formation:astrosim2017gpu4dummies

Différences

Ci-dessous, les différences entre deux révisions de la page.

Lien vers cette vue comparative

Les deux révisions précédentesRévision précédente
Prochaine révision		Révision précédente
formation:astrosim2017gpu4dummies [2017/07/07 12:22] – equemeneformation:astrosim2017gpu4dummies [2017/07/10 16:52] (Version actuelle) – [NBody, a simplistic simulator] equemene
Ligne 392: Ligne 392:
   * ''Pi_FP32_MWC_xPU_OpenCL_1_1_1_1_01000000_Device0_InMetro_titan.npz''   * ''Pi_FP32_MWC_xPU_OpenCL_1_1_1_1_01000000_Device0_InMetro_titan.npz''
   * ''Pi_FP32_MWC_xPU_OpenCL_1_1_1_1_01000000_Device0_InMetro_titan''   * ''Pi_FP32_MWC_xPU_OpenCL_1_1_1_1_01000000_Device0_InMetro_titan''
 +
 +=== Exercice #7 : explore ''PiXPU.py'' with several simple configurations pour ''PR=1'' ===
 +
 +  * Without any parameters (the default ones) : 
 +    * what is the selected device ? How many itops (iterative operations per second) do you reach ?
 +  * With only the device parameter as ''-d 1'' to select ''#1'' for all the available devices :
 +    * What are the different ratios between the devices ? Which one is the most powerful ?
 +  * With the selector of device and increasing the number of iterations and the number of redo :
 +    * What arrive to itops values ? What is the typical variability on results ?
 +
 +<code>/scratch/$USER/PiXPU.py</code>
 +
 +<code>
 +/scratch/$USER/PiXPU.py -d 1
 +/scratch/$USER/PiXPU.py -d 2
 +/scratch/$USER/PiXPU.py -d 3
 +</code>
 +
 +<code>
 +/scratch/$USER/PiXPU.py -d 0 -i 100000000 -r 10
 +/scratch/$USER/PiXPU.py -d 1 -i 100000000 -r 10
 +/scratch/$USER/PiXPU.py -d 2 -i 100000000 -r 10
 +/scratch/$USER/PiXPU.py -d 3 -i 100000000 -r 10
 +</code>
 +
 +=== Exercice #8 : explore ''PiXPU.py'' by increasing the Parallel Rate ''PR'' ===
 +
 +  * With a PR from ''1'' to ''64'' set by ''-b'' and ''-e'', a the number of iterations of 1 billion, and 10 times and on default device
 +    * How decrease the elapsed time of 
 +  * With the selector of device and increasing the number of iterations and the number of redo :
 +    * What arrive to itops values ? What is the typical variability on results ?
 +
 +<code>./PiXPU.py -d 0 -b 1 -e 32 -i 1000000000 -r 10</code>
 +
 +In this case, we define a gnuplot config file as follow. Adapt to your files and configuration.
 +<code>
 +set xlabel 'Parallel Rate'
 +set ylabel 'Itops'
 +plot 'Pi_FP32_MWC_xPU_OpenCL_1_64_1_1_1000000000_Device0_InMetro_titan' using 1:9 title 'CPU with OpenCL'
 +</code>
 +
 +{{ :formation:pimc_1_64_cpu.png?600 |}}
 +
 +=== Exercice #9 : explore ''PiXPU.py'' with large PR on GPU (mostly power of 2) ===
 +
 +  * Explore with ''PR'' from ''2048'' to ''32768'' with a 128 step
 +  * For which ''PR'' the itops is the higher on you device ?
  
 To explore on this platform the GPU device (device #1) from 2048 to 32768 as parallel rates with a step of 128 and 1000000000 iterations: <code> To explore on this platform the GPU device (device #1) from 2048 to 32768 as parallel rates with a step of 128 and 1000000000 iterations: <code>
Ligne 401: Ligne 448:
   * ''Pi_FP32_MWC_xPU_OpenCL_2048_32768_1_1_1000000000_Device1_InMetro_titan''   * ''Pi_FP32_MWC_xPU_OpenCL_2048_32768_1_1_1000000000_Device1_InMetro_titan''
  
-In this case, you can define a gnuplot confi file+In this case, you can define a gnuplot config file
 <code> <code>
 set xlabel 'Parallel Rate' set xlabel 'Parallel Rate'
Ligne 410: Ligne 457:
 {{ :formation:pimc_2048_32768_gtx1080ti.png?600 |}} {{ :formation:pimc_2048_32768_gtx1080ti.png?600 |}}
  
 +=== Exercice #10 : explore ''PiXPU.py'' with around a large ''PR'' ===
 +
 +<code>./PiXPU.py -d 1 -b $((2048-8)) -e $((2048+8)) -i 10000000000 -r 10</code>
 +
 +  * ''Pi_FP32_MWC_xPU_OpenCL_2040_2056_1_1_10000000000_Device1_InMetro_titan''
 +  * ''Pi_FP32_MWC_xPU_OpenCL_2040_2056_1_1_10000000000_Device1_InMetro_titan.npz''
 +
 +In this case, you can define a gnuplot config file
 +<code>
 +set xlabel 'Parallel Rate'
 +set ylabel 'Itops'
 +plot 'Pi_FP32_MWC_xPU_OpenCL_2040_2056_1_1_10000000000_Device1_InMetro_titan' using 1:9 title 'GTX 1080 Ti'
 +</code>
 +
 +{{ :formation:pimc_2040_2056_gtx1080ti.png?600 |}}
 ==== NBody, a simplistic simulator ==== ==== NBody, a simplistic simulator ====
  
 +The ''NBody.py'' code is a implementation of N-Body kepkerian system on OpenCL devices. 
 +
 +It's available on:
 +  * on file: ''/scratch/AstroSim2017/NBody.py'' on workstations
 +  * on website: [[http://www.cbp.ens-lyon.fr/emmanuel.quemener/documents/Astrosim2017/NBody.py|NBody.py]]
 +
 +Launch the code with a ''N=2'' on ''1000'' iterations with a graphical output
 +<code>
 +python NBody.py -n 2 -g -i 1000 
 +</code>
 +
 +{{ :formation:nbody_n2_gpu.png?600 |}}
 +
 +
 +=== Exercice #10 : explore ''NBody.py'' with different devices ===
 +
 +=== Exercice #11 : explore ''NBody.py'' with steps and iterations ===
 +
 +=== Exercice #12 : explore ''NBody.py'' with Double Precision ===
  
 ===== Exploration with production codes ===== ===== Exploration with production codes =====
  
 ==== PKDGRAV3 ==== ==== PKDGRAV3 ====
 +
  
formation/astrosim2017gpu4dummies.1499430149.txt.gz · Dernière modification : (modification externe)
Rechercher
Translations
Navigation
Piste :
Piste :
toolbox
Centre Blaise Pascal, ENS de Lyon - 46, allée d'Italie - 69364 Lyon cedex 07 - France
Téléphone : +33 (0)4 26 23 38 01 - Email : cbp @ ens-lyon.fr
Sauf mention contraire, le contenu de ce wiki est placé sous les termes de la licence suivante : CC Attribution-Noncommercial-Share Alike 4.0 International
Modified Scanlines on DokuWiki under the hood