Laguna S 2.1 NVFP4 Reaches Reported 145 Tok/s on 4x RTX 5090
Sign in✓ Signed in
Local tests have reportedly pushed Poolside's Laguna S 2.1 NVFP4 to 108 tok/s on one RTX PRO 6000 and 145 tok/s on 4x RTX 5090, putting a 117.6B-parameter model within reach of local hardware. The figures are promising but require methodological caution: the available comparison does not specify the prompt length, batch size, runtime configuration, or whether the measurements represent prefill, decode, or another workload.
Continue reading
The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in✓ Signed in — this article isn’t included in your current plan.
Unlocking the full article...