[ Skip to the content ]

Institute of Formal and Applied Linguistics Wiki


[ Back to the navigation ]

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision
Previous revision
gpu [2019/10/18 09:57]
vodrazka [Servers with GPU units]
gpu [2022/09/20 09:39] (current)
rosa [GPU at ÚFAL] todo slurm
Line 3: Line 3:
 This page summarizes which UFAL servers have some GPU card, and suggests basic diagnostic commands, paths to installed tools, etc., simply everything necessary at the very beginning of using GPUs for experiments. This page summarizes which UFAL servers have some GPU card, and suggests basic diagnostic commands, paths to installed tools, etc., simply everything necessary at the very beginning of using GPUs for experiments.
  
 +**TODO: IN 2022 MOVING FROM SGE TO SLURM** (see [[slurm|slurm guidelines]]) -- **commands like ''qsub'' and ''qstat'' will stop working!**
 ===== Servers with GPU units ===== ===== Servers with GPU units =====
 GPU cluster ''gpu-ms.q'' at Malá Strana: GPU cluster ''gpu-ms.q'' at Malá Strana:
  
 | machine | GPU type | GPU driver version | [[https://en.wikipedia.org/wiki/CUDA#GPUs_supported|cc]] | GPU cnt | GPU RAM (GB) | machine RAM (GB)| | machine | GPU type | GPU driver version | [[https://en.wikipedia.org/wiki/CUDA#GPUs_supported|cc]] | GPU cnt | GPU RAM (GB) | machine RAM (GB)|
-dll3 |  GeForce GTX 1080 Ti |  396.24 |  6.|  10 |  11.|  248.+dll1 |  Quadro RTX 5000 |   455.23.05 |  7.|  8 |  16 |  366.2 | 
-| dll4 |  GeForce GTX 1080 Ti |  396.24 |  6.1 |  10 |  11.0 |  248.+| dll3 |  NVIDIA RTX A4000 |   510.73.08 |  8.6 |  10 |  16 |  248.
-| dll5 |  GeForce GTX 1080 Ti |  396.24 |  6.1 |  10 |  11.0 |  248.+| dll4 |  GeForce GTX 1080 Ti |   455.23.05 |  6.1 |  10 |  11 |  248.
-| dll6 |  GeForce GTX 1080 Ti  396.24 |  6.1 |  10 |  11.0 |  248.+| dll5 |  GeForce GTX 1080 Ti |   455.23.05 |  6.1 |  10 |  11 |  248.
-| dll7 |  GeForce RTX 2080 Ti |  418.39 |  7.5 |   |  11.|  122.0 | +| dll6 |  NVIDIA RTX A4000   510.73.08 |  8.6 |  |  16 |  248.
-kronos |  Tesla K40c  418.39 |  3.|  |  11.0 |  122.0 | +| dll7 |  NVIDIA RTX A4000 |   510.73.08 |  8.6 |  8 |  16 |  248.8 | 
-                                                                        +| dll8 |  Quadro RTX 5000 |   455.23.05 |  7.5 |  8 |  16 |  366.2 | 
 +| dll9 |  GeForce RTX 3090 |   455.23.05 |  8.6 |  4 |  25 |  183.0 | 
 +dll10 |  GeForce RTX 3090   455.23.05 |  8.|  |  25 |  183.0 |
  
 GPU cluster ''gpu-troja.q'' at Troja: GPU cluster ''gpu-troja.q'' at Troja:
  
 | machine | GPU type | GPU driver version | [[https://en.wikipedia.org/wiki/CUDA#GPUs_supported|cc]] | GPU cnt | GPU RAM (GB) | machine RAM (GB)| | machine | GPU type | GPU driver version | [[https://en.wikipedia.org/wiki/CUDA#GPUs_supported|cc]] | GPU cnt | GPU RAM (GB) | machine RAM (GB)|
-| tdll1 |  Quadro P5000 |  410.48 |  6.1 |  8 |  17 |  245 |                                                                            +| tdll1 |  Quadro P5000 |   455.23.05 |  6.1 |  8 |  16 |  245.0 
-| tdll2 |  Quadro P5000 |  410.48 |  6.1 |  8 |  17 |  245 |                                                                            +| tdll2 |  Quadro P5000 |   455.23.05 |  6.1 |  8 |  16 |  245.0 
-| tdll3 |  Quadro P5000 |  410.48 |  6.1 |  8 |  17 |  245 |                                                                            +| tdll3 |  Quadro P5000 |   455.23.05 |  6.1 |  8 |  16 |  245.0 
-| tdll4 |  Quadro P5000 |  410.48 |  6.1 |  8 |  17 |  245 |                                                                            +| tdll4 |  Quadro P5000 |   455.23.05 |  6.1 |  8 |  16 |  245.0 
-| tdll5 |  Quadro P5000 |  410.48 |  6.1 |  8 |  17 |  245 | +| tdll5 |  Quadro P5000 |   455.23.05 |  6.1 |  8 |  16 |  245.0 |
  
