Differences
This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision Next revision Both sides next revision | ||
gpu [2017/10/12 13:42] ufal [How to use cluster] |
gpu [2018/04/24 10:01] vodrazka [Set-up CUDA and CUDNN] |
||
---|---|---|---|
Line 5: | Line 5: | ||
===== Servers with GPU units ===== | ===== Servers with GPU units ===== | ||
GPU cluster '' | GPU cluster '' | ||
- | + | | machine | GPU type | GPU driver version | |
- | | machine | + | | dll1 | |
- | | iridium | + | | dll2 | GeForce GTX 1080 | |
- | | titan-gpu | + | | dll3 | GeForce GTX 1080 Ti | 375.66 | |
- | | twister1; twister2; kronos | + | | dll4 | |
- | | dll1; dll2 | GeForce GTX 1080 | | + | | dll5 | GeForce GTX 1080 Ti | |
- | | titan | + | | dll6 | GeForce GTX 1080 Ti | |
- | | dll3; dll4; dll5 | GeForce GTX 1080 Ti | | + | | gpu | GeForce GTX TITAN Z | 381.22 | 3.5 | 2 | 6 | 31 | |
- | | dll6 | GeForce GTX 1080 Ti | | + | | iridium | Quadro K2000 | 367.48 | 3.0 | |
+ | | kronos | ||
+ | | titan | GeForce GTX 1080 | | ||
+ | | twister1 | Tesla K40c | 367.48 | | ||
+ | | twister2 | Quadro P5000 | 367.48 | 6.1 | 1 | 17 | 47 | | ||
Desktop machines: | Desktop machines: | ||
Line 23: | Line 27: | ||
All machines have CUDA8.0 and should support both Theano and TensorFlow. | All machines have CUDA8.0 and should support both Theano and TensorFlow. | ||
- | === Disk space === | + | [[https://ufaladm2.ufal.hide.ms.mff.cuni.cz/munin/ufal.hide.ms.mff.cuni.cz/lrc-headnode.ufal.hide.ms.mff.cuni.cz/index.html# |
- | All the GPU machines are at Malá Strana (not at Troja), so you should not use ''/ | + | |
- | - '' | + | |
- | - ''/ | + | |
- | - ''/ | + | |
- | - ''/ | + | |
- | === Individual acquisitions: | ||
- | There is an easy way to get one high-end GPU: [[https:// | + | ===== Rules ===== |
- | + | * First, read [[internal:Linux network]] and [[:Grid]]. | |
- | Take care, however, to coordinate | + | * 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 '' |
- | + | * Always specify | |
- | Known NVIDIA Academic Hardware Grants: | + | * If you need more than one GPU card (on a single machine), always require as many CPU cores ('' |
- | + | * For interactive jobs, you can use '' | |
- | * Ondřej Plátek | + | * Note that the dll machines have typically 10 cards, but "just" |
- | * Jan Hajič jr. - granted (early 2016) | + | |
- | + | ||
- | + | ||
- | | + | |
===== How to use cluster ===== | ===== How to use cluster ===== | ||
- | |||
- | In this section will be explained how to use cluster properly. | ||
==== Set-up CUDA and CUDNN ==== | ==== Set-up CUDA and CUDNN ==== | ||
- | You can add following | + | You should |
CUDNN_version=6.0 | CUDNN_version=6.0 | ||
Line 63: | Line 55: | ||
export CPATH=$CUDA_DIR/ | export CPATH=$CUDA_DIR/ | ||
fi | fi | ||
+ | |||
+ | When not using Theano, just Tensorflow this can be simplified to '' | ||
+ | |||
+ | TensorFlow 1.5 precompiled binaries need CUDA 9.0, for this you need to | ||
+ | |||
+ | export LD_LIBRARY_PATH=/ | ||
+ | |||
+ | You also need to use '' | ||
+ | |||
+ | **Testing configuration (so far on twister1 & 2, titan and titan-gpu only)** | ||
+ | |||
+ | Multiple versions of '' | ||
+ | System default version for cuda library is configured in ''/ | ||
+ | |||
+ | / | ||
+ | / | ||
+ | |||
+ | Actual version used depends on the link in ''/ | ||
+ | |||
+ | ls -l /opt | ||
+ | ... | ||
+ | lrwxrwxrwx 1 root root 8 dub 9 12:30 cuda -> cuda-9.0 | ||
+ | ... | ||
+ | | ||
+ | This means that the system is using '' | ||
+ | |||
+ | If system default version does not work for you, you can set library path from your '' | ||
+ | |||
+ | / | ||
+ | | ||
+ | In order to link specific version of '' | ||
+ | |||
+ | / | ||
+ | |||
+ | Please check if required CUDNN_version is available for given CUDA_version. | ||
+ | |||
+ | |||
+ | |||
+ | |||
+ | |||
==== TensorFlow Environment ==== | ==== TensorFlow Environment ==== | ||
Line 92: | Line 124: | ||
==== Using cluster ==== | ==== Using cluster ==== | ||
- | Rule number one, always use the GPU queue (never log in machine by ssh). Always use qsub or qsubmit with proper arguments. | + | As an alternative to '' |
- | + | ||
- | For testing and using the cluster interactively | + | |
- | + | ||
- | qrsh -q gpu.q -l gpu=1, | + | |
- | + | ||
- | For running experiments you must use qsub command: | + | |
- | + | ||
- | qsub -q gpu.q -l gpu=1, | + | |
- | + | ||
- | Cleaner way to use cluster is with / | + | |
qsubmit --gpumem=2G --queue=" | qsubmit --gpumem=2G --queue=" | ||
| | ||
- | It is recommended to use priority -100 if you are not rushing for the results and don't need to leap over your colleagues jobs. | + | It is recommended to use priority |
==== Basic commands ==== | ==== Basic commands ==== | ||
Line 145: | Line 167: | ||
| titan | GeForce GTX 1080 Ti | | | titan | GeForce GTX 1080 Ti | | ||
| dll1 | (2 GPU) GeForce GTX 1080; cc6.1 | | | dll1 | (2 GPU) GeForce GTX 1080; cc6.1 | | ||
+ | | twister2 | ||
| dll2 | GeForce GTX 1080; cc6.1 | | | dll2 | GeForce GTX 1080; cc6.1 | | ||
| titan-gpu | | titan-gpu | ||
Line 162: | Line 185: | ||
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. | 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; Cuda capability |
- | | Tesla K40c; cc3.5 | | + | | GeForce GTX 1080 Ti; cc6.1 | 11 GB | 00:55:56 | 2300 | dll5 | |
- | | 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 1080; cc6.1 | 8 GB | 01:10:57 | 1700 | dll1 | | + | | Quadro P5000 |
| GeForce GTX Titan Z; cc3.5 | 6 GB | 02:20:47 | 1100 | titan-gpu | | | 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 | | + | | Quadro K2000; cc3.0 | 2 GB | 28:15:26 | 50 | iridium |
===== Links ===== | ===== Links ===== | ||
Line 176: | Line 199: | ||
GPU specs for those GPUs we have: | GPU specs for those GPUs we have: | ||
* [[http:// | * [[http:// | ||
+ | |||
+ | ==== Individual acquisitions: | ||
+ | |||
+ | There is an easy way to get one high-end GPU: [[https:// | ||
+ | |||
+ | 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 " | ||
+ | |||
+ | Known NVIDIA Academic Hardware Grants: | ||
+ | |||
+ | * Ondřej Plátek - granted (2015) | ||
+ | * Jan Hajič jr. - granted (early 2016) | ||
+ | |||
+ | |||
+ |