Announcement
d-Matrix Corsair and Nvidia H100 are claimed to reach 845 tokens per second per user
d-Matrix says the pairing uses attention-FFN disaggregation for AI inference and claims roughly 50% lower latency.
TLDR
d-Matrix says pairing Corsair with Nvidia H100 uses attention-FFN disaggregation for AI inference. The company claims 845 tokens per second per user and roughly 50% lower latency.
Combined views
—
2 Sources, first seen ago
— likes— comments— saves— reposts
