- Infos im HLRS Wiki sind nicht rechtsverbindlich und ohne Gewähr -
- Information contained in the HLRS Wiki is not legally binding and HLRS is not responsible for any damages that might result from its use -
NEC SX-ACE HW: Difference between revisions
From HLRS Platforms
Jump to navigationJump to search
No edit summary |
m (fixed broken Link (cmp. https://meta.wikimedia.org/wiki/Help:Advanced_editing#External_links)) |
||
| (2 intermediate revisions by one other user not shown) | |||
| Line 5: | Line 5: | ||
** 1 GHz frequency | ** 1 GHz frequency | ||
** 16 vectore pipes per core, each doing 2 add and 2 multiply per cycle for a peak performance of 64 GFlops per core | ** 16 vectore pipes per core, each doing 2 add and 2 multiply per cycle for a peak performance of 64 GFlops per core | ||
** 8 vector registers, 64 vector data registers | |||
** 1MB ADB (vector cache) per core, enhanced in size, speed and functionality from SX-9 | ** 1MB ADB (vector cache) per core, enhanced in size, speed and functionality from SX-9 | ||
** 64 GB of memory per node | ** 64 GB of memory per node | ||
| Line 12: | Line 13: | ||
** SUPER-UX Unix based OS | ** SUPER-UX Unix based OS | ||
** NQSII batch system | ** NQSII batch system | ||
** [http://de.nec.com/de_DE/en/documents/SX-ACE-brochure.pdf | ** [http://de.nec.com/de_DE/en/documents/SX-ACE-brochure.pdf NEC SX-ACE brochure] | ||
| Line 25: | Line 26: | ||
* Network | * Network | ||
** NEC proprietary highspeed network using IXS crossbar for message passing between nodes | ** NEC proprietary highspeed network using IXS crossbar for message passing between nodes | ||
** 4GB/s theoretical MPI bandwidth per node | ** 4GB/s theoretical MPI bandwidth per node (single lane configuration) | ||
** 40G Juniper QFabric network for 10G ethernet connection of filesystem using QOS to separate metadata and data traffic | ** 40G Juniper QFabric network for 10G ethernet connection of filesystem using QOS to separate metadata and data traffic | ||
Latest revision as of 17:37, 23 February 2015
Hardware
- 64 SX-ACE nodes with a single vector processor
- 4 cores for 256GFlops peak performance
- 1 GHz frequency
- 16 vectore pipes per core, each doing 2 add and 2 multiply per cycle for a peak performance of 64 GFlops per core
- 8 vector registers, 64 vector data registers
- 1MB ADB (vector cache) per core, enhanced in size, speed and functionality from SX-9
- 64 GB of memory per node
- 32 memory channels for 256GB/s bandwidth, usuable by one core or sharable
- around 260 Watt power consumption per node during Linpack
- nodes are connected through IXS crossbar for message passing with 4GB/s per node
- SUPER-UX Unix based OS
- NQSII batch system
- NEC SX-ACE brochure
- Frontend for Cross Compilation
- Dual socket NEC Express5800/R120e-1M with 2x Intel E5-2697 v2 @ 2.70GHz, running Redhat Linux
- Filesystem
- 250TB NEC ScaTeFS (scalable technology filesystem)
- 2 servers (shared object and metadata servers)
- 4 NEC iStorage M300 raid arrays with 3TB spindles
- Network
- NEC proprietary highspeed network using IXS crossbar for message passing between nodes
- 4GB/s theoretical MPI bandwidth per node (single lane configuration)
- 40G Juniper QFabric network for 10G ethernet connection of filesystem using QOS to separate metadata and data traffic