BENCHMARK_RESULTS_f277721d.md

Transaction Benchmark Results - Commit f277721d

Generated: 2026-02-20

---

Machine Characteristics

Property	Value
Model	MacBook Pro (Mac16,6)
Chip	Apple M4 Max
CPU Cores	16 (12 performance + 4 efficiency)
Memory	128 GB
Architecture	ARM64 (aarch64-apple-darwin)
OS	Darwin 24.6.0
Firmware	11881.140.96

---

Commit Characteristics

Property	Value
Commit Hash	f277721d3ae89b121d30c535c592e124425fa2ca
Commit Message	Adjust output_note add_asset input order
Author	François Garillot (francois@garillot.net)
Date	Sun Feb 15 06:08:09 2026 -0500
Branch	huitseeker/huitseeker/update-miden-deps-p3

---

Build Configuration

Property	Value
Rust Version	rustc 1.93.1 (stable)
Cargo Version	cargo 1.93.0
Features	concurrent
Build Profile	release (for benchmarks)

---

Timing Benchmarks

These benchmarks measure wall-clock time using Criterion.

Execute Transaction (without proving)

Transaction Type	Mean Time	Range
Consume single P2ID note	2.60 ms	[2.58 ms - 2.62 ms]
Consume two P2ID notes	3.00 ms	[2.97 ms - 3.03 ms]

Execute and Prove Transaction

Transaction Type	Mean Time	Range	Notes
Execute and prove single P2ID note	597 ms	[588 ms - 611 ms]	Proof size: ~218KB
Execute and prove two P2ID notes	1.13 s	[1.12 s - 1.15 s]	Proof size: ~218KB

---

Comparison with next Branch

Metric	next Branch	f277721d	Delta
Execute single P2ID	6.25 ms	2.60 ms	-58.4%
Execute two P2ID	7.88 ms	3.00 ms	-61.9%
Execute+Prove single P2ID	460.35 ms	597 ms	+29.7%
Execute+Prove two P2ID	467.18 ms	1.13 s	+142%
miden-vm version	0.20.3	0.22.0	Updated

---

Summary

Performance Changes

Execution (improved):

• Single P2ID note consumption: ~58% faster
• Two P2ID notes consumption: ~62% faster

Proving (regressed):

• Single P2ID proving: ~30% slower
• Two P2ID proving: ~142% slower

Notes

• The execution performance improvements come from updated miden-vm dependencies (v0.22.0)
• The proving time regression may be due to changes in the STARK prover implementation or different proof options
• Benchmarks were fixed to use prove_async() instead of prove() to avoid Tokio runtime nesting issues
• The proving async function is now properly awaited in the benchmark closures