Skip to content
KernelIndex
Search⌘K

gpt-5-2025-08-07

Competition author · flashinfer-ai/flashinfer-trace · Apache-2.0
Records held
72as of 2026-10-01 23:30 UTC
Entries
47
Runs
1,183
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
RMSNorm h7168bf16 · [7168] · batch 32
8.05µs
1.9% faster · was 8.21 µs
B200
2025-10-16
RMSNorm h512bf16 · [512] · batch 18
6.19µs
0.2% faster · was 6.20 µs
B200
2025-10-16
RMSNorm h512bf16 · [512] · batch 14521
14.5µs
0.4% faster · was 14.5 µs
B200
2025-10-16
RMSNorm h128bf16 · [128] · batch 316
6.18µs
0.7% faster · was 6.22 µs
B200
2025-10-16
RMSNorm h128bf16 · [128] · batch 256
6.17µs
0.1% faster · was 6.18 µs
B200
2025-10-16
RMSNorm h128bf16 · [128] · batch 24
6.17µs
0.4% faster · was 6.20 µs
B200
2025-10-16
RMSNorm h128bf16 · [128] · batch 32
6.18µs
0.3% faster · was 6.20 µs
B200
2025-10-16
RMSNorm h128bf16 · [128] · batch 4
6.17µs
0.6% faster · was 6.21 µs
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 14
4.58ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 14107
40.3ms
B200
2025-10-16
Show all 40 records ›
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 11948
31.2ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 62
7.24ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 59
7.27ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 58
10.7ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 57
9.72ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 56
10.4ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 55
9.50ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 54
8.80ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 53
9.29ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 52
6.40ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 7
3.09ms
72.1% faster · was 11.1 ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 1
1.51ms
85.8% faster · was 10.6 ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 32
8.27ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 80
12.0ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 901
16.2ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 16
4.99ms
B200
2025-10-16
MoE FP8 block scale ds routing topk8 ng8 kg4 e32 h7168 i2048bf16/fp32/fp8_e4m3 · [256] · seq len 15
2.36ms
B200
2025-10-16
100.7µs
42.4% faster · was 174.8 µs
B200
2025-10-21
90.7µs
47.8% faster · was 173.8 µs
B200
2025-10-21
GQA paged decode h32 kv8 d128 ps1bf16 · [1, 32, 128] · num pages 18 · num kv indices 2
8.19µs
20.7% faster · was 10.3 µs
B200
2025-10-16
GQA paged decode h32 kv4 d128 ps1bf16 · [16, 32, 128] · num pages 5422 · num kv indices 5372
101.9µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [16, 32, 128] · num pages 29084 · num kv indices 19815
216.6µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [16, 32, 128] · num pages 6510 · num kv indices 6460
109.7µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [16, 32, 128] · num pages 30172 · num kv indices 20903
221.3µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [16, 32, 128] · num pages 7598 · num kv indices 7548
117.3µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [16, 32, 128] · num pages 31260 · num kv indices 21991
227.4µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [16, 32, 128] · num pages 8686 · num kv indices 8636
122.7µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [16, 32, 128] · num pages 32348 · num kv indices 23079
230.1µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [64, 32, 128] · num pages 55007 · num kv indices 54991
329.9µs
B200
2026-03-28
GQA paged decode h32 kv4 d128 ps1bf16 · [64, 32, 128] · num pages 57183 · num kv indices 57167
336.6µs
B200
2026-03-28

