Cisco and Pure shove mini AI in FlashStack converged systems
Entry-level AIRI equivalent
Pure Storage and Cisco have linked arms to build a converged FlashStack system for AI, a kind of AIRI microMINI but one that will run at half the speed.
FlashStack is a line of reference architecture converged systems using Pure Storage all-flash arrays, Cisco UCS servers and Nexus switches. It has typically competed with the NetApp-Cisco FlexPod.
AIRI is a Pure Storage and Nvidia system combining a FlashBlade unstructured data all-flash array (with 15 blades), two Arista 100Gbit switches, and 4 Nvidia DGX-1 GPU servers, each with 8 Tesla V100 GPUs. The thing puts out up to four petaFLOPS.
The later AIRI Mini system puts out 2 petaFLOPS, and has a smaller FlashBlade array (7 blades), two DGX-1s and two Cisco Nexus 100Gbit switches.
Now Cisco and Pure are developing the equivalent of a miniaturised AIRI Mini, but in the FlashStack line and using a Cisco server.
This FlashStack for AI validated design will feature:
- Cisco UCS C480ML M5 server (ML = Machine Learning)
- 4U x 2-socket
- 2 x Xeon SP (to 28 cores/CPU)
- Up to 3TB DRAM
- 8 x Nvidia SXM2 Tesla V100-32G Tensor Core GPUs with NVlink
- to 24 SAS/SATA drives (182TB)
- 6 x NVMe drives
- 4 x 100GbitE
- Cisco Nexus switches
It should be available later this year.
The C480ML M5 server is managed by Cisco's Intersight cloud-based management product. Pure Storage has developed APIs for its storage and now integrates with Intersight. This enables automated storage provisioning, and also management of Pure Storage functionality via the orchestration capability of Intersight.
All these shiny things will be presented at Cisco Live Barcelona, running from 28 January to 1 February. ®