Files
kernbench2/docs/sweeps/short_context_prefill_composite_sweep.csv
T
ywkang 711a9a257f gqa short-context: exclude KV deploy from wall; correct HBM peak; batch scaffold
Metric fixes (test harnesses, deploy artifact + wrong constant):
- wall = max(t_end) - min(t_start): exclude the one-time KV-cache deploy
  from the measured decode/prefill step (it was 92-99% of wall, mapping-
  invariant, and masked the real per-mapping separation).
- PEAK_PE_HBM_BPS 128 -> 256 B/ns (8 channels/PE x 32 GB/s); the old value
  was half the modeled per-PE HBM BW, so hbm_bw_util read >1.0 once wall
  was corrected. All six short-context sweep CSVs regenerated/repatched.

Result: the four KV mappings now separate along a {64,32,16,8}-active-PE
ladder (decode 8-kv/1-kv = 4.8x at 8K, 7.4x at 64K), not "modest/tied" as
before; decode is bandwidth-bound at 46-76% of the 256 GB/s per-PE ceiling.

Report (S5.2 rewrite):
- Replace the tied-wall / 8x-per-PE-util claims (both deploy artifacts)
  with the corrected separation and a density trade-off (per-CUBE KV
  footprint, Fig 16 top panel switched to a wall-invariant metric).
- Add a projected latency-vs-batched-throughput analysis (marked
  projected, not measured): dense mappings win throughput, 1-kv wins
  latency; converges at long context.
- Regenerate Fig 15/16/17/18; fix plot script hardcoded ROOT path.

Batch experiment (Part 2, cube_base):
- Add backward-compatible cube_base=0 scalar to the decode kernel so a
  batched user placed at DPPolicy.cube_start addresses 0-based shards.
  Default preserves single-user behavior (32 decode tests pass unchanged).
- New batch harness (skipped): concurrent B>=2 launches hit a sim routing
  issue (sip0.cube0.pe8); single-user path verified. Concurrency fix next.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-22 16:25:04 -07:00

1.8 KiB

1variantmodekv_per_cubeCS_kvwall_usn_pegemm_utilmath_countmath_pipeline_ushbm_bw_utilhbm_read_mbhbm_write_kbipcq_kbkv_cache_per_cube_mb
2compositeA118819235.312546999435405640.1916200493959853452480.00.029224716.062564.0114688.02.0
3compositeA1181638462.80411199463298640.21548245124172694108800.00.032736732.062564.0229376.04.0
4compositeA11832768117.78396202135924640.2297967187462748221440.00.034843564.062564.0458752.08.0
5compositeA11865536227.7376320068799640.23769780826259174446720.00.0360063128.062564.0917504.016.0
6compositeA224819235.873466499721395320.2178537164896468326240.00.057535616.062564.049152.04.0
7compositeA2241638467.67155399355386320.2309735047926852754400.00.060764132.062564.098304.08.0
8compositeA22432768129.35661752446276320.24166271963282046110720.00.063452564.062564.0196608.016.0
9compositeA22465536254.34501247150266320.24581313146006306223360.00.0644793128.062564.0393216.032.0
10compositeA442819246.88723350110906160.211407653217490913120.00.08804116.062564.016384.08.0
11compositeA4421638489.10037348339473160.2224978327880784127200.00.092300432.062564.032768.016.0
12compositeA44232768173.47797354247328160.2285551253787109555360.00.094628764.062564.065536.032.0
13compositeA44265536342.2399932809714160.23170453934016594111680.00.0958392128.062564.0131072.064.0
14compositeB81819268.7380459999949780.33417254836284596560.00.12010816.062564.00.016.0
15compositeB8116384134.2740459999921880.342141593041773413600.00.12249632.062564.00.032.0
16compositeB8132768265.3460459999982480.346270364150157627680.00.12373364.062564.00.064.0
17compositeB8165536527.490046000029980.348372344353921155840.00.124363128.062564.00.0128.0