GEMM n2048 k8192
43 eligible runs
gemm
General matrix multiply (GEMM) C = A @ B.T. Captured from Llama-3.2-1B mlp.down_proj.
Source baseline · unbeaten
15.9µsmean
torch / matmulff95bcFlashInfer-Bench baselines · Apache-2.0 · python
Reported evidence · last observed 2026-04-20. The source's designated baseline implementation. Reported by source; not independently reproduced.
Current records
Workloadm = 2 · fp1643 cases
m
dtype
scrolls · 43 cases total
Not measured on H100 for this workload. Challenges →
Source-native comparison · GPU NVIDIA B200 · Workload m = 2 · fp16 · CUDA 13.0 · Framework pytorch 2.11.0+cu130 · Protocol flashinfer-bench · mean · 1 results · last observed 2026-04-20Record history →
Estimated floor 4.20 µs · record 3.79× above itestimate, not evidence ›
DRAM 4.20 µs · compute 30 ns · bandwidth-bound on B200
every declared tensor crosses HBM exactly once (8,000 GB/s, B200 datasheet)
2·M·N·K with M=2, N=2048, K=8192 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
115.9µs1.00×Reported · Apache-2.0 · source2026-04-20
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
Implementations
Implementation
Runtime
Best latency
Evidence
Availability
Semantics
Inputs and outputs
afp16 [m, k]
bfp16 [n, k]
cfp16 [m, n]
Axes and behavior
kconstant = 8192
mvariable
nconstant = 2048
determinismunspecified
constraintsNo mutation or aliasing
Identity
Sources: FlashInfer-Bench (2026-04-20) · Apache-2.0last observed 2026-04-20How records are decidedJSON