NVDA, CRWV,and IBM set up a cluster of 2,496 NVIDIA GB200 “Blackwell” GPUs—the largest ever submitted in the MLPerf Training benchmark . The cluster was 34× larger than the next-biggest cloud-provider submission
They trained Llama 3.1 (405 billion parameters)…..one of the biggest open-source language models, in just 27.3 minutes.
Their training times were 2× faster than other similar-scale GPU setups .
More efficient use of space: This massive cluster fit in 39 racks, compared to the ~156 racks needed for comparable GPUs, saving on space and energy .
NVDASPY top things to keep awareness of walking into Monday: - US - China negotiations to start at 11am UK time Monday - LA protests and Trump bringing in National Guards, not sure if it will affect the markets - Huang trying to capture EU markets, speaking in Germany - regional elections
thoughts on if we will enter the week green or red?
NVDA NVIDIA News: Next Place of Business After Middle East Is The EU; CEO Jensen Huang Set To Announce Germany’s Biggest “AI Factory” Next Week
Muhammad Zuhair NVIDIA CEO’s next trip is likely to be to the EU, where he is set to visit Germany first to meet with the local leaders and announce a large-scale AI project. Jensen will also meet the German Vice Chancellor Friedrich Merz in Berlin.
According to the report, the EU wants to catch up in the AI race with the rest of the world, but the major hurdle the region faces is that it has no option other than NVIDIA for its chip needs.
NVDA, 
CRWV,and 
IBM set up a cluster of 2,496 NVIDIA GB200 “Blackwell” GPUs—the largest ever submitted in the MLPerf Training benchmark . The cluster was 34× larger than the next-biggest cloud-provider submission 
