ac4k
Competition author · sol-execbench/user/ac4k · license unknown
Records held
38as of 2026-10-01 03:09 UTC
Entries
429
Runs
429
Is this you? Claim this project ›
A claim grants attribution and metadata maintenance, never the right to edit evidence. Evidence claims are reviewed by a maintainer.
Current records
Operation · workload
History
Record
Implementation
Hardware
Since
Attention QKV projection with bias and reshapesuite of 16 cases
9.12µs
18.6% faster · was 11.2 µs
B200
2026-06-23
Adaptive instance normalization 2d backwardsuite of 16 cases
209.8µs
2.1% faster · was 214.3 µs
B200
2026-07-03
Attention output projection with residualsuite of 16 cases
34.6µs
4.9% faster · was 36.3 µs
B200
2026-06-26
Conv gated projection with causal convsuite of 16 cases
99.1µs
0.4% faster · was 99.5 µs
B200
2026-07-02
Embedding with initial LayerNorm backwardsuite of 16 cases
42.9µs
7.3% faster · was 46.3 µs
B200
2026-06-29
Fused RoPE with QK norm and KV cache updatesuite of 13 cases
9.38µs
6.8% faster · was 10.1 µs
B200
2026-06-26
Show all 38 records ›Showing all 38 records ⌄
Attention QKV projection with GQA reshapesuite of 16 cases
24.0µs
23.2% faster · was 31.3 µs
B200
2026-06-24
Grouped query attention with RoPE and QK normsuite of 16 cases
212.8µs
2.3% faster · was 217.8 µs
B200
2026-07-02
LM head projection with logit slicingsuite of 16 cases
375.6µs
1.4% faster · was 381.1 µs
B200
2026-06-19
Expert output weighted index add accumulationsuite of 16 cases
22.7µs
3.9% faster · was 23.7 µs
B200
2026-06-30
Attention QKV projection with bias and reshapesuite of 16 cases
9.07µs
12.8% faster · was 10.4 µs
B200
2026-07-12
Conv gated projection with causal convsuite of 16 cases
99.1µs
5.9% faster · was 105.3 µs
B200
2026-07-12
Encoder layer dual residual norm chainsuite of 16 cases
155.2µs
9.3% faster · was 171.1 µs
B200
2026-07-16
Vision cu seqlens variable length attentionsuite of 16 cases
69.6µs
2.0% faster · was 71.1 µs
B200
2026-07-02
Cross attention QKV projection with GQA repeat backwardsuite of 16 cases
149.5µs
1.6% faster · was 152.0 µs
B200
2026-07-02
Fused RoPE with QK norm and KV cache updatesuite of 13 cases
9.43µs
8.2% faster · was 10.3 µs
B200
2026-07-10
Attention softmax with softcapping and dropoutsuite of 16 cases
8.04µs
3.6% faster · was 8.34 µs
B200
2026-06-29
Expert output weighted index add accumulationsuite of 16 cases
22.6µs
3.8% faster · was 23.5 µs
B200
2026-07-10
MTP shifted embedding with dual RMS norm fusionsuite of 16 cases
111.8µs
5.3% faster · was 118.0 µs
B200
2026-06-30
Vision patch merger spatial shuffle MLPsuite of 15 cases
138.4µs
2.5% faster · was 141.9 µs
B200
2026-06-26
Measured entries
Implementation
Operation
Runtime
Best latency
Evidence
Availability
Show all 40 entries ›Showing all 40 entries ⌄
unknown
5.01µs
Reported
license unknown · no source
unknown
5.11µs
Reported
license unknown · no source
unknown
5.12µs
Reported
license unknown · no source
unknown
5.46µs
Reported
license unknown · no source
unknown
5.70µs
Reported
license unknown · no source
unknown
5.89µs
Reported
license unknown · no source
unknown
5.90µs
Reported
license unknown · no source
Showing 40 of 429; the full list is in the API dossier below.
Record activity
Nov
Dec
Jan
Feb
Mar
44Apr
26May
53Jun
58Jul
2Aug
Sep
Oct
Sources: NVIDIA SOL-ExecBench (2026-08-06)last observed 2026-08-06JSON