Speed Bench

Speed Bench (report v3.0.0)

Web: speed-bench.html · JSON: field-exec-full-bench.json

Grok16 distro 3.0.0 · speed_demo suite v1.0.0 · 3s execution window

Version stamps

Field	Value
Report	3.0.0
Distro	3.0.0 (v3.0.0)
g16 pkg	Grok16-16.2.0
Bench suite	speed_demo @ 1.0.0
Schema	grok16-field-exec-full-bench/v3

Doctrine: data/grok16-speed-bench-version.json

Winners (reference host)

Category	Winner	Compile	Execution
Fastest execution	C++ g16 belt_2_0	2,494 ms	85.3M ops/s
Fastest compile	C g16 belt_2_0	318 ms	79.5M ops/s
Best C	C g16 belt_2_0	318 ms	79.5M ops/s
Best CMake	CMake host g++ -O2	3,682 ms	82.6M ops/s
Best Python	host CPython 3	—	778K ops/s
Best amortized first-run	C g16 belt_2_0	318 ms	60.3M effective

Python has no compile — interpreter lane only.

Reproduce

cd Grok16
git checkout v3.0.0
export G16_PREFIX="$(pwd)"
SPEED_DEMO_TARGET_SEC=3 ./scripts/grok16-toolchain.sh exec-full-bench
./scripts/grok16-toolchain.sh exec-compare

Related

• Uncompiled Execution — chamber, not compiling, compile ahead
• CMake and Linking — toolchain file, g16-ld
• Performance — belt triad, field-nexus-bench