Skip to content
KernelIndex
Search⌘K

FP8 Mamba2 SSM discretization

25 eligible runs
ssm

FP8-quantized SSM parameter discretization in Mamba2 that computes dt (time step), dA (discretized A matrix), and dB (discretized B matrix) using FP8 linear projections with blockwise scaling.

Fastest reported
16.3µsmean · 79% of SOL
Amir M. Mir | SF TensorLicense unknown · unknown

Reported evidence · last observed 2026-07-27. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Source-native comparison · GPU NVIDIA B200 · Workload suite of 6 cases · mean latency · Protocol SOL-ExecBench evaluation stack v1.1 · mean · 16 results · last observed 2026-08-14Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
16.3µs79% SOL
1.00×
Reported · license unknown · no source
2026-07-27

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

no sourcelicense unknownno install recipeRun detail →
2
16.5µs79% SOL
1.01×
Reported · license unknown · no source
2026-08-14

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

no sourcelicense unknownno install recipeRun detail →
3
16.5µs79% SOL
1.01×
Reported · license unknown · no source
2026-08-14

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

no sourcelicense unknownno install recipeRun detail →
4
16.5µs79% SOL
1.01×
Reported · license unknown · no source
2026-08-01

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

no sourcelicense unknownno install recipeRun detail →
5
16.6µs79% SOL
1.01×
Reported · license unknown · no source
2026-08-14

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

no sourcelicense unknownno install recipeRun detail →
6
16.6µs79% SOL
1.01×
Reported · license unknown · no source
2026-08-14

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

no sourcelicense unknownno install recipeRun detail →
7
17.8µs78% SOL
1.09×
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 →
8
18.5µs77% SOL
1.13×
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 →
9
19.0µs76% SOL
1.16×
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 →
10
20.2µs75% SOL
1.24×
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 →
11
20.6µs75% SOL
1.26×
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 →
12
21.7µs74% SOL
1.33×
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 →
13
26.9µs69% SOL
1.64×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
14
30.7µs67% SOL
1.88×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
15
40.8µs60% SOL
2.50×
Reported · license unknown · no source
2026-07-10

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

no sourcelicense unknownno install recipeRun detail →
16
73.0µs46% SOL
4.47×
Reported · license unknown · no source
2026-07-10

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
16.3µs
1.00×
Reported
license unknown · no source
unknown
16.5µs
1.01×
Reported
license unknown · no source
unknown
16.5µs
1.01×
Reported
license unknown · no source
unknown
16.5µs
1.01×
Reported
license unknown · no source
unknown
16.6µs
1.01×
Reported
license unknown · no source
unknown
16.6µs
1.01×
Reported
license unknown · no source
unknown
17.8µs
1.09×
Reported
license unknown · no source
unknown
18.5µs
1.13×
Reported
license unknown · no source
Show all 25 implementations ›
unknown
18.5µs
1.13×
Reported
license unknown · no source
unknown
18.6µs
1.14×
Reported
license unknown · no source
unknown
19.0µs
1.16×
Reported
license unknown · no source
unknown
25.4µs
1.55×
Reported
license unknown · no source
unknown
39.6µs
2.43×
Reported
license unknown · no source
unknown
18.7µs
1.14×
Reported
license unknown · no source
unknown
20.2µs
1.24×
Reported
license unknown · no source
unknown
20.6µs
1.26×
Reported
license unknown · no source
unknown
21.7µs
1.33×
Reported
license unknown · no source
unknown
26.9µs
1.64×
Reported
license unknown · no source
unknown
26.7µs
1.63×
Reported
license unknown · no source
unknown
30.1µs
1.84×
Reported
license unknown · no source
unknown
30.7µs
1.88×
Reported
license unknown · no source
unknown
30.9µs
1.89×
Reported
license unknown · no source
unknown
40.8µs
2.50×
Reported
license unknown · no source
unknown
73.0µs
4.47×
Reported
license unknown · no source
unknown
283.1µs
17.32×
Reported
license unknown · no source

Semantics

Inputs and outputs
hidden_statesbf16 [batch_size, seq_len, num_heads]
bbf16 [batch_size, seq_len, n_groups, ssm_state_size]
dt_proj_weightbf16 [num_heads, num_heads]
dt_biasbf16 [num_heads]
a_logfp32 [num_heads]
time_step_limit_minfp32 scalar
time_step_limit_maxfp32 scalar
dtbf16 [batch_size, seq_len, num_heads, head_dim]
dafp32 [batch_size, seq_len, num_heads, head_dim, ssm_state_size]
dbfp32 [batch_size, seq_len, num_heads, head_dim, ssm_state_size]
Axes and behavior
seq_lenvariable
head_dimconstant = 128
n_groupsconstant = 8
num_headsconstant = 128
batch_sizevariable
ssm_state_sizeconstant = 128
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias023_fp8_mamba2_ssm_discretizationmodelnvidia-nemotron-3-nano-30b-a3b-fp8sha256da084c3a59ac…
Sources: NVIDIA SOL-ExecBench (2026-08-14)last observed 2026-08-14How records are decidedJSON