This is an old revision of the document!
Table of Contents
GPU at ÚFAL
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.
Servers with GPU units
GPU cluster gpu.q
at Malá Strana:
machine | GPU type | cc | GPUs | GPU RAM | Comment |
iridium | Quadro K2000 | cc3.0 | 1 | 2 GB | driver(iridium)=367.48 |
titan-gpu | GeForce GTX Titan Z | cc3.5 | 2 | 6 GB | driver(titan-gpu)=381.22 |
twister1; twister2; kronos | Tesla K40c | cc3.5 | 1 | 12 GB | driver(twister*)=367.48, driver(kronos)=384.81 |
dll1; dll2 | GeForce GTX 1080 | cc6.1 | 8 | 8 GB | driver(dll1)=375.66, driver(dll2)=387.26 |
titan | GeForce GTX 1080 | cc6.1 | 1 | 8 GB | driver(titan)=381.22 |
dll4; dll5 | GeForce GTX 1080 Ti | cc6.1 | 10 | 11 GB | driver(dll4)=375.66, driver(dll5)=384.69 |
dll3; dll6 | GeForce GTX 1080 Ti | cc6.1 | 9 | 11 GB | driver(dll3)=375.66, driver(dll6)=384.69 |
Desktop machines:
machine | GPU type | cc | GPUs | GPU RAM | Comment |
victoria; arc | GeForce GT 630 | cc3.0 | 1 | 2 GB | desktop machine |
athena | GeForce GTX 1080 | cc6.1 | 1 | 8 GB | Tom's desktop machine |
Not used at the moment: GeForce GTX 570 (from twister2)
All machines have CUDA8.0 and should support both Theano and TensorFlow.
Rules
- First, read Linux network and Grid.
- 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
(orqrsh
), never viassh
. You can ssh to any machine e.g. to runnvidia-smi
orhtop
, 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_vmem=12G
. - 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 you GPU memory requirements (e.g.gpu_ram=2G
). Thus e.g.qsub -q gpu.q -l gpu=1,gpu_cc_min3.5=1,gpu_ram=2G
- 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.qsub -q gpu.q -l gpu=4,gpu_cc_min3.5=1,gpu_ram=7G -pe smp 4
Warning: currently, this does not work, so you can omit the
-pe smp X
part. Milan Fučík is working on a fix. - 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
is necessary, otherwise the variable$CUDA_VISIBLE_DEVICES
will not be set correctly. E.g.qrsh -q gpu.q -l gpu=1,gpu_ram=2G -pty yes bash
In general: don't reserve a GPU (as described above) without actually using it for longer time. (E.g. try separating steps which need GPU and steps which do not and execute those separately on our GPU resp. CPU cluster.) Ondřej Bojar has a script /home/bojar/tools/servers/watch_gpus for watching reserved but unused GPU on most machines which will e-mail you, but don't rely on in only.
- Note that the dll machines have typically 10 cards, but “just” 250 GB RAM. So the expected (maximal)
mem_free
requirement for jobs with 1 GPU is 25GB. If your 1-GPU job takes e.g. 80 GB and you submit three such jobs on the same machine, you have effectively blocked the whole machine and seven GPUs remain unused.
How to use cluster
Set-up CUDA and CUDNN
You should add the following commands into your ~/.bashrc
CUDNN_version=6.0 CUDA_version=8.0 CUDA_DIR_OPT=/opt/cuda-$CUDA_version if [ -d "$CUDA_DIR_OPT" ] ; then CUDA_DIR=$CUDA_DIR_OPT export CUDA_HOME=$CUDA_DIR export THEANO_FLAGS="cuda.root=$CUDA_HOME,device=gpu,floatX=float32" export PATH=$PATH:$CUDA_DIR/bin export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CUDA_DIR/cudnn/$CUDNN_version/lib64:$CUDA_DIR/lib64 export CPATH=$CUDA_DIR/cudnn/$CUDNN_version/include:$CPATH fi
When not using Theano, just Tensorflow this can be simplified to export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/cuda-8.0/cudnn/6.0/lib64:/opt/cuda-8.0/lib64
. Note that on some machines (dll*, twister*), this is the current default even without setting LD_LIBRARY_PATH, but on other machines (kronos, titan, titan-gpu, iridium) you need to set LD_LIBRARY_PATH explicitly.
TensorFlow Environment
Majority people at UFAL use TensorFlow. To start using it you need to create python virtual environment (virtualenv or use Anaconda for it). Into the environment you must place TensorFlow. The TF is either in CPU or GPU version.
pip install tensorflow pip install tensorflow-gpu
You can use prepared environment by adding into your ~/.bashrc
export PATH=/a/merkur3/kocmanek/ANACONDA/bin:$PATH
And then you can activate your environment:
source activate tf1 source activate tf1cpu
This environment have TensorFlow 1.0 and all necessary requirements for NeuralMonkey.
Pytorch Environment
If you want to use pytorch, there is a ready-made environment in
/home/hajicj/anaconda3/envs/pytorch/bin
It does rely on the CUDA and CuDNN setup above.
