Skip to content
KernelIndex
Search⌘K

AMD FP8 blockwise GEMM

600 eligible runs
gemm

FP8-blockwise matrix multiply optimized for MI300: A (m×k) and B (n×k) in e4m3fnuz column-major with fp32 blockwise scales (128-wide k blocks), output C (m×n) in bf16 row-major. Shapes are drawn from DeepSeek-R1.

Fastest reported
33.1µsmean of 100
submission 27020KernelBot user 3 · License unknown · python

Reported evidence · last observed 2025-05-25. Reported by source; not independently reproduced. License unknown.

Current records

Source-native comparison · GPU AMD Instinct MI300X · Workload k = 512 · m = 1024 · n = 4096 · seed = 17512 · fp32/fp8_e4m3fnuz · Protocol reference-kernels eval.py · mean · 91 results · last observed 2025-08-22Record history →
Estimated floor 1.64 µs · record 20.15× above itestimate, not evidence ›
DRAM 498 ns · compute 1.64 µs · compute-bound on MI300X
every declared tensor crosses HBM exactly once (5,300 GB/s, MI300X datasheet)
2·M·N·K with M=1024, N=4096, K=512 at the dense fp8 peak (2,615 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
1
submission 27020KernelBot user 3
33.1µs
1.00×
Reported · license unknown · source
2025-05-25stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
2
submission 27258KernelBot user 3
33.5µs
1.01×
Reported · license unknown · source
2025-05-25stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
3
submission 28436KernelBot user 3
33.8µs
1.02×
Reported · license unknown · source
2025-05-27stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
4
submission 29878KernelBot user 376
34.9µs
1.05×
Reported · license unknown · source
2025-06-01stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
5
submission 25056KernelBot user 3
34.9µs
1.06×
Reported · license unknown · source
2025-05-19stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
6
submission 29876KernelBot user 376
35.0µs
1.06×
Reported · license unknown · source
2025-06-01stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
7
submission 25269KernelBot user 376
35.2µs
1.07×
Reported · license unknown · source
2025-05-20stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
8
submission 18889KernelBot user 376
35.6µs
1.07×
Reported · license unknown · source
2025-05-09stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
9
submission 30673KernelBot user 44
35.7µs
1.08×
Reported · license unknown · source
2025-06-02stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
10
submission 27935KernelBot user 3
35.8µs
1.08×
Reported · license unknown · source
2025-05-26stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
11
submission 30689KernelBot user 44
35.8µs
1.08×
Reported · license unknown · source
2025-06-02stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
12
submission 30771KernelBot user 44
36.0µs
1.09×
Reported · license unknown · source
2025-06-02stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
13
submission 9828KernelBot user 3
36.2µs
1.10×
Reported · license unknown · source
2025-04-27stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
14
submission 29949KernelBot user 44
36.5µs
1.10×
Reported · license unknown · source
2025-06-01stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
15
submission 25255KernelBot user 376
36.6µs
1.10×
Reported · license unknown · source
2025-05-20stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
16
submission 18667KernelBot user 376
36.6µs
1.11×
Reported · license unknown · source
2025-05-09stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
17
submission 27162KernelBot user 160
36.8µs
1.11×
Reported · license unknown · source
2025-05-25stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
18
submission 30688KernelBot user 44
36.9µs
1.12×
Reported · license unknown · source
2025-06-02stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
19
submission 34192KernelBot user 16
37.4µs
1.13×
Reported · license unknown · source
2025-08-22stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
20
submission 19014KernelBot user 376
37.4µs
1.13×
Reported · license unknown · source
2025-05-09stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
21
submission 30110KernelBot user 44
37.5µs
1.13×
Reported · license unknown · source
2025-06-01stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
22
submission 31540KernelBot user 292
38.4µs
1.16×
Reported · license unknown · source
2025-06-06stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
23
submission 27143KernelBot user 292
39.2µs
1.18×
Reported · license unknown · source
2025-05-25stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
24
submission 29932KernelBot user 44
39.4µs
1.19×
Reported · license unknown · source
2025-06-01stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
25
submission 24658KernelBot user 292
40.1µs
1.21×
Reported · license unknown · source
2025-05-19stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
26
submission 18833KernelBot user 160
40.3µs
1.22×
Reported · license unknown · source
2025-05-09stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
27
submission 9306KernelBot user 3
40.4µs
1.22×
