NVFP4 group GEMM
NVFP4 block-scaled group matrix multiply for B200: g independent GEMMs, each with its own (m, n, k) problem size, over nvfp4 e2m1 operands with fp8 scale factors per 16 elements of k and fp16 outputs. Per-group m and n are divisible by the kernel's mma tile, k by 256, and the batch dim l is always 1. Tensor shapes below are per group.
Reported evidence · last observed 2026-02-21. Reported by source; not independently reproduced. License unknown.
Current records
Not measured on H100 for this workload. Challenges →
1submission 505727gau.nernst13.1µs1.00×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
2submission 503798gau.nernst13.2µs1.01×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
3submission 506188nataliakokoromyti13.8µs1.05×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
4submission 506472guaguabear13.8µs1.06×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
5submission 506313guaguabear13.8µs1.06×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
6submission 506244guaguabear13.9µs1.06×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
7submission 506057guaguabear13.9µs1.06×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
8submission 506292Ouye Xie14.1µs1.08×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
9submission 506136Ouye Xie14.1µs1.08×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
10submission 504270Ouye Xie14.2µs1.09×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
11submission 498274Ouye Xie14.5µs1.11×Reported · license unknown · source2026-02-18stale
Measured exactly what you asked. Reported by source; not independently reproduced.
12submission 489410billcarson14.8µs1.13×Reported · license unknown · source2026-02-12stale
Measured exactly what you asked. Reported by source; not independently reproduced.
13submission 487724billcarson14.9µs1.14×Reported · license unknown · source2026-02-09stale
Measured exactly what you asked. Reported by source; not independently reproduced.
14submission 487719billcarson15.0µs1.14×Reported · license unknown · source2026-02-09stale
Measured exactly what you asked. Reported by source; not independently reproduced.
15submission 496386Ouye Xie15.1µs1.15×Reported · license unknown · source2026-02-18stale
Measured exactly what you asked. Reported by source; not independently reproduced.
16submission 503945crUst15.1µs1.15×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
17submission 501386gau.nernst15.1µs1.16×Reported · license unknown · source2026-02-19stale
Measured exactly what you asked. Reported by source; not independently reproduced.
18submission 502744crUst15.2µs1.16×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
19submission 487893Ouye Xie15.2µs1.16×Reported · license unknown · source2026-02-10stale
Measured exactly what you asked. Reported by source; not independently reproduced.
20submission 502637crUst15.2µs1.16×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
21submission 506484macto15.2µs1.16×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
22submission 502927guaguabear15.3µs1.17×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
23submission 503628jiab_8528115.4µs1.18×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
24submission 502394macto15.4µs1.18×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
25submission 496799macto15.4µs1.18×Reported · license unknown · source2026-02-18stale
Measured exactly what you asked. Reported by source; not independently reproduced.
26submission 491864jiab_8528115.5µs1.18×Reported · license unknown · source2026-02-15stale
Measured exactly what you asked. Reported by source; not independently reproduced.
27submission 494691macto15.5µs1.18×Reported · license unknown · source2026-02-17stale
Measured exactly what you asked. Reported by source; not independently reproduced.
28submission 501994guaguabear15.7µs1.20×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
29submission 493788macto15.7µs1.20×Reported · license unknown · source2026-02-17stale
Measured exactly what you asked. Reported by source; not independently reproduced.
30submission 490587jiab_8528115.8µs1.20×Reported · license unknown · source2026-02-14stale
Measured exactly what you asked. Reported by source; not independently reproduced.
31submission 493783macto15.8µs1.21×Reported · license unknown · no source2026-02-17stale
Measured exactly what you asked. Reported by source; not independently reproduced.
3215.9µs1.22×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
33submission 493047macto15.9µs1.22×Reported · license unknown · source2026-02-16stale
Measured exactly what you asked. Reported by source; not independently reproduced.
