Rosetta: The Engine Behind Cray’s Slingshot Exascale-Era Interconnect

Re:Rosetta: The Engine Behind Cray’s Slingshot Exascale-Era ... by the_svd_doctor on Wednesday May 06, 2020 @12:48AM (#23087991)

For many problems in HPC the bottleneck is the network. The compute node are relatively classical commodity CPU(/GPU). But without a (good, custom) network, you can't achieve much.

[ Parent | on HN ]

Re:Rosetta: The Engine Behind Cray’s Slingshot Exascale-Era ... by gnufx on Wednesday May 06, 2020 @05:46AM (#23089303)

For example, some of us have speculated that there are periods when around half of the time on the UK Archer system (with the previous generation of Cray interconnect) might be spent in one MPI collective. It's a pity there's no measurement of such things.

[ Parent | on HN ]