Ollamometer Benchmarks with Strix Halo + Deepseek and GPT-OSS

Benchmarking DeepSeek and GPT-OSS LLMs on the GMKtec EVO X2 (Strix Halo, Ryzen AI Max+ 395, 128GB RAM) using Ollamometer — a custom Python Flask app for measuring token speeds, time to first token, and memory usage across different model sizes. Ollamometer (open source): https://github.com/hakehardware/ollamometer Products in this video (affiliate links support the channel at no extra cost to you): - The mini PC being benchmarked — GMKtec EVO-X2 (Ryzen AI Max+ 395): https://www.amazon.com/dp/B0F53MLYQ6?tag=hakehardware-20 - Store dozens of LLM models locally — Crucial P3 4TB NVMe SSD: https://www.amazon.com/dp/B0B25P44CL?tag=hakehardware-20 - Keep long AI benchmarks running during outages — Tripp Lite 1500VA UPS: https://www.amazon.com/dp/B000DZRY9C?tag=hakehardware-20 - Keep your NVMe cool under heavy AI workloads — ELUTENG SSD Heatsink Kit: https://www.amazon.com/dp/B0CR69GG3W?tag=hakehardware-20 My other AMD mini PCs (upcoming benchmark comparisons): - GMKtec M7 (680M) with Oculink: https://www.amazon.com/dp/B0DQKRXKS5?tag=hakehardware-20 - GMKtec K12 (780M) with Oculink: https://www.amazon.com/dp/B0FK2BLV7J?tag=hakehardware-20 - GMKtec X1 (890M) with Oculink: https://www.amazon.com/dp/B0DS5XLRVQ?tag=hakehardware-20 My main system: - ASUS RTX 5060 Ti 16GB: https://www.amazon.com/dp/B0F7WB6LSH?tag=hakehardware-20 - CORSAIR Vengeance DDR5 64GB: https://www.amazon.com/dp/B09WN2H42Y?tag=hakehardware-20 - Samsung 990 EVO Plus 4TB: https://www.amazon.com/dp/B0DHLBDSP7?tag=hakehardware-20 - GIGABYTE Z790 AORUS Elite AX: https://www.amazon.com/dp/B0BH9DXY38?tag=hakehardware-20 0:00 - Intro and System Overview 0:39 - What is Ollamometer 0:51 - DeepSeek vs GPT-OSS Models 1:16 - Benchmark Methodology 1:58 - Starting the Benchmark 2:53 - Loading Different Model Sizes 4:04 - Benchmark Results and Token Speeds 5:00 - Memory Usage by Model Size 5:36 - Time to First Token Analysis 6:06 - Real-World Usability Assessment 7:02 - Comparing Benchmark Results 7:46 - Upcoming Tests with More Hardware 8:40 - Strix Halo Cooling Tips 9:25 - Wrap Up Related Videos: - Local AI with a Mini PC is Totally Usable: https://youtu.be/ve4pR7OrEsQ - Strix Halo + ROCm 7.1 + Ubuntu: https://youtu.be/vX7cYahvcPI - Ollama AMD GPU on Windows — Custom Build (680M/780M/890M): https://youtu.be/AgLL-mmlG9Y - Testing Qwen3-coder-30b with a Mini-PC: https://youtu.be/y7IHTVa8lL8 #strixhalo #localai #ollama #llmbenchmark #hakehardware