Using cluster
As an alternative to qsub
, you can use /home/bojar/tools/shell/qsubmit
qsubmit --gpumem=2G --queue="gpu.q" WHAT_SHOULD_BE_RUN
It is recommended to use priority lower than the default -100 if you are not rushing for the results and don't need to leap over your colleagues jobs.
Basic commands
lspci # is any such hardware there? nvidia-smi # more details, incl. running processes on the GPU # nvidia-* are typically located in /usr/bin watch nvidia-smi # For monitoring GPU activity in a separate terminal (thanks to Jindrich Libovicky for this!) nvcc --version # this should tell CUDA version # nvcc is typically installed in /usr/local/cuda/bin/ theano-test # dela to vubec neco uzitecneho? :-) # theano-* are typically located in /usr/local/bin/ /usr/local/cuda/samples/1_Utilities/deviceQuery/deviceQuery # shows CUDA capability etc. ssh dll1; ~popel/bin/gpu_allocations # who occupies which card on a given machine
Select GPU device
The variable CUDA_VISIBLE_DEVICES constrains tensorflow and other toolkits to compute only on the selected GPUs. Do not set this variable yourself (unless debugging SGE), it is set for you automatically by SGE if you ask for some GPUs (see above).
To list available devices, use:
/opt/cuda/samples/1_Utilities/deviceQuery/deviceQuery | grep ^Device
Performance tests
* 980 vs 1080 vs Titan X (not the Titan Z we have)
In the following table is the experiment conducted by Tom Kocmi. You can replicate experiment: /a/merkur3/kocmanek/Projects/GPUBenchmark (you will need to prepare environment of TensorFlow11 or use my ANACONDA). The benchmark uses 2GB model of seq2seq machine translation in Neural Monkey (De > EN). If not specified, the benchmark had an access only to one GPU.
machine | Setup; CPU/GPU; Capability [cc] | Walltime | Note |
athena | GeForce GTX 1080; cc6.1 | 9:55:58 | Tom's desktop |
dll2 | GeForce GTX 1080; cc6.1 | 10:19:40 | |
titan | GeForce GTX 1080 Ti | 10:45:11 | (new result with correct CUDA version) |
dll1 | (2 GPU) GeForce GTX 1080; cc6.1 | 12:34:34 | Probably only one GPU was used |
dll2 | GeForce GTX 1080; cc6.1 | 13:01:05 | Only one GPU was used |
titan-gpu | (2 GPU) GeForce GTX Titan Z; cc3.5 | 16:05:24 | Probably only one GPU was used |
kronos-dev | Tesla K40c; cc3.5 | 22:41:01 | |
twister2 | Tesla K40c; cc3.5 | 22:43:10 | |
twister1 | Tesla K40c; cc3.5 | 24:19:45 | |
helena1 | 16x cores CPU | 46:33:14 | |
belzebub | 16x cores CPU | 52:36:56 | |
iridium | Quadro K2000; cc3.0 | 59:47:58 | |
helena7 | 8x cores CPU | 60:39:17 | |
arc | GeForce GT 630; cc3.0 | 103:42:30 | (approximated after 66 hours) |
lucifer4 | 8x cores CPU | 134:41:22 |
Second Benchmark
The previous benchmark only compares the speed of processing units within the GPUs and do not take into account the size of memory. Therefore I have conducted another benchmark, this time for each graphic card I have increased the batch size as much as possible so the model still could fit into the GPU (the previous benchmark model had batch size 20). This way the results should be more representative of the power for each GPU.
GPU; Cuda capability | GPU RAM | Walltime | Batch size | Machine |
Tesla K40c; cc3.5 | 12 GB | |||
GeForce GTX 1080 Ti; cc6.1 | 11 GB | 00:55:56 | 2300 | dll5 |
GeForce GTX 1080; cc6.1 | 8 GB | 01:10:57 | 1700 | dll1 |
GeForce GTX Titan Z; cc3.5 | 6 GB | 02:20:47 | 1100 | titan-gpu |
Quadro K2000; cc3.0 | 2 GB | 28:15:26 | 50 | iridium |
Links
GPU specs for those GPUs we have:
Individual acquisitions: NVIDIA Academic Hardware Grants
There is an easy way to get one high-end GPU: ask NVIDIA for an Academic Hardware Grant. All it takes is writing a short grant application (at most ~2 hrs of work from scratch; if you have a GAUK, ~15 minutes of copy-pasting). Due to the GPU housing issues (mainly rack space and cooling), Milan F. said we should request the Tesla-line cards (2017 check with Milan about this issue). If you want to have a look at an application, feel free to ask at hajicj@ufal.mff.cuni.cz :)
Take care, however, to coordinate the grant applications a little, so that not too many arrive from UFAL within a short time: these grants are explicitly not intended to build GPU clusters, they are “seeding” grants meant for researchers to try out GPUs (and fall in love with them, and buy a cluster later). If you are planning to submit the hardware grant, have submitted one, or have already been awarded one, please add yourself here.
Known NVIDIA Academic Hardware Grants:
- Ondřej Plátek - granted (2015)
- Jan Hajič jr. - granted (early 2016)