This shows you the differences between two versions of the page.
Both sides previous revision Previous revision Next revision | Previous revision | ||
bayes:cluster [2015/05/19 16:11] wphase |
bayes:cluster [2017/09/14 11:11] (current) wphase |
||
---|---|---|---|
Line 1: | Line 1: | ||
- | ====== Project BAYES - Resources ====== | + | ====== Computational resources ====== |
==== GPU nodes - Mésocentre de l'Université de Strasbourg ==== | ==== GPU nodes - Mésocentre de l'Université de Strasbourg ==== | ||
== Current configuration == | == Current configuration == | ||
- | * NVIDIA K80 (pri2015gpu queue): | + | * NVIDIA K80: |
- | * funded by the IDEX project BAYES | + | * funded by the IDEX project BAYES |
- | * 4 compute nodes with 4 K80/node - 32 GPUs | + | * 5 compute nodes with 4 K80/node - 40 GPUs |
- | * nodes hpc-n443-446 | + | * nodes hpc-n443-446, hpc-n517 |
- | * NVIDIA K40 (publicgpu or grantgpu queue): | + | * pri2015gpu partition |
+ | * NVIDIA K40: | ||
* funded by the IDEX project BAYES + NVIDIA donation | * funded by the IDEX project BAYES + NVIDIA donation | ||
* 2 nodes with 4 K40 GPU/node - 8 GPUs | * 2 nodes with 4 K40 GPU/node - 8 GPUs | ||
- | * nodes hpc-n223-224 | + | * nodes hpc-447-448 |
- | * NVIDIA K20 (publicgpu or grantgpu queue): | + | * pri2015gpu partition |
+ | * NVIDIA K20: | ||
* 2 nodes with 4 K20 GPU cards (2 from Z. Duputel INSU project) | * 2 nodes with 4 K20 GPU cards (2 from Z. Duputel INSU project) | ||
+ | * publicgpu or grantgpu queue | ||
* Other (public queue): | * Other (public queue): | ||
* M2050, M2070 and K20 GPUs. | * M2050, M2070 and K20 GPUs. | ||
- | |||
- | == Planned configuration (16K80 + 7K40 = 39 graphics processors) == | ||
- | * Must be compatible with existing NEC nodes of the HPC cluster | ||
- | * The GPUs must be able to host G and M matrices larger than 5GB. | ||
- | * 4 compute nodes | ||
- | * 4 GPU NVIDIA K80 per node (total: 16 K80 GPUs, equivalent to 32 graphics processors) | ||
- | * Each GPU will have its own dedicated PCI-E slot | ||
- | * 7 GPU cards to be installed on existing nodes | ||
- | * NVIDIA K40 | ||
- | * Each GPU card will have its own dedicated PCI-E slot | ||
- | * We also request 5 years of software and hardware support | ||
- | * Budget: 145 000 euros VAT included | ||
- | * September 2014: Currently working with Transtech, Nec and Dell for pre-pricings : | ||
- | * Nec/Transtech working on the same nodes as the ones installed | ||
- | * Dell working on a solution based on a C8000 chassis and C8220x nodes. If the C8000 chassis is full, could be the same density as the NEC nodes. Otherwise may be less interresting |