Files: docs/performance_analysis.md and docs/index.md
The two performance documents report conflicting CPU speed ratios for c6288:
index.md (line 52): ratio 1.2x — VAJAX is 1.2× slower than VACASK on CPU (88.060 ms vs 76.390 ms)performance_analysis.md (line 20): ratio 0.57x — VAJAX is faster than VACASK on CPU (163.684 ms vs 288.800 ms)
This also leads to different GPU speedup figures:
index.md: 4.4x GPU speedup (19.81 ms vs 88.06 ms)performance_analysis.md: 8.3x GPU speedup (19.81 ms vs 163.68 ms)
CLAUDE.md states the authoritative figure as '1.2x on CPU', matching index.md.
The performance_analysis.md tables appear to use different benchmark run data (20 steps at line 20 vs 1k steps in index.md), and the inconsistency should be resolved or explained.
Files:
docs/performance_analysis.mdanddocs/index.mdThe two performance documents report conflicting CPU speed ratios for c6288:
index.md(line 52): ratio 1.2x — VAJAX is 1.2× slower than VACASK on CPU (88.060 ms vs 76.390 ms)performance_analysis.md(line 20): ratio 0.57x — VAJAX is faster than VACASK on CPU (163.684 ms vs 288.800 ms)This also leads to different GPU speedup figures:
index.md: 4.4x GPU speedup (19.81 ms vs 88.06 ms)performance_analysis.md: 8.3x GPU speedup (19.81 ms vs 163.68 ms)CLAUDE.md states the authoritative figure as '1.2x on CPU', matching
index.md.The
performance_analysis.mdtables appear to use different benchmark run data (20 steps at line 20 vs 1k steps inindex.md), and the inconsistency should be resolved or explained.