Managed by the Infra team we have access to several machines. Their hardware is:
2 x Intel Xeon Gold 6242 CPU
512GB RAM
4 x 1.92TB (SATA SSD)
4 x Nvidia Quadro RTX6000 24GB
More details, also about how to access them is available at https://github.com/filecoin-project/lotus-dcops/issues/16.
If you want to use CUDA on those machines, you need to set (e.g. in your ~/.profile
):
export PATH="/usr/local/cuda/bin:${PATH}"
export LD_LIBRARY_PATH=/usr/local/cuda/lib64
We use the machines for different things, here's a list:
worker-gpu-3
: Volker is using it for development. WARNING: this machine seems to be broken: https://github.com/protocol/netops/issues/5worker-gpu-4
:worker-gpu-5
:worker-gpu-6
: This runs Artems buildbot (2023-07-25): Kind of unusable to to network connectivity issues: https://github.com/protocol/netops/issues/66#issuecomment-1496803933worker-gpu-7
: vmx using for development (2023-04-06)worker-gpu-8
: ‣ is using it for testing out new Nvidia drivers (currently broken, infra is on it)Additional machines were recently provided by our request.
1000GB RAM; AMD EPYC 7F32 8-Core; no GPU; 4 x 1.92TB (SATA SSD):
worker-cpu-2-1
(10.2.2.13)worker-cpu-2-2
(10.2.2.12)1000GB RAM; AMD EPYC 7402 24-Core; 1 x Nvidia Quadro RTX6000 24GB; 1 x 1.92TB (SATA SSD):