NVFP4 Mamba2 out projection

38 eligible runs

quantization

FP4-quantized Mamba2 output projection that transforms the normalized SSM output back to hidden dimension. Uses FP4 weights with block-scaled quantization (16 elements per block). Performs matrix multiplication from intermediate_size (2688) to hidden_size (2688). Note: dimensions are padded to 2816 to satisfy 128-multiple constraint for FP4 quantization.

Fastest reported

13.9µsmean · 74% of SOL

Amir M. Mir | SF TensorLicense unknown · unknown

Reported evidence · last observed 2026-08-16. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyhidden_size = 1920 · sf_vec_size = 16 · half_intermediate = 960 · intermediate_size = 1920 · num_hidden_blocks = 120 · num_intermediate_blocks = 12016 cases

m

dtype

128fp4/fp8_e4m3 256fp4/fp8_e4m3 384fp4/fp8_e4m3 512fp4/fp8_e4m3 640fp4/fp8_e4m3 1024fp4/fp8_e4m3 1408fp4/fp8_e4m3 1664fp4/fp8_e4m3 2176fp4/fp8_e4m3 2304fp4/fp8_e4m3 3328fp4/fp8_e4m3 3584fp4/fp8_e4m3 4352fp4/fp8_e4m3 4608fp4/fp8_e4m3 4864fp4/fp8_e4m3 8192fp4/fp8_e4m3

scrolls · 16 cases total

Aggregatessuite of 16 cases · mean latency

HardwareBest knownImplementationRuns

NVIDIA B200 · env 113.9 µsAmir M. Mir | SF Tensor33 NVIDIA B200 · env 214.3 µsRecursive5

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload suite of 16 cases · mean latency · Protocol SOL-ExecBench evaluation stack v1.1 · mean · 33 results · last observed 2026-08-16Record history →

