multi

8×H100 SXM (NVSwitch, NVLink4 · ~900 GB/s/GPU)● graded on busbw (NVLink bus-bandwidth efficiency)

The multi-GPU sibling of hard. A coding agent takes a PyTorch + NCCL reference for a distributed op and rewrites it as a fast, fine-grained NVLink kernel (CUDA / Triton / NVSHMEM / CUDA symmetric memory / ParallelKittens) that beats the NCCL baseline. The graded quantity is busbw — achieved NVLink bus bandwidth ÷ NVLink peak, never TFLOPS — so every problem is deliberately communication-dominated. Six-problem deck, all busbw-graded, each forbidding its bare collective. No runs yet.

showing 6 of 6 rows
filesconversation
AllReduce + Residual-no runno runclean--
referencesolution
trace0/0
awaiting first run
ReduceScatter + RMSNorm-no runno runclean--
referencesolution
trace0/0
awaiting first run
AllGather + fp8 Dequant-no runno runclean--
referencesolution
trace0/0
awaiting first run
MoE All-to-All-no runno runclean--
referencesolution
trace0/0
awaiting first run
Ulysses All-to-All-no runno runclean--
referencesolution
trace0/0
awaiting first run
fp8 ReduceScatter Grad-no runno runclean--
referencesolution
trace0/0
awaiting first run

Methodology and the full problem deck live in the spec. Results will populate here as sweeps complete on the 8×H100 node.