORM_NO (see docs). Trying to run both instances on the same GPU fails because the device runs out of memory.

from veros.

nutrik commented on May 31, 2024

List of environment variables related to bohrium:

export BH_STACK=opencl
export BH_OPENCL_DEVICE_TYPE=gpu
export BH_OPENCL_PLATFORM_NO=0

module load bohrium/06032018
module load veros/06032018

I tried numbers 0 and 1 for BH_OPENCL_PLATFORM_NO variable, but there is only 0 available.

terminate called after throwing an instance of 'std::runtime_error'
  what():  No such OpenCL platform. Tried to fetch #1 out of 0.

Output of clinfo:

Platform #0
  Name:                                  NVIDIA CUDA
  Version:                               OpenCL 1.2 CUDA 9.1.84

  Device #0
    Name:                                Tesla P100-PCIE-16GB
    Type:                                GPU
    Version:                             OpenCL 1.2 CUDA
    Global memory size:                  15 GB 920 MB 896 kB 
    Local memory size:                   48 kB 
    Max work group size:                 1024
    Max work item sizes:                 (1024, 1024, 64)

  Device #1
    Name:                                Tesla P100-PCIE-16GB
    Type:                                GPU
    Version:                             OpenCL 1.2 CUDA
    Global memory size:                  15 GB 920 MB 896 kB 
    Local memory size:                   48 kB 
    Max work group size:                 1024
    Max work item sizes:                 (1024, 1024, 64)

from veros.

dionhaefner commented on May 31, 2024

Interesting. Bohrium actually only offers an option to choose the OpenCL platform (such as CPU vs. GPU), not the device on that platform to be used for computations (is seems to choose the first it finds: https://github.com/bh107/bohrium/blob/9f022432ead25bfccb68a13d8fa032b638d7b6bf/ve/opencl/engine_opencl.cpp#L62). You should raise this issue with Bohrium. Up until then, I'm afraid it's not possible to run Veros on multiple devices on the same platform in parallel.

from veros.

Recommend Projects