Reported · license unknown · source
2025-04-26stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
28
submission 18884KernelBot user 160
40.7µs
1.23×
Reported · license unknown · source
2025-05-09stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
29
submission 21570KernelBot user 212
40.9µs
1.24×
Reported · license unknown · source
2025-05-13stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
30
submission 30355KernelBot user 415
41.1µs
1.24×
Reported · license unknown · source
2025-06-01stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
31
submission 24311KernelBot user 292
41.4µs
1.25×
Reported · license unknown · source
2025-05-18stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
32
submission 7548KernelBot user 376
41.8µs
1.26×
Reported · license unknown · source
2025-04-23stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
33
submission 33220KernelBot user 346
42.0µs
1.27×
Reported · license unknown · source
2025-07-11stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
34
submission 23187KernelBot user 7
42.9µs
1.30×
Reported · license unknown · source
2025-05-16stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
35
submission 6008KernelBot user 3
43.1µs
1.30×
Reported · license unknown · source
2025-04-20stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
36
submission 19003KernelBot user 376
43.1µs
1.30×
Reported · license unknown · source
2025-05-09stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
37
submission 33210KernelBot user 346
43.4µs
1.31×
Reported · license unknown · source
2025-07-10stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
38
submission 23139KernelBot user 423
43.9µs
1.33×
Reported · license unknown · source
2025-05-16stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
39
submission 29705KernelBot user 87
44.0µs
1.33×
Reported · license unknown · source
2025-06-01stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
40
submission 18428KernelBot user 292
44.2µs
1.33×
Reported · license unknown · source
2025-05-08stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
41
submission 30576KernelBot user 415
44.3µs
1.34×
Reported · license unknown · source
2025-06-02stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
42
submission 30578KernelBot user 415
44.5µs
1.34×
Reported · license unknown · source
2025-06-02stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
43
submission 21778KernelBot user 334
45.0µs
1.36×
Reported · license unknown · source
2025-05-13stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
44
submission 17835KernelBot user 292
45.0µs
1.36×
Reported · license unknown · source
2025-05-08stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
45
submission 17326KernelBot user 292
45.0µs
1.36×
Reported · license unknown · source
2025-05-07stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
46
submission 30582KernelBot user 7
45.0µs
1.36×
Reported · license unknown · source
2025-06-02stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
47
submission 32127KernelBot user 213
45.2µs
1.37×
Reported · license unknown · source
2025-06-15stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
48
submission 17827KernelBot user 292
46.0µs
1.39×
Reported · license unknown · source
2025-05-08stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
49
submission 18398KernelBot user 160
46.3µs
1.40×
Reported · license unknown · source
2025-05-08stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
50
submission 22782KernelBot user 334
46.9µs
1.42×
Reported · license unknown · source
2025-05-15stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
51
submission 13557KernelBot user 160
47.7µs
1.44×
Reported · license unknown · source
2025-04-30stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
52
submission 10058KernelBot user 160
48.1µs
1.45×
Reported · license unknown · source
2025-04-27stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
53
submission 33182KernelBot user 346
48.6µs
1.47×
Reported · license unknown · source
2025-07-09stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
54
submission 5810KernelBot user 3
48.6µs
1.47×
Reported · license unknown · source
2025-04-20stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
55
submission 9617KernelBot user 160
48.9µs
1.48×
Reported · license unknown · source
2025-04-26stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
56
submission 8734KernelBot user 160
48.9µs
1.48×
Reported · license unknown · source
2025-04-25stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
57
submission 21793KernelBot user 334
49.0µs
1.48×
Reported · license unknown · source
2025-05-13stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
58
submission 21355KernelBot user 334
49.7µs
1.50×
Reported · license unknown · source
2025-05-13stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
59
submission 21230KernelBot user 7
49.9µs
1.51×
Reported · license unknown · source
2025-05-12stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →
60
submission 5798KernelBot user 3
51.0µs
1.54×
Reported · license unknown · source
2025-04-20stale

