H3-Image Day-0 Support: 4K in 0.87s with Vera Rubin
First Rubin data point: vLLM's day-0 MiniMax H3-Image support, 4K in 0.87s or 6,115 images/hour on four GPUs.
3 items · 2 sources · 2 days
Operational story trace
Follow in this browser to see new updates on your Live feed.
Latest change
CoreWeave reports a 2.7x speedup for Vera Rubin over GB200 on the MLPerf DeepSeek-V3 benchmark (via HPCwire), the first Rubin number in this thread on a standardized suite.
This is a run of separate vendor benchmark posts for NVIDIA's Vera Rubin, not one event. On Nov 12, vLLM published two day-0 serving results, for MiniMax H3-Image and DeepSeek-V4.1-Flash on a Vera Rubin NVL72 rack.
Arc
First Rubin data point: vLLM's day-0 MiniMax H3-Image support, 4K in 0.87s or 6,115 images/hour on four GPUs.
Source of the 58x claim: DeepSeek-V4.1-Flash on one NVL72 rack vs GB300 NVL72 at a P90 of 387 tokens/s per user.
First standardized-benchmark figure: CoreWeave's MLPerf DeepSeek-V3 run at 2.7x over GB200, reported by HPCwire.
First Rubin data point: vLLM's day-0 MiniMax H3-Image support, 4K in 0.87s or 6,115 images/hour on four GPUs.
Source of the 58x claim: DeepSeek-V4.1-Flash on one NVL72 rack vs GB300 NVL72 at a P90 of 387 tokens/s per user.
First standardized-benchmark figure: CoreWeave's MLPerf DeepSeek-V3 run at 2.7x over GB200, reported by HPCwire.
What to watch — open questions
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.