Graphcore, a start-up based in the UK, introduced its second-generation Intelligence Processing Unit (IPU) platform with greater processing power, more memory and built-in scalability for handling extremely large Machine Intelligence workloads.
The IPU-Machine M2000 is a plug-and-play Machine Intelligence compute blade that delivers one PetaFlop of Machine Intelligence compute and includes integrated networking technology. Each blade is powered by four of Graphcore’ new 7nm Colossus Mk2 GC200 IPU processors. Each chip contains more than 59.4 billion transistors on a single 823sqmm die. TSMC is Graphcore fab partner. Each IPU has 900MB of ultra-high-speed SRAM In-Processor Memory.
Multiple blades can work in unison in massive datacenter-scale systems of up to 64,000 IPUs – an “IPU-POD” that a maximum configuration would represent 16 ExaFlops of Machine Intelligence compute power.
Graphcore has developed its own low-latency IPU-Fabric technology to connects IPUs across the entire datacenter. A dedicated IPU-Gateway chip delivers 2.8 Tbps of bandwidth for each IPU-Machine M2000. The overall bandwidth grows to many Petabits/sec when multiple IPU-Machine M2000 systems are connected together.
https://www.graphcore.ai/posts/introducing-second-generation-ipu-systems-for-ai-at-scale
