AI-rewritten: This is a summary of an article from Tom’s Hardware, rewritten by AI (Qwen, running locally) to make it easier to read. The facts come from the original article – read it for the full story.

---

Ahead of Nvidia’s expected RTX Spark launch later this week, AMD has shared benchmarks for its new Gorgon Halo chips to highlight their generative AI capabilities. The company is comparing its Ryzen AI Max+ Pro 495 chip against Intel’s Core Ultra X9 388H using the ComfyUI software to measure performance. AMD utilized a top-spec configuration with 192GB of memory, while the Intel system used 64GB. Results showed throughput ranging from 1.1x to 32.2x in favor of AMD, though the highest figure was identified as an outlier that may stem from optimization issues or hardware limitations on the Intel side.

The benchmarks focused on specific AI models rather than gaming performance, which AMD notes will remain similar to last-generation Strix Halo chips. In tests with the GLM 5.3 Flash model containing 320 billion parameters, AMD achieved a peak throughput of 20 tokens per second using a mixed-quantization format. Another test with the Qwen 3.8 Flash Next multimodal model yielded up to 42 tokens per second. However, experts warn that performance likely drops significantly as context length increases, which could make running these large models locally difficult despite the high initial numbers.

AMD states its Gorgon Halo range primarily competes with Nvidia’s RTX Spark devices and Apple’s M-series SoCs, rather than Intel’s Panther Lake chips. While Nvidia’s devices support up to 128GB of unified memory, AMD’s Gorgon Halo supports up to 192GB, allowing for larger local models at a lower speed. The first Gorgon Halo devices are currently available for purchase, with top-line configurations priced around $7,099. AMD previously claimed it had shipped tens of millions of AI PCs but clarified that the actual number was over half a million.