Files
kernbench2/docs/sweeps/short_context_decode_composite_fused_sweep.csv
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_fusedA11881922.2471645000005376640.116655456242704286560.00.45768816.06258.057.752.0
3composite_fusedA118163844.090972500001895640.4907507933676371129613.9036159999929370.50171432.06258.057.754.0
4composite_fusedA118327686.65601250000007640.8261186408347434257641.710847999967630.6160664.06258.057.758.0
5composite_fusedA1186553611.786092499984429641.05893212704674513697.325312001436960.695438128.06258.057.7516.0
6composite_fusedA22481923.8395887500001407320.52288099864609426246.9518079999955370.53469316.031258.024.754.0
7composite_fusedA224163846.404628750001605320.858544064722257126420.8554239999894050.64031832.031258.024.758.0
8composite_fusedA2243276811.534708749999874321.0820101547616419254448.662655999962240.71063864.031258.024.7516.0
9composite_fusedA2246553621.794868749984772321.213345411922135104104.277120001539590.751966128.031258.024.7532.0
10composite_fusedA44281925.915787499999919160.929488423979272460810.4277119999933060.69339916.0156258.08.258.0
11composite_fusedA4421638411.045867500001041161.1298951395303998124824.331327999987640.74217832.0156258.08.2516.0
12composite_fusedA4423276821.306027499999384161.2411841672219757252852.138559999959540.76926664.0156258.08.2532.0
13composite_fusedA4426553641.82634749998455161.29996452601033145088107.75302400159090.783573128.0156258.08.2564.0
14composite_fusedB81819210.85690499999985481.14956076339739760012.165663999992190.7552816.00781258.00.016.0
15composite_fusedB811638421.11706500000122681.2522906947682175124026.0692799999867580.77624432.00781258.00.032.0
16composite_fusedB813276841.63738499999977481.3058641410537761252053.87651199995820.78717764.00781258.00.064.0
17composite_fusedB816553682.6780249999731381.333230879731835080109.490976001616570.792762128.00781258.00.0128.0