Skip to content
KernelIndex
Search⌘K
Sign in

GQA paged prefill causal h24 kv4 d128 ps1

20 eligible runs
gqa-paged-attention

Batched Grouped Query Attention prefill with a paged KV cache. Causal mask is applied. Captured from Mixtral 8x22B at TP=2. 24 q-heads, 4 kv-heads, head_dim=128, page_size=1.

Source baseline · unbeaten
35.8µsmean
flashinfer / wrapper0b1670FlashInfer-Bench baselines · Apache-2.0 · python

Reported evidence · last observed 2026-03-31. The source's designated baseline implementation. Reported by source; not independently reproduced.

No package · vendor the source →View source →Run detail →

Current records

Workloadtotal_q = 154 · num_pages = 155 · len_indptr = 2 · num_kv_indices = 154 · bf1620 cases
num_kv_indices
total_q
num_pages
len_indptr
dtype
797968102bf16828231933bf161541541552bf1617017022033bf1624724770574bf16332332313510bf1641341358854bf165595597182bf1655955958782bf16731731319120bf1698798780604bf161307130720292bf1617281728588246bf161825182524794bf161834183467307bf161932193251932bf1624062406673016bf1647434743819330bf16476347638193125bf164812481281938bf16
scrolls · 20 cases total

Not measured on H100 for this workload. Challenges →

Source-native comparison · GPU NVIDIA B200 · Workload total_q = 154 · num_pages = 155 · len_indptr = 2 · num_kv_indices = 154 · bf16 · CUDA 12.8 · Framework pytorch 2.9.1+cu128 · Protocol flashinfer-bench · mean · 1 results · last observed 2026-03-31Record history →
Estimated floor 158 ns · record 226.35× above itestimate, not evidence ›
DRAM 158 ns · bandwidth-bound on B200
every declared tensor crosses HBM exactly once (8,000 GB/s, B200 datasheet)
no arithmetic formula for this family: bandwidth floor only
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
flashinfer / wrapper0b1670baselineFlashInfer-Bench baselines
35.8µs
1.00×
Reported · Apache-2.0 · source
2026-03-31stale

Measured exactly what you asked. The source's designated baseline implementation. Reported by source; not independently reproduced.

source mirroredApache-2.0no install recipeView source →Run detail →

Implementations

Implementation
Runtime
Best latency
Evidence
Availability
flashinfer / wrapper0b1670FlashInfer-Bench baselines
python
35.8µs
1.00×
Reported
Apache-2.0 · source

Semantics

Inputs and outputs
qbf16 [total_q, num_qo_heads, head_dim]
k_cachebf16 [num_pages, page_size, num_kv_heads, head_dim]
v_cachebf16 [num_pages, page_size, num_kv_heads, head_dim]
qo_indptrint32 [len_indptr]
kv_indptrint32 [len_indptr]
kv_indicesint32 [num_kv_indices]
sm_scalefp32 scalar
outputbf16 [total_q, num_qo_heads, head_dim]
lsefp32 [total_q, num_qo_heads]
Axes and behavior
total_qvariable
head_dimconstant = 128
num_pagesvariable
page_sizeconstant = 1
len_indptrvariable
num_kv_headsconstant = 4
num_qo_headsconstant = 24
num_kv_indicesvariable
determinismunspecified
constraintsNo mutation or aliasing
Identity
aliasgqa_paged_prefill_causal_h24_kv4_d128_ps1modelmixtral-8x22bsha256689156f6551d…
Sources: FlashInfer-Bench (2026-03-31) · Apache-2.0last observed 2026-03-31How records are decidedJSON