zkVM Comparison

This page provides benchmark comparisons between ethrex and other implementations, along with a feature matrix for supported zkVM backends.

ethrex vs RSP (SP1)

RSP (RISC Zero/SP1 Prover) is Succinct's implementation for proving Ethereum blocks. Here's how ethrex compares on mainnet blocks:

Block	Gas Used	ethrex (SP1)	RSP	Difference
23769082	7.9M	2m 23s	1m 27s	+63%
23769083	44.9M	12m 24s	7m 49s	+59%
23769084	27.1M	8m 40s	Failed	—
23769085	22.2M	6m 40s	Failed	—
23769086	28.4M	7m 36s	4m 45s	+60%
23769087	17.5M	6m 02s	Failed	—
23769088	17.5M	4m 50s	2m 59s	+61%
23769089	23.9M	8m 14s	4m 44s	+74%
23769090	24.2M	8m 11s	4m 40s	+75%
23769091	24.4M	7m 02s	Failed	—
23769092	21.7M	6m 35s	4m 01s	+64%

note

RSP failed on several blocks with block gas used mismatch errors. ethrex successfully proved all blocks.

Hardware:

ethrex: AMD EPYC 7713 64-Core, 128GB RAM, RTX 4090
RSP: AMD EPYC 7F72 24-Core, 64GB RAM, RTX 4090

zkVM Backend Comparison

ethrex supports multiple zkVM backends with varying features and maturity levels:

Feature	SP1	RISC Zero	ZisK	OpenVM
Status	Production	Production	Experimental	Experimental
GPU Acceleration	✓	✓	✓	✓
L2 Prover	✓	✓	Planned	Planned
Proof Aggregation	✓	✓	✓	✓

Precompile Support

ZK proving of Ethereum precompiles varies by backend:

Precompile	SP1	RISC Zero	ZisK
ecrecover	✓	✓	✓
SHA256	✓	✓	✓
RIPEMD160	✓	✓	✓
identity	✓	✓	✓
modexp	✓	✓	✓
ecAdd	✓	✓	✓
ecMul	✓	✓	✓
ecPairing	✓	✓	✓
blake2f	✓	✓	✓
KZG point evaluation	✓	⚠️	⚠️
BLS12-381	✓	⚠️	✓
P256 verify	✓	✓	⚠️

Legend: ✓ = Supported with patches, ⚠️ = Limited or requires unstable features

Optimization Impact

ethrex has undergone significant optimization for zkVM proving:

Optimization	Impact	Description
Jumpdest analysis	-15% cycles	Optimized jump destination validation
Trie caching	-50% hash calls	Cache initial node hashes during trie construction
Trie hashing	-75% trie cycles	Improved traversal and RLP encoding
Trie operations	-93% get/insert cycles	Eliminated unnecessary node cloning
Serialized tries	-22% total cycles	Pre-serialize resolved tries, skip decoding
ecPairing patch	-10% total cycles	138k → 6k cycles per operation
ecMul patch	-10% total cycles	Accelerated scalar multiplication

See prover_performance.md for detailed optimization history.

Reproduction Instructions

ethrex Benchmarks

Clone ethrex-replay
Run the prover:

cargo r -r -F "sp1,gpu" -p ethrex-replay -- blocks \
  --action prove \
  --zkvm sp1 \
  --from 23769082 \
  --to 23769092 \
  --rpc-url <RPC_WITH_DEBUG_EXECUTIONWITNESS>

RSP Benchmarks

Clone rsp
Run with CUDA:

SP1_PROVER=cuda cargo r -r \
  --manifest-path bin/host/Cargo.toml \
  --block-number <BLOCK> \
  --rpc-url <RPC> \
  --prove

Hardware Recommendations

Use Case	Minimum	Recommended
Development	32GB RAM, 8 cores	64GB RAM, 16 cores
Production (CPU)	64GB RAM, 32 cores	128GB RAM, 64 cores
Production (GPU)	64GB RAM, RTX 3090	128GB RAM, RTX 4090

GPU proving is significantly faster and recommended for production workloads. All modern NVIDIA GPUs with 24GB+ VRAM are supported.

Keyboard shortcuts

Ethrex