Inferact and Google reportedly make a preliminary TPUv7 Kimi K3 2.8T model submission
SemiAnalysis says Inferact led the submission end to end using the TorchTPU vLLM stack.
TLDR
SemiAnalysis says Inferact, the maker of vLLM, led the preliminary submission with Google using the TorchTPU vLLM stack. It claims the submission is already delivering strong performance per total cost of ownership against GB200 NVL72, B300 and MI355 in the ultra-high-interactivity regime, crediting a megakernel approach for external customers.
Combined views
—
5 Sources, first seen ago
