Speed Bench

Speed Bench (report v4.7.1)

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

Grok16 distro 4.7.1 · speed_demo suite v1.1.0 · schema v5 · 11 runners · 3s window · 2026-06-27

Version stamps

Field	Value
Report	4.7.1
Distro	4.7.1 (v4.7.1)
g16 pkg	Grok16-16.2.0
Bench suite	speed_demo @ 1.1.0
Schema	grok16-field-exec-full-bench/v5

Doctrine: data/grok16-speed-bench-version.json · Plate meld: data/grok16-plate-meld-bench-doctrine.json

Methodology

1. field-plate-meld.py fuse — plate fusion (4 plates on this witness)
2. g16-compiler-sense-plate.py cycle — sense profile ladder
3. Compile every runner (gcc, g16 belt_1_0/2_0, sense field_opt, CMake)
4. Execute speed_demo @ 3s — axis field_execution_ops_per_sec
5. Post-meld re-exec same ELF — verify hot path
6. Cross-ref bench-all from data/bench/latest.json

Kernel

256×16 die · 240 frames/epoch · 512 prog_ops/frame · φ=0.618 · FieldX86 + entropy + wave + NEXUS

Winners (cold exec, BSP rocket)

Category	Winner	Value
Fastest execution	C++ — host g++ -O2	95.3M ops/s
Best g16 C++	C++ — g16 belt_2_0	92.6M ops/s
Fastest compile	C — g16 belt_1_0	357 ms
Best amortized first-run	C — g16 belt_1_0	66.4M eff.
Best Python	host CPython 3	800K ops/s

All executions (reference host @ 2026-06-27)

Runner	Profile	Compile (ms)	ops/s
C++ — host g++ -O2	—	1,588	95.3M
CMake — host g++ -O2	—	2,174	93.0M
C++ — g16 belt_2_0	belt_2_0	1,883	92.6M
C — g16 belt_1_0	belt_1_0	357	90.1M
C++ — g16 sense field_opt	field_opt	1,774	89.6M
C — g16 belt_2_0	belt_2_0	449	87.5M
C++ — g16 belt_1_0	belt_1_0	1,768	87.1M
C — host gcc -O2	—	496	82.6M
Python — host CPython 3	—	—	800K
Python — gpy-16 GrokVM	—	—	751K

Plate meld verdict

Metric	Value
Meld generation	2
Plates fused	4
Sense vs belt_2_0 compile	−109 ms
Sense vs belt_2_0 exec ratio	0.967
Post-meld belt_2 re-exec	82.9M ops/s

Professional conclusion: Plate meld does not block the hot path. Compiler-sense trades a small exec delta for faster wave-convert on this host; profile ladder unlocks when meld generation ≥ 2.

Chart refresh

./scripts/grok16-toolchain.sh bench-refresh
# or: bench-charts after exec-comprehensive-bench

Charts: docs/assets/speed-bench-chart.svg, triad-chart.svg, compare-chart.svg, bench-all-chart.svg

Reproduce

cd Grok16
export G16_PREFIX="$(pwd)"
export GROK16_SG_ROOT=/path/to/SG
export NEXUS_STATE_DIR=$GROK16_SG_ROOT/NewLatest/state
G16_PLATE_MELD_CMD=fuse SPEED_DEMO_TARGET_SEC=3 ./scripts/grok16-toolchain.sh exec-comprehensive-bench

Related

• Uncompiled Execution
• CMake and Linking
• Performance — belt triad + bench-all charts