BREAKING
CoreWeave Benchmarks Vera Rubin NVL72
0
x
tokens per MW
0
Rubin GPUs per rack
0
Vera CPUs per rack
Live Silicon, Not Projections
CoreWeave test
live
●
Model: DeepSeek-R1
●
Measured on live silicon
●
Own cooling and networking
NVIDIA claim
vendor
●
Up to 10x vs GB200 NVL72
●
Benchmarked on Kimi-K2 Thinking
●
~1/10 cost per million tokens
How the Deployment Rolled Out
1
June 1 bring-up
↓
2
Dell XE9812 ships
↓
3
System validation
↓
4
Live benchmark
0
x
throughput per MW
0
production ramp year
Efficiency Now Governs AI Capacity
AI NEWS BLITZ
CoreWeave says it ran the first live benchmarks on NVIDIA's Vera Rubin platform.