Report
Vera Rubin reportedly delivers roughly 3x GB300's GLM-5.2 throughput per megawatt
ArtificialAnlys reports roughly 2.8x GB300's throughput per megawatt on Kimi K3 at about 67 tokens per second per user.
TLDR
ArtificialAnlys says Vera Rubin serves roughly three times GB300's GLM-5.2 throughput per megawatt at about 165 tokens per second per user, and roughly 2.8 times GB300's Kimi K3 throughput per megawatt at about 67 tokens per second per user. It also reports the fastest GLM-5.2 and Kimi K3 results on its leaderboard, at 408 and 255 tokens per second per user, respectively.
Combined views
—
1 Source, first seen ago
