Skip to content
KernelIndex
Search⌘K
Sign in

Mamba discretization and segsum

28 eligible runs
ssm

Mamba SSM parameter discretization and segment sum computation for chunk-based parallel processing. Discretizes continuous-time SSM parameters using timestep dt, computes cumulative sums for decay factors, and performs segment sum operations for intra-chunk attention-like computations.

Fastest reported
19.6µsmean · 95% of SOL
RIAC as wellLicense unknown · unknown

Reported evidence · last observed 2026-08-22. Reported by source; not independently reproduced. No public source. License unknown.

No public sourceRun detail →

Current records

Workloadsuite of 16 cases · mean latencyd_state = 256 · head_dim = 160 · n_groups = 1 · num_heads = 16 · chunk_size = 25616 cases
batch_size
num_chunks
dtype
11bf1612bf1613bf1618bf1622bf16216bf1641bf1642bf1643bf1645bf1688bf16161bf16162bf16321bf16322bf16641bf16
scrolls · 16 cases total
Aggregatessuite of 16 cases · mean latency
HardwareBest knownImplementationRuns
NVIDIA B200 · env 119.6 µsRIAC as well19NVIDIA B200 · env 229.1 µsRecursive9

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 · 19 results · last observed 2026-08-23Record history →
#
Implementation
Latency
vs #1
Trust
Observed
1
RIAC as well
19.6µs95% SOL
1.00×
Reported · license unknown · no source
2026-08-22

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

no sourcelicense unknownno install recipeRun detail →
2
Geometric
25.0µs92% SOL
1.28×
Reported · license unknown · no source
2026-08-23

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

no sourcelicense unknownno install recipeRun detail →
3
Hyra
25.8µs91% SOL
1.31×
Reported · license unknown · no source
2026-07-17

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

no sourcelicense unknownno install recipeRun detail →
4
Recursive
26.6µs91% SOL
1.36×
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 →
5
ac4k
27.2µs91% SOL
1.39×
Reported · license unknown · no source
2026-07-21

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

no sourcelicense unknownno install recipeRun detail →
6
Q H
30.2µs89% SOL
1.54×
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 →
7
Amir M. Mir | SF Tensor
30.4µs88% SOL
1.55×
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 →
8
doubleAI
31.6µs87% SOL
1.61×
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 →
9
Geometric
31.8µs86% SOL
1.62×
Reported · license unknown · no source
2026-08-22

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

no sourcelicense unknownno install recipeRun detail →
10
Q H
36.8µs85% SOL
1.88×
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
Proud Wren @ NVIDIA
36.8µs83% 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 →
12
Geometric
38.5µs81% SOL
1.97×
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 →
13
Geometric
38.6µs81% SOL
1.97×
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 →
14
Geometric
48.3µs78% SOL
2.47×
Reported · license unknown · no source
2026-08-05

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

no sourcelicense unknownno install recipeRun detail →
15
Amir M. Mir | SF Tensor
48.5µs77% SOL
2.48×
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 →
16
Amir M. Mir | SF Tensor
52.6µs75% SOL
2.68×
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 →
17
Runic Kingfisher
63.7µs69% SOL
3.25×
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 →
18
Amir M. Mir | SF Tensor
63.8µs70% SOL
3.25×
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 →
19
Cursor
67.2µs70% SOL
3.43×
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
RIAC as well
unknown
19.6µs
1.00×
Reported
license unknown · no source
Geometric
unknown
25.0µs
1.28×
Reported
license unknown · no source
Q H
unknown
30.2µs
1.54×
Reported
license unknown · no source
doubleAI
unknown
29.9µs
1.53×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
30.4µs
1.55×
Reported
license unknown · no source
Proud Wren @ NVIDIA
unknown
36.8µs
1.88×
Reported
license unknown · no source
Cursor
unknown
65.6µs
3.35×
Reported
license unknown · no source
Runic Kingfisher
unknown
63.7µs
3.25×
Reported
license unknown · no source
Show all 28 implementations ›
Recursive
unknown
29.1µs
1.48×
Reported
license unknown · no source
doubleAI
unknown
31.6µs
1.61×
Reported
license unknown · no source
Geometric
unknown
31.8µs
1.62×
Reported
license unknown · no source
Q H
unknown
36.8µs
1.88×
Reported
license unknown · no source
Proud Wren @ NVIDIA
unknown
36.9µs
1.88×
Reported
license unknown · no source
doubleAI
unknown
39.5µs
2.02×
Reported
license unknown · no source
Geometric
unknown
38.5µs
1.97×
Reported
license unknown · no source
Geometric
unknown
38.6µs
1.97×
Reported
license unknown · no source
Geometric
unknown
48.3µs
2.47×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
48.5µs
2.48×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
52.6µs
2.68×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
60.5µs
3.09×
Reported
license unknown · no source
Cursor
unknown
67.2µs
3.43×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
63.8µs
3.25×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
99.0µs
5.05×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
148.4µs
7.57×
Reported
license unknown · no source
Amir M. Mir | SF Tensor
unknown
303.1µs
15.47×
Reported
license unknown · no source
Hyra
unknown
25.8µs
1.31×
Reported
license unknown · no source
Recursive
unknown
26.6µs
1.36×
Reported
license unknown · no source
ac4k
unknown
27.2µs
1.39×
Reported
license unknown · no source

Semantics

Inputs and outputs
hidden_statesbf16 [batch_size, num_chunks, chunk_size, num_heads, head_dim]
dtbf16 [batch_size, num_chunks, chunk_size, num_heads]
a_logbf16 [num_heads]
bbf16 [batch_size, num_chunks, chunk_size, n_groups, d_state]
cbf16 [batch_size, num_chunks, chunk_size, n_groups, d_state]
lbf16 [batch_size, num_heads, num_chunks, chunk_size, chunk_size]
gbf16 [batch_size, num_chunks, chunk_size, chunk_size, num_heads]
mbf16 [batch_size, num_chunks, chunk_size, chunk_size, num_heads]
a_cumsumfp32 [batch_size, num_heads, num_chunks, chunk_size]
decay_statesfp32 [batch_size, num_heads, num_chunks, chunk_size]
Axes and behavior
d_stateconstant = 256
head_dimconstant = 160
n_groupsconstant = 1
num_headsconstant = 16
batch_sizevariable
chunk_sizeconstant = 256
num_chunksvariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
alias044_mamba_discretization_and_segsummodelgranite-4-0-microsha256e17b608d82e7…
Sources: NVIDIA SOL-ExecBench (2026-08-23)last observed 2026-08-23How records are decidedJSON