Reaction
Vera Rubin NVL72 claimed to deliver up to 7.3x GB300 NVL72's performance per dollar for agentic inference at 1 million context length
SemiAnalysis says AgentX captures multi-turn sessions, long contexts, high prefix reuse and bursts of short-lived sub-agents.
TLDR
SemiAnalysis claims Vera Rubin NVL72 delivers up to 7.3x the performance per dollar of GB300 NVL72 for agentic inference workloads at a 1 million context length. It calls AgentX an industry-standard inference benchmark, saying it captures long, multi-turn sessions, high prefix reuse and short-lived sub-agent bursts that create bursty KV-cache patterns.
Combined views
—
3 Sources, first seen ago