#

Implementation

Latency

vs #1

Trust

Observed

1

Amir M. Mir | SF Tensor

13.9µs74% SOL

1.00×

Reported · license unknown · no source

2026-08-16

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

no sourcelicense unknownno install recipeRun detail →

2

14.2µs73% SOL

1.02×

Reported · license unknown · no source

2026-08-02

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

no sourcelicense unknownno install recipeRun detail →

3

Amir M. Mir | SF Tensor

14.7µs72% SOL

1.05×

Reported · license unknown · no source

2026-08-15

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

no sourcelicense unknownno install recipeRun detail →

4

14.7µs72% SOL

1.06×

Reported · license unknown · no source

2026-08-12

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

no sourcelicense unknownno install recipeRun detail →

5

15.3µs72% SOL

1.10×

Reported · license unknown · no source

2026-07-13

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

no sourcelicense unknownno install recipeRun detail →

6

Amir M. Mir | SF Tensor

15.7µs71% SOL

1.13×

Reported · license unknown · no source

2026-08-15

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

no sourcelicense unknownno install recipeRun detail →

7

Amir M. Mir | SF Tensor

15.9µs71% SOL

1.14×

Reported · license unknown · no source

2026-08-13

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

no sourcelicense unknownno install recipeRun detail →

8

Amir M. Mir | SF Tensor

16.1µs70% SOL

1.16×

Reported · license unknown · no source

2026-08-03

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

no sourcelicense unknownno install recipeRun detail →

9

17.2µs69% SOL

1.23×

Reported · license unknown · no source

2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

10

Amir M. Mir | SF Tensor

18.6µs67% SOL

1.34×

Reported · license unknown · no source

2026-07-19

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

no sourcelicense unknownno install recipeRun detail →

11

22.2µs63% SOL

1.60×

Reported · license unknown · no source

2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

12

22.4µs63% SOL

1.61×

Reported · license unknown · no source

2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

13

22.4µs63% SOL

1.61×

Reported · license unknown · no source

2026-08-09

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

no sourcelicense unknownno install recipeRun detail →

14

22.5µs63% SOL

1.62×

Reported · license unknown · no source

2026-07-16

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

no sourcelicense unknownno install recipeRun detail →

15

25.7µs59% SOL

1.85×

Reported · license unknown · no source

2026-08-03

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

no sourcelicense unknownno install recipeRun detail →

16

30.4µs55% SOL

2.19×

Reported · license unknown · no source

2026-08-11

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

no sourcelicense unknownno install recipeRun detail →

17

31.6µs54% SOL

2.27×

Reported · license unknown · no source

2026-07-22

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

no sourcelicense unknownno install recipeRun detail →

18

37.0µs50% SOL

2.66×

Reported · license unknown · no source

2026-08-03

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

no sourcelicense unknownno install recipeRun detail →

19

40.1µs48% SOL

2.88×

Reported · license unknown · no source

2026-07-18

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

no sourcelicense unknownno install recipeRun detail →

20

Amir M. Mir | SF Tensor

40.5µs48% SOL

2.91×

Reported · license unknown · no source

2026-07-14

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

no sourcelicense unknownno install recipeRun detail →

21

44.1µs46% SOL

3.17×

Reported · license unknown · no source

2026-07-13

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

no sourcelicense unknownno install recipeRun detail →

22

46.8µs44% SOL

3.36×

Reported · license unknown · no source

2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

23

47.3µs44% SOL

3.40×

Reported · license unknown · no source

2026-08-04

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

no sourcelicense unknownno install recipeRun detail →

24

48.5µs43% SOL

3.49×

Reported · license unknown · no source

2026-08-10

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

no sourcelicense unknownno install recipeRun detail →

25

49.8µs43% SOL

3.58×

Reported · license unknown · no source

2026-08-09

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

no sourcelicense unknownno install recipeRun detail →

26

50.8µs42% SOL

3.65×

Reported · license unknown · no source

2026-07-14

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

no sourcelicense unknownno install recipeRun detail →

27

65.1µs36% SOL

4.68×

Reported · license unknown · no source

2026-08-02

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

no sourcelicense unknownno install recipeRun detail →

28

68.2µs35% SOL

4.90×

Reported · license unknown · no source

2026-08-11

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

no sourcelicense unknownno install recipeRun detail →

29

69.9µs35% SOL

5.02×

Reported · license unknown · no source

2026-08-11

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

no sourcelicense unknownno install recipeRun detail →

30

70.2µs35% SOL

5.04×

Reported · license unknown · no source

2026-08-03

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

no sourcelicense unknownno install recipeRun detail →

31

70.7µs35% SOL

5.08×

Reported · license unknown · no source

2026-08-07

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

no sourcelicense unknownno install recipeRun detail →

32

237.6µs14% SOL

17.08×

Reported · license unknown · no source

2026-08-09

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

no sourcelicense unknownno install recipeRun detail →

33

241.0µs13% SOL

17.32×

Reported · license unknown · no source

2026-08-09

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

no sourcelicense unknownno install recipeRun detail →

Implementations

Implementation

Runtime

Best latency

Evidence

Availability

unknown

14.7µs

1.06×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

15.9µs

1.14×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

16.1µs

1.16×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

13.9µs

1.00×

Reported

license unknown · no source

unknown

14.2µs

1.02×

Reported

license unknown · no source

unknown

14.3µs

1.03×

Reported

license unknown · no source

unknown

14.8µs

1.06×

Reported

license unknown · no source

unknown

17.2µs

1.23×

Reported

license unknown · no source

Show all 38 implementations ›

unknown

38.3µs

2.75×

Reported

license unknown · no source

unknown

40.1µs

2.88×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

14.7µs

1.05×

Reported

license unknown · no source

unknown

15.3µs

1.10×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

15.7µs

1.13×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

18.6µs

1.34×

Reported

license unknown · no source

unknown

19.0µs

1.36×

Reported

license unknown · no source

unknown

22.2µs

1.60×

Reported

license unknown · no source

unknown

22.4µs

1.61×

Reported

license unknown · no source

unknown

22.4µs

1.61×

Reported

license unknown · no source

unknown

22.5µs

1.62×

Reported

license unknown · no source

unknown

25.7µs

1.85×

Reported

license unknown · no source

unknown

30.4µs

2.19×

Reported

license unknown · no source

unknown

31.6µs

2.27×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

31.7µs

2.28×

Reported

license unknown · no source

unknown

37.0µs

2.66×

Reported

license unknown · no source

Amir M. Mir | SF Tensor

unknown

40.5µs

2.91×

Reported

license unknown · no source

unknown

44.1µs

3.17×

Reported

license unknown · no source

unknown

46.8µs

3.36×

Reported

license unknown · no source

unknown

47.3µs

3.40×

Reported

license unknown · no source

unknown

48.5µs

3.49×

Reported

license unknown · no source

unknown

49.8µs

3.58×

Reported

license unknown · no source

unknown

50.8µs

3.65×

Reported

license unknown · no source

unknown

65.1µs

4.68×

Reported

license unknown · no source

unknown

68.2µs

4.90×

Reported

license unknown · no source

unknown

69.9µs

5.02×

Reported

license unknown · no source

unknown

70.7µs

5.08×

Reported

license unknown · no source

unknown

70.2µs

5.04×

Reported

license unknown · no source

unknown

237.6µs

17.08×

Reported

license unknown · no source

unknown

241.0µs

17.32×

Reported

license unknown · no source

Semantics

Inputs and outputs

hidden_statesfloat4_e2m1fn_x2 [1, m, half_intermediate]

hidden_states_scalefp8_e4m3 [1, m, num_intermediate_blocks]

global_decode_scale_inputfp32 scalar

weightfloat4_e2m1fn_x2 [1, hidden_size, half_intermediate]

weight_scalefp8_e4m3 [1, hidden_size, num_intermediate_blocks]

global_decode_scale_weightfp32 scalar

outputfp32 [1, m, hidden_size]

Axes and behavior

mvariable

hidden_sizeconstant = 1920

sf_vec_sizeconstant = 16

half_intermediatehalf_intermediate = intermediate_size // 2

intermediate_sizeconstant = 1920

num_hidden_blocksnum_hidden_blocks = hidden_size // sf_vec_size

num_intermediate_blocksnum_intermediate_blocks = intermediate_size // sf_vec_size

determinismunspecified

constraintsNo mutation or aliasing

Identity

alias026_nvfp4_mamba2_out_projectionmodelnvidia-nemotron-3-nano-30b-a3b-nvfp4sha2566014d60249ee…

Sources: NVIDIA SOL-ExecBench (2026-08-16)last observed 2026-08-16How records are decided JSON