 Desktop machines: Desktop machines:
Line 38: Line 41:
   * All the rules from [[:Grid]] apply, even more strictly than for CPU because there are too many GPU users and not as many GPUs available. So as a reminder: always use GPUs via ''qsub'' (or ''qrsh''), never via ''ssh''. You can ssh to any machine e.g. to run ''nvidia-smi'' or ''htop'', but not to start computing on GPU. Don't forget to specify you RAM requirements with e.g. ''-l mem_free=8G,act_mem_free=8G,h_data=12G''.   * All the rules from [[:Grid]] apply, even more strictly than for CPU because there are too many GPU users and not as many GPUs available. So as a reminder: always use GPUs via ''qsub'' (or ''qrsh''), never via ''ssh''. You can ssh to any machine e.g. to run ''nvidia-smi'' or ''htop'', but not to start computing on GPU. Don't forget to specify you RAM requirements with e.g. ''-l mem_free=8G,act_mem_free=8G,h_data=12G''.
     * **Note that you need to use ''h_data'' instead of ''h_vmem'' for GPU jobs.** CUDA driver allocates a lot of "unused" virtual memory (tens of GB per card), which is counted in ''h_vmem'', but not in ''h_data''. All usual allocations (''malloc'', ''new'', Python allocations) seem to be included in ''h_data''.     * **Note that you need to use ''h_data'' instead of ''h_vmem'' for GPU jobs.** CUDA driver allocates a lot of "unused" virtual memory (tens of GB per card), which is counted in ''h_vmem'', but not in ''h_data''. All usual allocations (''malloc'', ''new'', Python allocations) seem to be included in ''h_data''.
-  * Always specify the number of GPU cards (e.g. ''gpu=1''), the minimal Cuda capability you need (e.g. ''gpu_cc_min3.5=1'') and your GPU memory requirements (e.g. ''gpu_ram=2G''). Thus e.g. <code>qsub -q 'gpu*' -l gpu=1,gpu_cc_min3.5=1,gpu_ram=2G</code> +  * Always specify the number of GPU cards (e.g. ''gpu=1'') and your GPU memory requirements (e.g. ''gpu_ram=2G''). Thus e.g. <code>qsub -q 'gpu*' -l gpu=1,gpu_ram=2G</code> 
-  * If you need more than one GPU card (on a single machine), always require as many CPU cores (''-pe smp X'') as many GPU cards you need. E.g. <code>qsub -q 'gpu*' -l gpu=4,gpu_cc_min3.5=1,gpu_ram=7G -pe smp 4</code>+  * If you need more than one GPU card (on a single machine), always require at least as many CPU cores (''-pe smp X'') as many GPU cards you need. E.g. <code>qsub -q 'gpu*' -l gpu=4,gpu_ram=7G -pe smp 4</code>
   * For interactive jobs, you can use ''qrsh'', but make sure to end your job as soon as you don't need the GPU (so don't use qrsh for long training). **Warning: ''-pty yes bash -l'' is necessary**, otherwise the variable ''$CUDA_VISIBLE_DEVICES'' will not be set correctly. E.g. <code>qrsh -q 'gpu*' -l gpu=1,gpu_ram=2G -pty yes bash -l</code>In general: don't reserve a GPU (as described above) without actually using it for longer time. (E.g. try separating steps which need GPU and steps which do not and execute those separately on our GPU resp. CPU cluster.) Ondřej Bojar has a script /home/bojar/tools/servers/watch_gpus for watching reserved but unused GPU on most machines which will e-mail you, but don't rely on it only.   * For interactive jobs, you can use ''qrsh'', but make sure to end your job as soon as you don't need the GPU (so don't use qrsh for long training). **Warning: ''-pty yes bash -l'' is necessary**, otherwise the variable ''$CUDA_VISIBLE_DEVICES'' will not be set correctly. E.g. <code>qrsh -q 'gpu*' -l gpu=1,gpu_ram=2G -pty yes bash -l</code>In general: don't reserve a GPU (as described above) without actually using it for longer time. (E.g. try separating steps which need GPU and steps which do not and execute those separately on our GPU resp. CPU cluster.) Ondřej Bojar has a script /home/bojar/tools/servers/watch_gpus for watching reserved but unused GPU on most machines which will e-mail you, but don't rely on it only.
   * Note that the dll machines have typically 10 cards, but "just" 250 GB RAM (DLL7 has only 128 GB). So the expected (maximal) ''mem_free'' requirement for jobs with 1 GPU is 25GB. If your 1-GPU job takes e.g. 80 GB and you submit three such jobs on the same machine, you have effectively blocked the whole machine and seven GPUs remain unused. If you really need to submit more high-memory jobs, send each on a different machine.   * Note that the dll machines have typically 10 cards, but "just" 250 GB RAM (DLL7 has only 128 GB). So the expected (maximal) ''mem_free'' requirement for jobs with 1 GPU is 25GB. If your 1-GPU job takes e.g. 80 GB and you submit three such jobs on the same machine, you have effectively blocked the whole machine and seven GPUs remain unused. If you really need to submit more high-memory jobs, send each on a different machine.
Line 53: Line 56:
 You need to set library path from your ''~/.bashrc'': You need to set library path from your ''~/.bashrc'':
  
