Google Cloud Posts 10 MLPerf v6.1 Training Results on Ironwood TPUs and GB300 GPUs
Start here
Google Cloud published 10 MLPerf Training v6.1 results covering its seventh-generation Ironwood TPUs and NVIDIA Grace Blackwell GPUs, including what it describes as the first MLPerf benchmark for agentic reinforcement learning. On Ironwood, 8,192 TPUs reached the pre-training target in 1.70 minutes, with 96% scaling efficiency at 4,096 chips. For the coding benchmark, 640 TPUs hit the target in 18.02 minutes, scoring 0.765 against a 0.69 goal. On A4X Max instances, 256 NVIDIA GB300 GPUs reached target quality in 31.96 minutes. The figures are Google Cloud's own submitted benchmark results; they show performance on these specific configurations rather than across all of its services.
Generated from reports · updated 24 min ago
Report timeline
Follow the reports to see the event from different sides.
- Hacker News · AIGoogle Cloud MLPerf v6.1 Results: Ironwood TPU and NVIDIA GB300 GPU Performance
Google Cloud published 10 MLPerf Training v6.1 results spanning seventh-generation Ironwood TPUs and NVIDIA Grace Blackwell GPUs, including the first MLPerf benchmark for agentic reinforcement learning. On Ironwood, 8,192 TPUs reached the pre-training target in 1.70 minutes with 96% scaling efficiency at 4,096 chips, while 640 TPUs hit the coding benchmark target in 18.02 minutes at 0.765 versus a 0.69 goal. On A4X Max instances, 256 NVIDIA GB300 GPUs reached target quality in 31.96 minutes.
Heat for this event
Not enough continuous observations yet to draw a trend.