Procyon AI Image Generation / Int 8
This benchmark evaluates how well a system performs in AI-based image creation tasks using INT8 precision. It reflects the kind of workloads seen in modern generative applications, helping assess how efficiently a device can run image synthesis models.
Procyon AI Computer Vision 2.0 / fp16
This test focuses on a system's ability to run AI vision tasks using fp16 precision. By simulating practical workloads like object recognition and image analysis, it highlights how well a device manages optimized inference in real-time applications.
MLPerf Client 2.0.1 / Agentic
This benchmark from MLCommons simulates a short AI agent loop, where a local language model reads a long prompt, calls a tool on the CPU, and continues with the result. It measures how long a system needs to finish data analyst and software engineer tasks, showing how well a device can run local AI agents.
Procyon AI Text Generation
This benchmark evaluates how well a system runs large language models locally, using Phi-3.5, Mistral 7B, Llama 3.1 8B and Llama 2 13B. It measures how quickly the first token appears and how many tokens per second the model generates, reflecting the experience of using an AI chatbot or assistant on the device.