Files
kernbench2/docs/sweeps/short_context_prefill_composite_fused_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

2.1 KiB

1variantmodekv_per_cubeCS_kvwall_usn_pegemm_utilmath_countmath_pipeline_ushbm_bw_utilhbm_read_mbhbm_write_kbipcq_kbkv_cache_per_cube_mb
2composite_fusedA118819235.53158299944201640.3512557265116541480097.325312001436960.029044616.062564.0114688.02.0
3composite_fusedA1181638463.02314799463935640.41960303226187389920208.554240003079170.032622932.062564.0229376.04.0
4composite_fusedA11832768118.00299802135117640.460774462505422120160431.012096005439730.034778864.062564.0458752.08.0
5composite_fusedA11865536227.95666800683736640.483551970372369540640875.92780790591230.0359717128.062564.0917504.016.0
6composite_fusedA224819235.27508350025257320.5747630902258046240096.408704000771050.058511616.062564.049152.04.0
7composite_fusedA2241638465.58160349091561320.6533363888727434960206.590080001652240.062700532.062564.098304.08.0
8composite_fusedA22432768126.19361352364719320.6972691369268810080426.95283200156690.065042964.062564.0196608.016.0
9composite_fusedA22465536247.42072338639758320.720550152526249820320867.6783359506130.0662839128.062564.0393216.032.0
10composite_fusedA442819245.62825600103196160.7872118539828873120092.280384000420570.090470316.062564.016384.08.0
11composite_fusedA4421638487.81637548340065160.86283606655528672480197.743680000901240.0936532.062564.032768.016.0
12composite_fusedA44232768172.19397554247547160.90290731425133455040408.67027200001480.095334364.062564.065536.032.0
13composite_fusedA44265536340.9491752808783160.923549170385355410160830.52345597374440.096202128.062564.0131072.064.0
14composite_fusedB81819268.3378660000041182.036048828365075712121.61385600038990.12081116.062564.00.016.0
15composite_fusedB8116384133.8738660000013382.2092344147653021480260.60112000083550.12286232.062564.00.032.0
16composite_fusedB8132768264.945865999977182.29857442633239773016538.5756479960680.1239264.062564.00.064.0
17composite_fusedB8165536527.089865999976282.343956792936172760881094.52470397886620.124457128.062564.00.0128.0