AMD MXFP4 MoE
DeepSeek-R1 style MXFP4 fused MoE layer for MI355X: dynamic per-1x32 MXFP4 quantization of bf16 activations, fused gate+up a4w4 GEMM with SwiGLU, down-projection a4w4 GEMM, and weighted reduction over top-k routed plus always-selected shared experts. Expert dims are padded to 256-alignment; pre-shuffled CK-layout weight/scale copies and a config dict accompany the tensors listed here.
Reported evidence · last observed 2026-04-06. Reported by source; not independently reproduced. License unknown.
Current records
Not measured on H100, B200 for this workload. Challenges →
1submission 745202Danishlynx69.9µs1.00×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
2submission 714495Danishlynx70.5µs1.01×Reported · license unknown · source2026-04-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
3submission 749286Hamza70.6µs1.01×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
4submission 754731.jonnss76.7µs1.10×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
5submission 744553lgc033878.2µs1.12×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
6submission 743969lgc033881.6µs1.17×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
7submission 755012flower2123104.5µs1.49×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
8submission 754940guojun21104.6µs1.50×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
9submission 754739guojun21107.1µs1.53×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
10submission 754914bigmodel_wuzhigang107.3µs1.54×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
11submission 674077Maxwell Cipher107.3µs1.54×Reported · license unknown · source2026-03-30stale
Measured exactly what you asked. Reported by source; not independently reproduced.
12108.1µs1.55×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
13submission 754534shaw061434108.2µs1.55×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
14submission 750481shaw061434108.4µs1.55×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
15108.4µs1.55×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
16submission 754773flower2123108.7µs1.56×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
17submission 754347guojun21108.9µs1.56×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
18submission 754981NinoHeather109.3µs1.56×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
19109.5µs1.57×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
20submission 612426Ananda Sai A109.8µs1.57×Reported · license unknown · source2026-03-23stale
Measured exactly what you asked. Reported by source; not independently reproduced.
21submission 754170guojun21111.4µs1.59×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
22submission 730602lgc0338111.6µs1.60×Reported · license unknown · source2026-04-05stale
Measured exactly what you asked. Reported by source; not independently reproduced.
23submission 753670johnny.t.shi111.8µs1.60×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
24submission 715350shaw061434112.1µs1.60×Reported · license unknown · source2026-04-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
25submission 714877shaw061434113.2µs1.62×Reported · license unknown · source2026-04-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
26submission 676752johnny.t.shi113.2µs1.62×Reported · license unknown · source2026-03-31stale
Measured exactly what you asked. Reported by source; not independently reproduced.
27submission 625705johnny.t.shi113.5µs1.62×Reported · license unknown · source2026-03-24stale
Measured exactly what you asked. Reported by source; not independently reproduced.
28submission 728974lgc0338113.6µs1.62×Reported · license unknown · source2026-04-04stale
Measured exactly what you asked. Reported by source; not independently reproduced.
29submission 712180shaw061434113.7µs1.63×Reported · license unknown · source2026-04-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
30submission 624419johnny.t.shi113.8µs1.63×Reported · license unknown · source2026-03-24stale
Measured exactly what you asked. Reported by source; not independently reproduced.
31submission 543893John Hahn114.6µs1.64×Reported · license unknown · source2026-03-13stale
Measured exactly what you asked. Reported by source; not independently reproduced.
32submission 600310johnny.t.shi115.1µs1.65×Reported · license unknown · source2026-03-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
33submission 543522John Hahn115.9µs1.66×Reported · license unknown · source2026-03-13stale
Measured exactly what you asked. Reported by source; not independently reproduced.
34submission 673796Maxwell Cipher118.2µs1.69×Reported · license unknown · source2026-03-30stale
Measured exactly what you asked. Reported by source; not independently reproduced.
35submission 589916Danishlynx118.2µs1.69×Reported · license unknown · source2026-03-19stale
Measured exactly what you asked. Reported by source; not independently reproduced.
36submission 754241rosehulman.119.9µs1.72×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
37submission 753984guojun21120.3µs1.72×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
38submission 750364rosehulman.120.5µs1.72×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
39submission 670884Maxwell Cipher120.5µs1.72×Reported · license unknown · source2026-03-30stale
Measured exactly what you asked. Reported by source; not independently reproduced.
40submission 707949josusanmartin120.7µs1.73×Reported · license unknown · source2026-04-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
41submission 611955Ananda Sai A120.9µs1.73×Reported · license unknown · source2026-03-22stale
Measured exactly what you asked. Reported by source; not independently reproduced.
42submission 746413rosehulman.121.2µs1.73×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
43submission 611635Ananda Sai A121.3µs1.73×Reported · license unknown · source2026-03-22stale
Measured exactly what you asked. Reported by source; not independently reproduced.
44submission 754350NinoHeather121.5µs1.74×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
45submission 643890Maxwell Cipher121.6µs1.74×Reported · license unknown · source2026-03-27stale
Measured exactly what you asked. Reported by source; not independently reproduced.
46submission 643447Maxwell Cipher122.0µs1.74×Reported · license unknown · source2026-03-27stale
Measured exactly what you asked. Reported by source; not independently reproduced.
47submission 754624Aniket Sadashiva122.1µs1.75×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
48submission 732358guojun21122.3µs1.75×Reported · license unknown · source2026-04-05
Measured exactly what you asked. Reported by source; not independently reproduced.
49submission 611369Ananda Sai A122.8µs1.76×Reported · license unknown · source2026-03-22stale
Measured exactly what you asked. Reported by source; not independently reproduced.
50submission 682060Aniket Sadashiva122.9µs1.76×Reported · license unknown · source2026-03-31stale
Measured exactly what you asked. Reported by source; not independently reproduced.
51submission 754548champagnepapi123.0µs1.76×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
52submission 605993Ananda Sai A123.0µs1.76×Reported · license unknown · source2026-03-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
53submission 753928bigmodel_wuzhigang123.5µs1.77×Reported · license unknown · source2026-04-07
Measured exactly what you asked. Reported by source; not independently reproduced.
54submission 681568ooousay123.7µs1.77×Reported · license unknown · source2026-03-31stale
Measured exactly what you asked. Reported by source; not independently reproduced.
55submission 724956Barry_zhang123.8µs1.77×Reported · license unknown · source2026-04-04stale
Measured exactly what you asked. Reported by source; not independently reproduced.
56submission 602624Ananda Sai A124.0µs1.77×Reported · license unknown · source2026-03-21stale
Measured exactly what you asked. Reported by source; not independently reproduced.
57submission 745058rosehulman.124.5µs1.78×Reported · license unknown · source2026-04-06
Measured exactly what you asked. Reported by source; not independently reproduced.
58submission 586370Ananda Sai A124.7µs1.78×Reported · license unknown · source2026-03-18stale
Measured exactly what you asked. Reported by source; not independently reproduced.
59submission 598451ooousay124.8µs1.78×Reported · license unknown · source2026-03-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
60submission 582964Ananda Sai A124.8µs1.79×Reported · license unknown · source2026-03-18stale
Measured exactly what you asked. Reported by source; not independently reproduced.
540 more rows in this comparison group. Open all in search →