Measured exactly what you asked. Reported by source; not independently reproduced.

source mirroredlicense unknownno install recipeView source →Run detail →

31 more rows in this comparison group. Open all in search →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
submission 30771KernelBot user 44
python
34.4µs
1.04×
Reported
license unknown · source
submission 34192KernelBot user 16
python
35.9µs
1.09×
Reported
license unknown · source
submission 29878KernelBot user 376
python
34.9µs
1.05×
Reported
license unknown · source
submission 28436KernelBot user 3
python
33.7µs
1.02×
Reported
license unknown · source
submission 27162KernelBot user 160
python
36.8µs
1.11×
Reported
license unknown · source
submission 30578KernelBot user 415
python
44.5µs
1.34×
Reported
license unknown · source
submission 30582KernelBot user 7
python
45.0µs
1.36×
Reported
license unknown · source
submission 31540KernelBot user 292
python
38.2µs
1.15×
Reported
license unknown · source
Show all 98 implementations ›
submission 33220KernelBot user 346
python
39.4µs
1.19×
Reported
license unknown · source
submission 22782KernelBot user 334
python
46.9µs
1.42×
Reported
license unknown · source
submission 32127KernelBot user 213
python
45.2µs
1.37×
Reported
license unknown · source
submission 29705KernelBot user 87
python
44.0µs
1.33×
Reported
license unknown · source
submission 21570KernelBot user 212
python
40.9µs
1.24×
Reported
license unknown · source
submission 23394KernelBot user 333
python
61.4µs
1.86×
Reported
license unknown · source
submission 23139KernelBot user 423
python
43.9µs
1.33×
Reported
license unknown · source
submission 29791KernelBot user 126
python
79.1µs
2.39×
Reported
license unknown · source
submission 23712KernelBot user 322
python
77.1µs
2.33×
Reported
license unknown · source
submission 31736KernelBot user 364
python
76.5µs
2.31×
Reported
license unknown · source
submission 20350KernelBot user 267
python
54.6µs
1.65×
Reported
license unknown · source
submission 17892KernelBot user 105
python
70.4µs
2.13×
Reported
license unknown · source
submission 20620KernelBot user 395
python
58.9µs
1.78×
Reported
license unknown · source
submission 15295KernelBot user 115
python
62.6µs
1.89×
Reported
license unknown · source
submission 19189KernelBot user 44
python
54.7µs
1.65×
Reported
license unknown · source
submission 19191KernelBot user 44
python
50.6µs
1.53×
Reported
license unknown · source
submission 29932KernelBot user 44
python
37.7µs
1.14×
Reported
license unknown · source
submission 29949KernelBot user 44
python
33.9µs
1.02×
Reported
license unknown · source
submission 30110KernelBot user 44
python
35.9µs
1.09×
Reported
license unknown · source
submission 30673KernelBot user 44
python
34.4µs
1.04×
Reported
license unknown · source
submission 30688KernelBot user 44
python
34.4µs
1.04×
Reported
license unknown · source
submission 30689KernelBot user 44
python
34.5µs
1.04×
Reported
license unknown · source
submission 33458KernelBot user 16
python
—
—
license unknown · source
submission 34190KernelBot user 16
python
—
—
license unknown · source
submission 7548KernelBot user 376
python
41.8µs
1.26×
Reported
license unknown · source
submission 18667KernelBot user 376
python
36.6µs
1.11×
Reported
license unknown · source
submission 18889KernelBot user 376
python
35.6µs
1.07×
Reported
license unknown · source
submission 19003KernelBot user 376
python
43.1µs
1.30×
Reported
license unknown · source
submission 19014KernelBot user 376
python
37.4µs
1.13×
Reported
license unknown · source
submission 25255KernelBot user 376
python
36.6µs
1.10×
Reported
license unknown · source
submission 25269KernelBot user 376
python
35.2µs
1.07×
Reported
license unknown · source
submission 29876KernelBot user 376
python
35.0µs
1.06×
Reported
license unknown · source
submission 3700KernelBot user 3
python
—
—
license unknown · source
submission 5153KernelBot user 3
python
47.8µs
1.45×
Reported
license unknown · source
submission 5798KernelBot user 3
python
42.8µs
1.29×
Reported
license unknown · source
submission 5810KernelBot user 3
python
40.4µs
1.22×
Reported
license unknown · source
submission 6008KernelBot user 3
python
43.1µs
1.30×
Reported
license unknown · source
submission 9306KernelBot user 3
python
40.4µs
1.22×
Reported
license unknown · source
submission 9828KernelBot user 3
python
36.1µs
1.09×
Reported
license unknown · source
submission 25056KernelBot user 3
python
34.9µs
1.06×
Reported
license unknown · source
submission 27020KernelBot user 3
python
33.1µs
1.00×
Reported
license unknown · source
submission 27258KernelBot user 3
python
33.5µs
1.01×
Reported
license unknown · source
submission 27935KernelBot user 3
python
35.2µs
