// Blog
Notes from the datacenter.
Engineering write-ups on GPU infrastructure and applied AI.
Mar 12, 2026
Infrastructure
Scaling InfiniBand fabrics past 512 GPUs.
Non-blocking topologies, congestion control, and the tradeoffs we made when engineering our latest cluster.
Read →
Feb 27, 2026
AI Training
Notes on FP8 training with H200.
Numerical stability, loss scaling, and the throughput gains we measured across four representative workloads.
Read →
Feb 04, 2026
Inference
Achieving 38 ms p99 with Triton and TensorRT.
How we tuned our inference stack to serve 12M requests per day without breaking latency budgets.
Read →