-  CUDNN_version=7.0 +  CUDA_version=11.1 
-  CUDA_version=9.0+  CUDNN_version=8.0
   CUDA_DIR_OPT=/opt/cuda/$CUDA_version   CUDA_DIR_OPT=/opt/cuda/$CUDA_version
   if [ -d "$CUDA_DIR_OPT" ] ; then   if [ -d "$CUDA_DIR_OPT" ] ; then
Line 65: Line 68:
   fi   fi
  
-  * When not using Theano, just Tensorflow this can be simplified to ''export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/cuda/9.0/lib64:/opt/cuda/9.0/cudnn/7.0/lib64''.+  * When not using Theano, just Tensorflow this can be simplified to ''export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/cuda/10.1/lib64:/opt/cuda/11.1/cudnn/8.0/lib64''.
   * Note that the ''cudnn'' library is compiled for specific version of ''cuda''. If you need specific version of ''cudnn'', you can look in ''/opt/cuda/$CUDA_version/cudnn/'' whether it is available for a given ''$CUDA_version''.   * Note that the ''cudnn'' library is compiled for specific version of ''cuda''. If you need specific version of ''cudnn'', you can look in ''/opt/cuda/$CUDA_version/cudnn/'' whether it is available for a given ''$CUDA_version''.
  
Line 87: Line 90:
 This environment have TensorFlow 1.8.0 and all necessary requirements for NeuralMonkey. This environment have TensorFlow 1.8.0 and all necessary requirements for NeuralMonkey.
  
-==== Pytorch Environment ====+==== PyTorch Environment ====
  
-If you want to use pytorch, there is a ready-made environment in+Install PyTorch following the instructions on https://pytorch.org.
  
-  /home/hajicj/anaconda3/envs/pytorch/bin +At the time of writing, the recommended setup for CUDA 11.1 (supported by all GPU nodes) is:
-   +
-It does rely on the CUDA and CuDNN setup above.+
  
 +   pip3 install torch==1.10.0+cu111 torchvision==0.11.1+cu111 torchaudio==0.10.0+cu111 -f https://download.pytorch.org/whl/cu111/torch_stable.html
 ==== Using cluster ==== ==== Using cluster ====
  
Line 111: Line 113:
   watch nvidia-smi   watch nvidia-smi
     # For monitoring GPU activity in a separate terminal (thanks to Jindrich Libovicky for this!)     # For monitoring GPU activity in a separate terminal (thanks to Jindrich Libovicky for this!)
 +    # You can also use nvidia-smi -l TIME
   nvcc --version   nvcc --version
     # this should tell CUDA version     # this should tell CUDA version

[ Back to the navigation ] [ Back to the content ]