Skip to content
KernelIndex
Search⌘K
Sign in

GEMM fp4 e2m1 n2048 k2048

3 eligible runs
gemm

Unscaled narrow FP4 (E2M1) dense GEMM C = A @ B.T (N=2048, K=2048). Inputs are raw packed E2M1 values (int8, 2 fp4/byte) with NO block scales. This is the non-block-scaled narrow-precision GEMM path (a primitive / reference target, not a block-scaled NVFP4/MXFP4 kernel).

Source baseline · unbeaten
45.9µsmean
cublaslt_fp4_e2m1_scaled_mm_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 = 128 · int8m = 256 · int8m = 1024 · int8

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload m = 256 · int8 · CUDA 13.0 · Framework pytorch 2.11.0+cu130 · Protocol flashinfer-bench · mean · 1 results · last observed 2026-06-06Record history →
Estimated floor 295 ns · record 155.62× above itestimate, not evidence ›
DRAM 295 ns · bandwidth-bound on B200
every declared tensor crosses HBM exactly once (8,000 GB/s, B200 datasheet)
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
cublaslt_fp4_e2m1_scaled_mm_n2048_k2048baselineFlashInfer-Bench baselines
45.9µs
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 →

Scaling by m

50.0 µs100.0 µs1282561024m →49.5 µs45.9 µs102.7 µscublaslt_fp4_e2m1_scaled_mm_n2048_k2048
cublaslt_fp4_e2m1_scaled_mm_n2048_k2048best per workload · NVIDIA B200 · CUDA 13.0 · pytorch 2.11.0+cu130 · flashinfer-bench held constant

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
cublaslt_fp4_e2m1_scaled_mm_n2048_k2048FlashInfer-Bench baselines
python
45.9µs
1.00×
Reported
Apache-2.0 · source

Semantics

Inputs and outputs
a_fp4int8 [m, k_half]
b_fp4int8 [n, k_half]
cbf16 [m, n]
Axes and behavior
kconstant = 2048
mvariable
nconstant = 2048
k_halfconstant = 1024
determinismunspecified
constraintsNo mutation or aliasing
Identity
aliasgemm_fp4_e2m1_n2048_k2048sha2560127a9a81a87…
Sources: FlashInfer-Bench (2026-06-06) · Apache-2.0last observed 2026-06-06How records are decidedJSON