Files
club-3090/results/lucebox-pflash-niah-q8k-20260504-150600/gpu_monitor.csv
noonghunna ebca0c8921 docs(benchmarks): PFlash long-context bench — 131K source ceiling on 1× 3090 (#230)
Closes task #230. Measured PFlash NIAH compression at 16K-260K source
contexts on 1× 24 GB / 3090 single-card.

Result: PFlash works flawlessly up to 131K source. Compresses 131,068
tokens to 6,524 (5%) in 10.8s with NIAH key + answer both retained.
Vanilla llama.cpp pp131072 takes ~257s per Luce's published numbers,
so PFlash alone is ~24× faster at this context. End-to-end TTFT
(PFlash + target prefill on 6.5K) would be ~12-13s vs ~257s = ~20×.

Above 131K, drafter ephemeral forward-pass tensors (K_curr/V_curr/Q_last
at full sequence length) exceed 24 GB. K-cache quantization
(--pflash-k-type q8_0) doesn't help — the failing allocs are
forward-pass not cache, confirmed by separate bench at 200K/260K with
identical OOM at the same layer numbers.

@weicj's PR #78 claim of 24K → 262K dual-GPU phase split is neither
refuted nor reproduced. Their setup was 2× 22 GB Ti with target also
loaded co-resident on one card; the "24K" was target+drafter
combined. Our 131K is drafter-alone on 24 GB. Reproducing 262K
specifically would require investigation of their drafter config
(chunk_size, lookahead, BSA window) — drafter activation footprint
at 200K+ is the binding constraint regardless of GPU count.

Practical recommendation for 24 GB / 3090 single-card users: PFlash
is shippable for source contexts ≤ 131K. The ~24× TTFT speedup is
genuine and quality holds. Above 131K, fall back to vanilla llama.cpp
prefill or wait for upstream drafter optimizations.

Adds:
- BENCHMARKS.md "PFlash long-context compression on 1× 3090" subsection
  with full per-context table + drafter ceiling explanation
- results/lucebox-pflash-niah-20260504-150321/ (BF16 K cache run)
- results/lucebox-pflash-niah-q8k-20260504-150600/ (q8_0 K cache run)

This closes our active investigation of the Luce surface — three
benches done (DFlash same-card 73.97 mean, K8V4 same-card 74.68 mean,
PFlash compression ceiling 131K). Recommendation surface narrows to:
PFlash at ≤131K is the one piece of Luce that beats vLLM dual.yml on
TTFT for that workload class.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-04 15:08:12 +00:00

1.6 KiB

1tsphaseindextemp_cfan_pctpower_wpower_limit_wmem_used_mibmem_total_mibutil_pct
21777907191.039pflash_load043027.92230.001245760
31777907191.039pflash_load149022.60230.001245760
41777907192.089niah_ctx131068047083.13230.00186902457617
51777907192.089niah_ctx131068150025.62230.001245760
61777907193.135niah_ctx1310680510214.50230.00189442457699
71777907193.135niah_ctx131068150026.32230.001245760
81777907194.195niah_ctx1310680530227.52230.001913824576100
91777907194.195niah_ctx131068150023.25230.001245760
101777907195.247niah_ctx1310680530229.01230.00191382457699
111777907195.247niah_ctx131068149022.55230.001245760
121777907196.291niah_ctx1310680540227.81230.001894424576100
131777907196.291niah_ctx131068149022.45230.001245760
141777907197.337niah_ctx13106805356229.25230.00191382457699
151777907197.337niah_ctx131068149022.45230.001245760
161777907198.384niah_ctx13106805452227.66230.001913824576100
171777907198.384niah_ctx131068149022.55230.001245760
181777907199.430niah_ctx13106805454229.18230.00189442457699
191777907199.430niah_ctx131068149022.46230.001245760
201777907200.474niah_ctx13106805356211.57230.00210242457694
211777907200.474niah_ctx131068149022.56230.001245760
221777907201.614niah_ctx13106805358186.76230.00210242457693
231777907201.614niah_ctx131068149022.60230.001245760
241777907202.805niah_ctx13106805358186.52230.0020482457684
251777907202.805niah_ctx131068149022.61230.001245760