34submission 492388macto16.0µs1.22×Reported · license unknown · source2026-02-16stale
Measured exactly what you asked. Reported by source; not independently reproduced.
35submission 489191gau.nernst16.0µs1.22×Reported · license unknown · source2026-02-12stale
Measured exactly what you asked. Reported by source; not independently reproduced.
36submission 506083nataliakokoromyti16.0µs1.22×Reported · license unknown · source2026-02-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
37submission 490544jiab_8528116.1µs1.23×Reported · license unknown · source2026-02-14stale
Measured exactly what you asked. Reported by source; not independently reproduced.
3816.1µs1.23×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
39submission 490689macto16.1µs1.23×Reported · license unknown · source2026-02-14stale
Measured exactly what you asked. Reported by source; not independently reproduced.
4016.2µs1.24×Reported · license unknown · source2026-02-14stale
Measured exactly what you asked. Reported by source; not independently reproduced.
4116.4µs1.25×Reported · license unknown · source2026-02-14stale
Measured exactly what you asked. Reported by source; not independently reproduced.
42submission 505064nataliakokoromyti16.6µs1.27×Reported · license unknown · source2026-02-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
43submission 487955macto16.7µs1.27×Reported · license unknown · source2026-02-10stale
Measured exactly what you asked. Reported by source; not independently reproduced.
44submission 490628crUst16.8µs1.28×Reported · license unknown · source2026-02-14stale
Measured exactly what you asked. Reported by source; not independently reproduced.
45submission 487344macto16.9µs1.29×Reported · license unknown · source2026-02-09stale
Measured exactly what you asked. Reported by source; not independently reproduced.
46submission 487098jiab_8528117.1µs1.31×Reported · license unknown · source2026-02-08stale
Measured exactly what you asked. Reported by source; not independently reproduced.
47submission 482269Ouye Xie17.3µs1.32×Reported · license unknown · source2026-02-07stale
Measured exactly what you asked. Reported by source; not independently reproduced.
4817.6µs1.34×Reported · license unknown · source2026-02-13stale
Measured exactly what you asked. Reported by source; not independently reproduced.
4917.6µs1.35×Reported · license unknown · source2026-02-10stale
Measured exactly what you asked. Reported by source; not independently reproduced.
5017.7µs1.35×Reported · license unknown · source2026-02-09stale
Measured exactly what you asked. Reported by source; not independently reproduced.
51submission 498506Naturalseeker17.8µs1.36×Reported · license unknown · source2026-02-18stale
Measured exactly what you asked. Reported by source; not independently reproduced.
5217.8µs1.36×Reported · license unknown · source2026-02-09stale
Measured exactly what you asked. Reported by source; not independently reproduced.
53submission 498323Naturalseeker17.8µs1.36×Reported · license unknown · source2026-02-18stale
Measured exactly what you asked. Reported by source; not independently reproduced.
5417.9µs1.37×Reported · license unknown · source2026-02-09stale
Measured exactly what you asked. Reported by source; not independently reproduced.
5517.9µs1.37×Reported · license unknown · source2026-02-09stale
Measured exactly what you asked. Reported by source; not independently reproduced.
5617.9µs1.37×Reported · license unknown · source2026-02-08stale
Measured exactly what you asked. Reported by source; not independently reproduced.
57submission 495386Naturalseeker17.9µs1.37×Reported · license unknown · source2026-02-17stale
Measured exactly what you asked. Reported by source; not independently reproduced.
58submission 486998jiab_8528117.9µs1.37×Reported · license unknown · source2026-02-08stale
Measured exactly what you asked. Reported by source; not independently reproduced.
5918.1µs1.38×Reported · license unknown · source2026-02-08stale
Measured exactly what you asked. Reported by source; not independently reproduced.
6018.2µs1.39×Reported · license unknown · source2026-02-07stale
Measured exactly what you asked. Reported by source; not independently reproduced.
250 more rows in this comparison group. Open all in search →