AMD MLA decode
Multi-head latent attention decode for MI300: bf16 Q/K/V, non-paged pre-allocated latent KV cache, single decode step over a variable prefill length. Dimensions follow DeepSeek-R1/V3 (hidden 7168, 128 heads, kv-lora rank 512, rope dim 64).
Reported evidence · last observed 2025-06-03. Reported by source; not independently reproduced. License unknown.
Current records
Not measured on H100, B200 for this workload. Challenges →
Estimated floor 171.3 µs · record 10.41× above itestimate, not evidence ›
1submission 31179KernelBot user 71.78ms1.00×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
2submission 31138KernelBot user 4151.79ms1.00×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
3submission 31174KernelBot user 71.80ms1.01×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
4submission 31152KernelBot user 71.80ms1.01×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
5submission 31257KernelBot user 3761.82ms1.02×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
6submission 31256KernelBot user 3761.83ms1.03×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
7submission 31233KernelBot user 441.84ms1.03×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
8submission 30729KernelBot user 3761.84ms1.03×Reported · license unknown · source2025-06-02stale
Measured exactly what you asked. Reported by source; not independently reproduced.
9submission 30508KernelBot user 3761.87ms1.05×Reported · license unknown · source2025-06-02stale
Measured exactly what you asked. Reported by source; not independently reproduced.
10submission 30358KernelBot user 71.90ms1.07×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
11submission 30410KernelBot user 441.91ms1.07×Reported · license unknown · source2025-06-02stale
Measured exactly what you asked. Reported by source; not independently reproduced.
12submission 30332KernelBot user 71.91ms1.07×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
13submission 30353KernelBot user 4151.91ms1.07×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
14submission 30309KernelBot user 441.92ms1.08×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
15submission 29942KernelBot user 441.92ms1.08×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
16submission 30324KernelBot user 71.96ms1.10×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
17submission 30507KernelBot user 3761.98ms1.11×Reported · license unknown · source2025-06-02stale
Measured exactly what you asked. Reported by source; not independently reproduced.
18submission 29936KernelBot user 442.02ms1.13×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
19submission 30286KernelBot user 3762.02ms1.13×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
20submission 30308KernelBot user 72.07ms1.16×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
21submission 30208KernelBot user 3022.12ms1.19×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
22submission 30202KernelBot user 3022.14ms1.20×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
23submission 29860KernelBot user 3762.22ms1.24×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
24submission 29875KernelBot user 72.23ms1.25×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
25submission 29859KernelBot user 3762.24ms1.26×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
26submission 29702KernelBot user 1262.62ms1.47×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
27submission 29488KernelBot user 72.68ms1.50×Reported · license unknown · source2025-05-31stale
Measured exactly what you asked. Reported by source; not independently reproduced.
28submission 28260KernelBot user 1263.32ms1.87×Reported · license unknown · source2025-05-27stale
Measured exactly what you asked. Reported by source; not independently reproduced.
29submission 30453KernelBot user 3343.54ms1.99×Reported · license unknown · source2025-06-02stale
Measured exactly what you asked. Reported by source; not independently reproduced.
30submission 30440KernelBot user 3343.58ms2.01×Reported · license unknown · source2025-06-02stale
Measured exactly what you asked. Reported by source; not independently reproduced.
31submission 32259KernelBot user 2383.74ms2.10×Reported · license unknown · source2025-06-20stale
Measured exactly what you asked. Reported by source; not independently reproduced.
32submission 29823KernelBot user 3344.34ms2.44×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
33submission 29821KernelBot user 3344.39ms2.47×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
34submission 29820KernelBot user 3344.49ms2.52×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
35submission 29283KernelBot user 3344.65ms2.61×Reported · license unknown · source2025-05-30stale
Measured exactly what you asked. Reported by source; not independently reproduced.
36submission 29933KernelBot user 876.69ms3.76×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
37submission 29913KernelBot user 878.92ms5.01×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
38submission 31121KernelBot user 2669.76ms5.48×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
39submission 31106KernelBot user 2669.86ms5.53×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
40submission 31226KernelBot user 2559.94ms5.58×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
41submission 31156KernelBot user 20310.4ms5.82×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
42submission 27739KernelBot user 30211.4ms6.39×Reported · license unknown · source2025-05-26stale
Measured exactly what you asked. Reported by source; not independently reproduced.
43submission 30992KernelBot user 26612.6ms7.05×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
44submission 30988KernelBot user 26612.6ms7.05×Reported · license unknown · source2025-06-03stale
Measured exactly what you asked. Reported by source; not independently reproduced.
45submission 27386KernelBot user 8714.4ms8.07×Reported · license unknown · source2025-05-26stale
Measured exactly what you asked. Reported by source; not independently reproduced.
46submission 29478KernelBot user 25821.1ms11.83×Reported · license unknown · source2025-05-31stale
Measured exactly what you asked. Reported by source; not independently reproduced.
47submission 29855KernelBot user 16024.8ms13.94×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
48submission 29476KernelBot user 726.8ms15.02×Reported · license unknown · source2025-05-31stale
Measured exactly what you asked. Reported by source; not independently reproduced.
49submission 26172KernelBot user 126112.1ms62.91×Reported · license unknown · source2025-05-22stale
Measured exactly what you asked. Reported by source; not independently reproduced.
50submission 26088KernelBot user 302112.9ms63.35×Reported · license unknown · source2025-05-22stale
Measured exactly what you asked. Reported by source; not independently reproduced.
51submission 29868KernelBot user 395117.7ms66.02×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
52submission 28291KernelBot user 212131.4ms73.71×Reported · license unknown · source2025-05-27stale
Measured exactly what you asked. Reported by source; not independently reproduced.
53submission 28389KernelBot user 151135.3ms75.92×Reported · license unknown · source2025-05-27stale
Measured exactly what you asked. Reported by source; not independently reproduced.
54submission 26740KernelBot user 376136.4ms76.55×Reported · license unknown · source2025-05-24stale
Measured exactly what you asked. Reported by source; not independently reproduced.
55submission 26904KernelBot user 334248.6ms139.50×Reported · license unknown · source2025-05-25stale
Measured exactly what you asked. Reported by source; not independently reproduced.
56submission 26892KernelBot user 334281.8ms158.12×Reported · license unknown · source2025-05-25stale
Measured exactly what you asked. Reported by source; not independently reproduced.
57submission 26738KernelBot user 334332.2ms186.39×Reported · license unknown · source2025-05-24stale
Measured exactly what you asked. Reported by source; not independently reproduced.
58submission 30256KernelBot user 266949.3ms532.65×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
59submission 30233KernelBot user 266957.3ms537.14×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
60submission 30230KernelBot user 266961.1ms539.23×Reported · license unknown · source2025-06-01stale
Measured exactly what you asked. Reported by source; not independently reproduced.
19 more rows in this comparison group. Open all in search →