jamesob
Featured
local-llm
A comprehensive guide for building and configuring a high-end local machine to run state-of-the-art LLMs, with detailed hardware choices, BIOS tuning, and Docker-based model serving.
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
The benchmarks are fully reproducible with a single command, include automatic verification against sequential state roots to prevent inflated numbers, and show that signature verification and the virtual machine are not bottlenecks.
Who it is for
- Blockchain developers evaluating Quantova
- Performance engineers benchmarking execution engines
- Validator node operators assessing finality guarantees
- Researchers analyzing state machine throughput
Use cases
- Verifying Quantova node throughput claims on local hardware
- Comparing execution and finality characteristics across different workloads
- Understanding the impact of core count on parallel transaction execution
- Establishing baseline performance for deployment planning
Strengths
- Reproducible via a single bash command with automatic correctness checks
- Measured under realistic workloads including contended hot accounts
- Signature verification and VM throughput far exceed execution throughput, isolating the true bottleneck to serial state access
- Finality does not scale with validator set size due to fixed sampled committee
Considerations
- Throughput measured on Apple M4 (10 cores, 16 GB) – absolute numbers vary with hardware
- Finality measured in-process on a single host, excluding real network propagation delays
- Measurements are compute-path only and do not account for production networking overhead
README quick start
Quantova throughput and finality
Reproducible benchmarks for the Quantova node. Every figure is produced by a harness that checks its parallel result against the sequential state root before it reports, and the throughput figures rerun on your own machine with a single command.
Headline
Around 44,000 transactions a second of execution throughput on a single node, and block finality under one second across a live validator committee. Signature verification and the virtual machine both run far above throughput, so the compute path keeps margin and the bound is serial state access.
Execution throughput
Measured on an Apple M4, ten cores, 16 GB, release build. The full block path runs a module lattice signature verify, a virtual machine transfer, and an account write back.
| workload | parallel | speedup |
|---|---|---|
| independent transfers | 47,712 tx/s | 2.20x |
| a third contended on hot accounts | 47,021 tx/s | 2.14x |
| independent at half the cores | 36,469 tx/s |
Held constant over twenty seconds the floor is 41,431 a second and the median 44,085, so the figure is a rate the node sustains rather than a burst. Signature verification runs at 70,039 a second across the cores and the virtual machine alone at 1,381,895, both far past throughput, which places the ceiling on serial state access, the sparse Merkle state read and written in block order.
Block finality
Measured across a real four validator committee producing and finalising blocks one after another, each block built, attested, aggregated, and verified with a real finality certificate that every node checks.
| percentile | finality |
|---|---|
| median | 607 ms |
| p90 | 620 ms |
| p99 | 635 ms |
Finality lands under a second, and it does not grow with the validator set. A fixed sampled committee carries it, so the certificate stays flat as validators join. This figure is the in process compute floor across the committee, a production network adds propagation between hosts on top.
Honest scope
Execution throughput is the compute path of one node. Finality is measured in process across the committee on a single host, so it is the compute floor and leaves out wide area propagation. Finalised throughput at production scale, with validators running in parallel across real hosts, is the figure the network is built to carry, and it is bounded by execution rather t
Description
Reproducible execution throughput benchmark for the Quantova node
Related repositories
Similar projects matched by category, topics, and programming language.
bjan
Featured
claudemux
Claudemux is a lightweight, dependency-minimal message bus that lets multiple Claude Code sessions communicate in real-time over tmux, enabling one session to ask another a question and receive the answer without manual intervention.
0xwilliamortiz
Featured
openclaude-improved
OpenClaude is an open-source CLI coding agent that runs on any platform and supports a wide range of LLM providers, offering the same tools and workflows as Claude Code.