One of the facet bulletins made throughout TSMC’s Technology Symposium was that it already has clients readily available with product improvement progressing for its future 3nm course of node expertise. As we’ve reported on beforehand, TSMC is creating its 3nm for threat manufacturing subsequent 12 months, and excessive quantity manufacturing within the second half of 2022, so at the moment TSMC’s lead companions are already creating their future silicon on the preliminary variations of the 3nm PDKs.
One firm highlighted throughout TSMC’s shows was Graphcore. Graphcore is an AI silicon firm that makes the IPU, an ‘Intelligence Processing Unit’, to speed up ‘machine intelligence’. It lately introduced its second technology Colossus Mk2 IPU, constructed on TSMC’s N7 manufacturing course of, and that includes 59.2 billion transistors. The Mk2 has an efficient core depend of 1472 cores, that may run ~9000 threads for 250 Teraflops of FP16 AI coaching workloads. The firm places 4 of those chips collectively in a single 1U to allow 1 Petaflop, together with 450 GB of reminiscence and a customized low-latency cloth design between the IPUs.
A future technology of merchandise from Graphcore, in keeping with the TSMC presentation, is about to be developed with the TSMC 3nm course of in thoughts, skipping TSMC’s 5nm. No precise timescale was introduced, nor any indication of Graphcore’s technique. As we will see from the slide, the Colossus IPU line entails large high-transistor depend chips, utilizing the additional transistor finances afforded by the extra dense course of node.
We reached out to Graphcore for a press release, and acquired the next:
Nigel Toon, CEO & co-founder at Graphcore stated: “Graphcore was first to build a completely new kind of fully programmable processor, designed from the ground up for machine intelligence. Many of the innovative features of our IPU architecture and the high yields we see even at the cutting edge of the latest process node, are testament to the close technology partnership we enjoy with TSMC. With 59.4Bn transistors, and built using the latest TSMC 7nm technology, the MK2 IPU, which we announced in July, is the world’s most sophisticated processor. Each GC200 IPU has 1472 independent processor cores and an unprecedented 900MB of In-Processor memory delivering an 8x step up in real world performance vs. our MK1 products. We continue to work closely with TSMC as one of their technology innovation partners to explore the advantages of new process nodes and techniques, including N3, so we can continue to deliver more performance improvements to enable our customers to make new breakthroughs in AI.”
PCIe Accelerator with two IPUs
As it stands, Graphcore has a lot of merchandise constructed on its Mk1 and Mk2 IPUs, together with programs in partnership with Dell. Graphcore in Q1 2020 went by means of an prolonged Series D funding spherical earlier this 12 months, and has raised $450 million, valuating the corporate at $1.95 billion, with traders resembling BMW, Microsoft, the CEO of DeepMind, and a lot of VC corporations. According to TechCrunch, who reported this in February, the corporate nonetheless has $300m in money reserves. As the price to develop new silicon on the newest manufacturing node will increase, it will likely be attention-grabbing to see at what level Graphcore places an order in with TSMC’s 3nm, or if TSMC and Graphcore are working collectively to assist optimize the method for big scale chips and if TSMC will bear a few of that price.
