Graphcore, a $2.8B AI startup, looks like a failure! Their hardware and software simply don't measure up.
They are require twice as much 7nm silicon to achieve slower results with worse pricing, TCO, and software support.
"One of the negatives of MLPerf is that many firms will only submit results in the few models where they excel, and they will simply leave the rest empty."
Wish they said "we'll only publish your results if you submit for all categories".
Graphcore Looks Like A Complete Failure In Machine Learning Training Performance
This isn't directly mentioned in the article (and I don't think they're in MLPerf either), but do you have any analysis on SambaNova's hardware?
"One of the negatives of MLPerf is that many firms will only submit results in the few models where they excel, and they will simply leave the rest empty."
Wish they said "we'll only publish your results if you submit for all categories".
Has there been any comparison done with Brainchip's Akida Neural Processor (https://brainchipinc.com/products/)?
Clients and employees of SemiAnalysis may hold positions in companies referenced in this article. Isn't that obvious?