32 more records · Open in the records ledger →

Measured entries

Implementation
Operation
Runtime
Best latency
Evidence
Availability
gpt-5 / cuda1ad912gpt-5-2025-08-07
cuda
6.17µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / tritonbfd137gpt-5-2025-08-07
triton
6.17µs
Reported
Apache-2.0 · source
gpt-5 / triton13f897gpt-5-2025-08-07
triton
6.19µs
Reported
Apache-2.0 · source
gpt-5 / cudab963ecgpt-5-2025-08-07
cuda
6.27µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cuda8fff8agpt-5-2025-08-07
cuda
8.05µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cuda95c7fegpt-5-2025-08-07
cuda
8.19µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cudadccc70gpt-5-2025-08-07
cuda
9.35µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / triton21bcc2gpt-5-2025-08-07
triton
10.1µs
Reported
Apache-2.0 · source
gpt-5 / tritonb6738dgpt-5-2025-08-07
triton
10.5µs
Reported
Apache-2.0 · source
gpt-5 / cudaa83af5gpt-5-2025-08-07
cuda
11.5µs
Reproduction-ready
Apache-2.0 · source
Show all 40 entries ›
gpt-5 / triton159afdgpt-5-2025-08-07
triton
12.3µs
Reported
Apache-2.0 · source
gpt-5 / triton2f0daagpt-5-2025-08-07
triton
12.3µs
Reported
Apache-2.0 · source
gpt-5 / triton714ae0gpt-5-2025-08-07
triton
12.9µs
Reported
Apache-2.0 · source
gpt-5 / triton679e13gpt-5-2025-08-07
triton
13.3µs
Reported
Apache-2.0 · source
gpt-5 / cuda5fa8b5gpt-5-2025-08-07
cuda
13.5µs
Reproduction-ready
Apache-2.0 · source
triton
15.8µs
Reported
Apache-2.0 · source
triton
20.4µs
Reported
Apache-2.0 · source
gpt-5 / triton0de5b5gpt-5-2025-08-07
triton
20.9µs
Reported
Apache-2.0 · source
gpt-5 / cuda00b2ddgpt-5-2025-08-07
cuda
30.7µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / triton14f133gpt-5-2025-08-07
triton
32.6µs
Reported
Apache-2.0 · source
gpt-5 / cuda727b5dgpt-5-2025-08-07
cuda
36.9µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cudafd76fegpt-5-2025-08-07
cuda
45.5µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cudac0b7b7gpt-5-2025-08-07
cuda
45.8µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cuda5eb89cgpt-5-2025-08-07
cuda
46.3µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cuda0b3c0agpt-5-2025-08-07
cuda
46.7µs
Reproduction-ready
Apache-2.0 · source
gpt-5_triton_f88811gpt-5-2025-08-07
triton
57.1µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cuda8ba217gpt-5-2025-08-07
cuda
67.8µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / cuda21ea96gpt-5-2025-08-07
cuda
68.7µs
Reported
Apache-2.0 · source
gpt-5 / tritoncb1275gpt-5-2025-08-07
triton
86.8µs
Reported
Apache-2.0 · source
gpt-5_triton_793693gpt-5-2025-08-07
triton
88.2µs
Reported
Apache-2.0 · source
gpt-5 / cuda99c7efgpt-5-2025-08-07
cuda
90.7µs
Reproduction-ready
Apache-2.0 · source
cuda
93.4µs
Reproduction-ready
Apache-2.0 · source
gpt-5 / tritona41cd4gpt-5-2025-08-07
triton
101.2µs
Reported
Apache-2.0 · source
gpt-5 / triton88089agpt-5-2025-08-07
triton
121.9µs
Reported
Apache-2.0 · source
triton
143.4µs
Reported
Apache-2.0 · source
gpt-5 / triton7308c5gpt-5-2025-08-07
triton
146.3µs
Reported
Apache-2.0 · source
triton
202.5µs
Reported
Apache-2.0 · source
gpt-5 / triton13eb4bgpt-5-2025-08-07
triton
209.0µs
Reported
Apache-2.0 · source
gpt-5 / triton998d17gpt-5-2025-08-07
triton
235.9µs
Reported
Apache-2.0 · source
gpt-5 / tritone289b9gpt-5-2025-08-07
triton
259.7µs
Reported
Apache-2.0 · source

Showing 40 of 47; the full list is in the API dossier below.

Record activity

Nov
Dec
Jan
Feb
54Mar
Apr
May
Jun
Jul
Aug
Sep
Oct
Sources: FlashInfer-Bench (2026-03-28) · Apache-2.0last observed 2026-03-28JSON