1.06×
Reported
license unknown · source
submission 3803KernelBot user 160
python
—
—
license unknown · source
submission 5109KernelBot user 160
python
67.0µs
2.03×
Reported
license unknown · source
submission 5193KernelBot user 160
python
53.6µs
1.62×
Reported
license unknown · source
submission 8178KernelBot user 160
python
58.3µs
1.76×
Reported
license unknown · source
submission 8734KernelBot user 160
python
48.9µs
1.48×
Reported
license unknown · source
submission 9617KernelBot user 160
python
48.9µs
1.48×
Reported
license unknown · source
submission 10058KernelBot user 160
python
48.1µs
1.45×
Reported
license unknown · source
submission 13557KernelBot user 160
python
47.7µs
1.44×
Reported
license unknown · source
submission 18398KernelBot user 160
python
46.3µs
1.40×
Reported
license unknown · source
submission 18833KernelBot user 160
python
40.3µs
1.22×
Reported
license unknown · source
submission 18884KernelBot user 160
python
40.7µs
1.23×
Reported
license unknown · source
submission 30355KernelBot user 415
python
41.1µs
1.24×
Reported
license unknown · source
submission 30576KernelBot user 415
python
44.3µs
1.34×
Reported
license unknown · source
submission 10955KernelBot user 7
python
—
—
license unknown · source
submission 12305KernelBot user 7
python
124.5µs
3.76×
Reported
license unknown · source
submission 12307KernelBot user 7
python
113.3µs
3.43×
Reported
license unknown · source
submission 12401KernelBot user 7
python
113.6µs
3.43×
Reported
license unknown · source
submission 14641KernelBot user 7
python
56.4µs
1.70×
Reported
license unknown · source
submission 14649KernelBot user 7
python
57.9µs
1.75×
Reported
license unknown · source
submission 16188KernelBot user 7
python
56.7µs
1.71×
Reported
license unknown · source
submission 16190KernelBot user 7
python
49.9µs
1.51×
Reported
license unknown · source
submission 20839KernelBot user 7
python
53.6µs
1.62×
Reported
license unknown · source
submission 21230KernelBot user 7
python
49.9µs
1.51×
Reported
license unknown · source
submission 23187KernelBot user 7
python
42.9µs
1.30×
Reported
license unknown · source
submission 7722KernelBot user 292
python
64.6µs
1.95×
Reported
license unknown · source
submission 8915KernelBot user 292
python
51.4µs
1.55×
Reported
license unknown · source
submission 9802KernelBot user 292
python
49.7µs
1.50×
Reported
license unknown · source
submission 17326KernelBot user 292
python
43.4µs
1.31×
Reported
license unknown · source
submission 17827KernelBot user 292
python
44.6µs
1.35×
Reported
license unknown · source
submission 17835KernelBot user 292
python
43.5µs
1.31×
Reported
license unknown · source
submission 18428KernelBot user 292
python
42.8µs
1.29×
Reported
license unknown · source
submission 24311KernelBot user 292
python
41.4µs
1.25×
Reported
license unknown · source
submission 24658KernelBot user 292
python
39.8µs
1.20×
Reported
license unknown · source
submission 27143KernelBot user 292
python
39.2µs
1.18×
Reported
license unknown · source
submission 33182KernelBot user 346
python
47.0µs
1.42×
Reported
license unknown · source
submission 33210KernelBot user 346
python
39.4µs
1.19×
Reported
license unknown · source
submission 13410KernelBot user 334
python
—
—
license unknown · source
submission 16197KernelBot user 334
python
—
—
license unknown · source
submission 16780KernelBot user 334
python
113.7µs
3.44×
Reported
license unknown · source
submission 18206KernelBot user 334
python
99.8µs
3.02×
Reported
license unknown · source
submission 19663KernelBot user 334
python
78.0µs
2.36×
Reported
license unknown · source
submission 19674KernelBot user 334
python
73.7µs
2.23×
Reported
license unknown · source
submission 20999KernelBot user 334
python
56.9µs
1.72×
Reported
license unknown · source
submission 21025KernelBot user 334
python
54.4µs
1.64×
Reported
license unknown · source
submission 21355KernelBot user 334
python
49.6µs
1.50×
Reported
license unknown · source
submission 21778KernelBot user 334
python
44.8µs
1.35×
Reported
license unknown · source
submission 21793KernelBot user 334
python
49.0µs
1.48×
Reported
license unknown · source

Semantics

Inputs and outputs
afp8_e4m3fnuz [m, k] · col_major
bfp8_e4m3fnuz [n, k] · col_major
a_scalefp32 [m, k_blocks] · col_major
b_scalefp32 [n_blocks, k_blocks] · col_major
cbf16 [m, n] · row_major
Axes and behavior
kvariable
mvariable
nvariable
seedvariable
k_blocksk_blocks = k // 128
n_blocksn_blocks = n // 128
determinismunspecified
constraintsNo mutation or aliasing
Identity
aliasamd-fp8-mmmodeldeepseek-r1sha256a1fd649f9f2b…
Sources: GPU MODE KernelBot (2025-08-22) · June 9 Researcher Reciprocity License v1.0last observed 2025-08-22How records are decidedJSON