Skip to content
KernelIndex
Search⌘K
Sign in

GEMM MXFP4 n2048 k2048

9 eligible runs
gemm

MXFP4 dense GEMM C = A @ B.T. Inputs A and B are bf16 values pre-snapped to the MXFP4 (E2M1 data + UE8M0 block scale, block size 32) representable grid, so MXFP4 kernels are numerically exact against the high-precision reference.

Source baseline · unbeaten
273.6µsmean
mm_fp4_mxfp4_cudnn_n2048_k2048FlashInfer-Bench baselines · Apache-2.0 · python

Reported evidence · last observed 2026-06-06. The source's designated baseline implementation. Reported by source; not independently reproduced.

No package · vendor the source →View source →Run detail →

Current records

Workloadm = 16 · bf16m = 128 · bf16m = 1024 · bf16

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload m = 128 · bf16 · CUDA 13.0 · Framework pytorch 2.11.0+cu130 · Protocol flashinfer-bench · mean · 3 results · last observed 2026-06-06Record history →
Estimated floor 1.11 µs · record 245.59× above itestimate, not evidence ›
DRAM 1.11 µs · compute 477 ns · bandwidth-bound on B200
every declared tensor crosses HBM exactly once (8,000 GB/s, B200 datasheet)
2·M·N·K with M=128, N=2048, K=2048 at the dense bf16 peak (2,250 TFLOP/s)
headroom-v1: a lower bound from declared tensors and datasheet peaks. A kernel can sit well above it for good reasons.
#
Implementation
Latency
vs #1
Trust
Observed
1
mm_fp4_mxfp4_cudnn_n2048_k2048baselineFlashInfer-Bench baselines
273.6µs
1.00×
Reported · Apache-2.0 · source
2026-06-06

Measured exactly what you asked. The source's designated baseline implementation. Reported by source; not independently reproduced.

source mirroredApache-2.0no install recipeView source →Run detail →
2
mm_fp4_mxfp4_flashinferbaselineFlashInfer-Bench baselines
283.8µs
1.04×
Reported · Apache-2.0 · source
2026-06-06

Measured exactly what you asked. The source's designated baseline implementation. Reported by source; not independently reproduced.

source mirroredApache-2.0no install recipeView source →Run detail →
3
mm_fp4_mxfp4_flashinferbaselineFlashInfer-Bench baselines
334.6µs
1.22×
Reported · Apache-2.0 · source
2026-06-06

Measured exactly what you asked. The source's designated baseline implementation. Reported by source; not independently reproduced.

source mirroredApache-2.0no install recipeView source →Run detail →

Scaling by m

100.0 µs200.0 µs300.0 µs161281024m →273.6 µs273.6 µs274.8 µsmm_fp4_mxfp4_cudnn_n2048_k2048280.5 µs283.8 µs283.6 µsmm_fp4_mxfp4_flashinfer
mm_fp4_mxfp4_cudnn_n2048_k2048mm_fp4_mxfp4_flashinferbest per workload · NVIDIA B200 · CUDA 13.0 · pytorch 2.11.0+cu130 · flashinfer-bench held constant

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
mm_fp4_mxfp4_cudnn_n2048_k2048FlashInfer-Bench baselines
python
273.6µs
1.00×
Reproduction-ready
Apache-2.0 · source
mm_fp4_mxfp4_flashinferFlashInfer-Bench baselines
python
280.5µs
1.03×
Reproduction-ready
Apache-2.0 · source

Semantics

Inputs and outputs
abf16 [m, k]
bbf16 [n, k]
cbf16 [m, n]
Axes and behavior
kconstant = 2048
mvariable
nconstant = 2048
determinismunspecified
constraintsNo mutation or aliasing
Identity
aliasgemm_mxfp4_n2048_k2048sha2567f82a4bf5df6…
Sources: FlashInfer-Bench (2026-06-06) · Apache-2.0last observed 2026-06-06How records are decidedJSON