Post
88
We measured local LLM inference on real hardware — and published every record.
The **GEEKOM A9 Mega Local LLM Benchmarks** dataset is live on Hugging Face: 72 measured test points on physical A9 Mega 395 (Ryzen AI Max+ 395, 128 GB) and A9 Mega 388 (64 GB) mini PCs — Qwen, DeepSeek, Gemma and MiniCPM models, single device and dual/triple-node cascades, context lengths from 128 to 256K.
Every figure traces back to a public record with its full test command in the source repository on GitHub — including the coverage gaps and the run that failed. No simulations, no vendor estimates.
📊 Explore it interactively: geekom-ai/a9-mega-benchmark-explorer
📦 Load the dataset: geekom-ai/a9-mega-local-llm-benchmarks
🔧 Source of truth: https://github.com/GEEKOMOfficial/geekom-a9-mega-local-llm-benchmarks
The **GEEKOM A9 Mega Local LLM Benchmarks** dataset is live on Hugging Face: 72 measured test points on physical A9 Mega 395 (Ryzen AI Max+ 395, 128 GB) and A9 Mega 388 (64 GB) mini PCs — Qwen, DeepSeek, Gemma and MiniCPM models, single device and dual/triple-node cascades, context lengths from 128 to 256K.
Every figure traces back to a public record with its full test command in the source repository on GitHub — including the coverage gaps and the run that failed. No simulations, no vendor estimates.
📊 Explore it interactively: geekom-ai/a9-mega-benchmark-explorer
📦 Load the dataset: geekom-ai/a9-mega-local-llm-benchmarks
🔧 Source of truth: https://github.com/GEEKOMOfficial/geekom-a9-mega-local-llm-benchmarks