Description
ComputeArena is a platform dedicated to providing local AI benchmarks through community contributions. It hosts 694 benchmarks across 36 models and 17 chips, contributed by 8 community members. The platform allows users to decode and prefill tokens per second by model, quantisation, and chip, measured with BaseRT or llama.cpp.
Read More
The benchmarks are submitted by the community and are based on signed reports produced on community hardware. Users can browse results by model, quantisation, and chip, making it easier to compare performance across different configurations. The platform supports macOS and Linux, requiring curl, tar, and Python 3 for installation. Users can install the ComputeArena CLI and choose between BaseRT or llama.cpp as their runtime.
ComputeArena operates offline, meaning no account is required to run benchmarks. Signed reports remain on the user's machine until they decide to submit them. When ready, users can sign in, review their saved reports, and submit the benchmarks they wish to share publicly.
The platform provides headline rankings that compare compatible PP512 prefill and TG128 decode workloads, with the measured workload displayed alongside each result. This feature allows users to gain insights into the performance of different AI models and configurations, aiding in the selection of the best setup for their needs.
- CategoryMachine Learning & Data Science
- Pricing ModelFree
- DeploymentSelf-hosted
- PlatformsmacOS, Linux
ComputeArena's Core Features
Community-submitted benchmarks
Supports macOS and Linux
No account required for offline use
Signed reports for authenticity
Comparison by model, quantisation, and chip
Uses BaseRT or llama.cpp runtimes
Interactive installation menu
Headline rankings for performance comparison
How to use ComputeArena?
Install: Get the ComputeArena CLI for macOS or Linux
Choose Runtime: Select BaseRT or llama.cpp
Run Offline: Execute benchmarks without an account
Submit: Sign in and submit benchmarks when ready
ComputeArena's Use Cases
- Model Performance Comparison
- Community Benchmarking
- Offline Benchmarking
- Quantisation Analysis
- Chip Performance Evaluation
