NVIDIA GPU servers

Two BRCF research pods have NVIDIA GPU servers; however their use is restricted to the groups who own those pods. 

Servers

Hopefog pod

hfogcomp04.ccbb.utexas.edu compute server on the Hopefog pod (Ellington/Marcotte):

• Dell PowerEdge R750XA
• dual 24-core/48-thread CPUs (48 cores, 96 hyperthreads total)
• 512 GB system RAM
• 2 NVIDIA Ampere A100 GPUs w/80GB onboard RAM each

hfogcomp05.ccbb.utexas.edu

• GIGABYTE MC62-G40-00
• 32-core/64-thread AMD Ryzen CPU
• 512 GB RAM
• 4 NVIDIA RTX 6000 Ada GPUs, 48G RAM each

Wilke pod

wilkcomp03.ccbb.utexas.edu compute server on the Wilke pod:

• GIGABYTE MC62-G40-00 workstation
  
• AMD Ryzen 5975WX CPU (32 cores, 64 hyperthreads total)
• 512 GB system RAM
• 4 NVIDIA RTX 6000 Ada GPUs

GPU-enabled software

AlphaFold

The AlphaFold protein structure solving software is available on all AMD GPU servers. The /stor/scratch/AlphaFold directory has the large required database, under the data.3 sub-directory. There is also an AMD example script /stor/scratch/AlphaFold/alphafold_example_amd.sh and an alphafold_example_nvidia.sh script if the POD also has NVIDIA GPUs, (e.g. the Hopefog pod). Interestingly, our timing tests indicate that AlphaFold performance is quite similar on all the AMD and NVIDIA GPU servers.

TensorFlow and PyTorch

Two Python example scripts are located in /stor/scratch/GPU_info that can be used to ensure you have access to the server's GPUs from TensorFlow or PyTorch. Run them from the command line using time to compare the run times.

• Tensor Flow
  • time ( python3 /stor/scratch/GPU_info/tensorflow_example.py )
    • should take 30s or less with GPU, > 1 minute with CPUs only
    • this is a simple test, and on CPU-only servers multiple cores are used but only 1 GPU, one reason why the times are not more different
• PyTorch
  
  • time ( python3 /stor/scratch/GPU_info/pytorch_example.py )
    • takes ~30s or less to complete on wilkcomp03
    • takes ~1m to complete on hfogcomp04.

If GPUs are available and accessible, the output generated will indicate they are being used.

Note that our system-wide CUDA-enabled TensorFlow and PyTorch versions are only available in the default Python 3 command-line environment (e.g. python3 or python3.8 on the command line). They are not yet available in the global JupyterHub environment that uses the Python 3.9 kernel. If you need a different combination of Python and  TensorFlow/PyTorch versions, you'll need to construct an appropriate custom Conda environment (e.g. miniconda3 or anaconda).

GROMACS

An NVIDIA GPU-enabled version of the Molecular Dynamics (MD) GROMACS program is available on all NVIDIA GPU servers, and a CPU-only version is installed also.

The /stor/scratch/GROMACS directory has several useful resources:

• benchmarks/ - a set of MD benchmark files from https://www.mpinat.mpg.de/grubmueller/bench
• gromacs_nvidia_example.sh - a simple GROMACS example script taking advantage of the GPU, running the benchMEM.tpr benchmark by default.
• gromacs_cpu_example.sh - an GROMACS example script using the CPUs only.

Resources

CUDA

Both hfogcomp04 and wilkcomp03 have both CUDA 11.8 and CUDA 12.x installed, under version-specific subdirectories of /usr/local. 

To ensure CUDA 11 is made active:

export CUDA_HOME=/usr/local/cuda-11.8
export PATH=$CUDA_HOME/bin:$PATH

To ensure CUDA 12 is made active:

export CUDA_HOME=/usr/local/cuda-12
export PATH=$CUDA_HOME/bin:$PATH

Neither version is specified by default, and some (but not all) programs rely on these environment variables. So you should activate one or the other before running software that uses GPUs.

After setting these environment variables, type nvcc --version to ensure you have access to the desired version.

CUDA drivers are installed under /usr/lib/x86_64-linux-gnu/. To see what version is currently installed:
ls /usr/lib/x86_64-linux-gnu/libnvidia-gl*. See https://saturncloud.io/blog/where-did-cuda-get-installed-in-my-computer/.

Command-line diagnostics

Use nvidia-smi to verify access to the server's GPUs and to monitor GPU usage.

Sharing resources

Since there's no batch system on BRCF POD compute servers, it is important for users to monitor their resource usage and that of other users in order to share resources appropriately.

• Use top to monitor running tasks (or top -i to exclude idle processes)
  • commands while top is running include:
  • M - sort task list by memory usage
  • P - sort task list by processor usage
  • N - sort task list by process ID (PID)
  • T - sort task list by run time
  • 1 - show usage of each individual hyperthread
    • they're called "CPUs" but are really hyperthreads
    • this list can be long; non-interactive mpstat may be preferred
• Use mpstat to monitor overall CPU usage
  • mpstat -P ALL to see usage for all hyperthreads
  • mpstat -P 0 to see specific hyperthread usage
• Use free -g to monitor overall RAM memory and swap space usage (in GB)
• Use nvidia-smi to  monitorGPU usage