_cropped_story.jpg)
AI startup Graphcore has launched its 'Graphcloud' AI cloud services platform. Starting today, the firm will offer its IPU-POD systems that it launched last year in the form of cloud service. To do so, it will be leveraging cloud infrastructure provider, Cirrascale Cloud Services' technology.
IPU-PODs are clusters of Graphcore's IPU-M2000 units that are stackable, and each such M2000 unit further consists of four Colossus second-gen MK2 GC200 IPU processors. For those wondering, IPUs stand for Intelligence Processing Units and are Graphcore's in-house dedicated processors designed to handle AI and machine learning workloads.
For now, Graphcore is offering two of its IPU-PODs via Graphcloud:
-
IPU-POD16 - delivering 4 petaFLOPS of AI compute (16 GC200 IPUs in 4 IPU-M2000s),
-
IPU-POD64 - delivering 16 petaFLOPS of AI compute (64 GC200 IPUs in 16 IPU-M2000s).
A single Dell R6525 dedicated host server powered by AMD's EPYC 7532 will be handling an IPU-POD16 system. The POD16 will feature 14.4GB In-Processor Memory(on the IPU itself) and 512GB (8 x 64GB DIMMs) of Streaming Memory(surrounding the IPU chips). There's also 4TB of local NVMe storage provided.
The IPU-POD64 system is similarly built, just with four times of everything as on the POD-16. Inter-communication among the chips, etc., is handled by the IPU-Fabric interconnect architecture on both systems. They will also come pre-installed with Poplar.
Pricing of the two systems are as follows:
If you're interested and wish to have access to Graphcloud, you can request for it by visiting the following link on the official site.
