Benchmarks
Inspect the newest public uploads
Start with the newest results and narrow the list down from there with filters.
Community benchmark hub
Compare hardware for local LLM workloads
Explore public GPU and NPU benchmark uploads, jump into filtered leaderboards and publish your own reproducible results with screenshots.
Jump in
Find the fastest way into the benchmarks
Start with the newest uploads, jump straight into speed and efficiency views, filter for Apple hardware or browse the uploader leaderboard.
Views
Prompt Processing
Compare GPUs and NPUs by prompt throughput and see which local LLM hardware handles long prompts and large context windows the fastest.
Views
Token Generation
See which GPUs and NPUs deliver the highest token generation speed after the first token when streaming output and response speed matter most.
Views
Efficiency per Watt
Compare GPUs and NPUs by token generation per watt to find local LLM hardware that balances speed and lower power consumption.
Apple
Open Apple benchmark results
Open the benchmark list with Apple already selected and compare the available results right away.
Leaderboard
See the most active uploaders
See who has contributed the most results and jump from there into matching benchmark views.
Quick answers
FAQ for local LLM benchmarks and the upcoming software
Get the core answers about comparable hardware results, standardized tests, and the upcoming LocalLLMBench software in one place.
What does LocalLLMBench compare?
LocalLLMBench compares GPU and NPU performance for local LLMs using public benchmark results, filters, and measurable hardware context.
Why is a standardized test important?
Fair hardware comparisons need reproducible measurements with the same benchmark, ideally a standardized test such as Llama-Bench on llama.cpp.
Is a dedicated benchmark software coming?
Yes. A simpler LocalLLMBench software is in development to make standardized benchmarks easier to run and prepare for upload.