AMD has published benchmark results for its Gorgon Halo processors days before Nvidia is expected to unveil RTX Spark devices. The company is pitching the chips as the memory-capacity choice for local AI workloads, with the top Ryzen AI Max+ Pro 495 supporting up to 192GB of unified memory versus 128GB for RTX Spark. However, the benchmarks AMD shared do not compare against Nvidia; they pit the 495 against Intel's Core Ultra X9 388H.
In ComfyUI tests, AMD reported performance advantages ranging from 1.1x to 32.2x, with the high end an outlier. The company also cited token throughput of up to 20 tokens per second on a 320-billion-parameter GLM model and up to 42 tokens per second on Qwen 3.8 Flash Next, both using aggressive quantization. Tom's Hardware notes that "up to" figures can decline sharply as context length grows, and that the comparison systems were not equivalent: the AMD machine had 192GB of memory, while the Intel system had 64GB.
Tom's Hardware points out that Gorgon Halo is largely a refresh of the previous Strix Halo generation, with identical core counts and microarchitectures, a 100MHz boost-clock increase, and the memory bump. In its own review of the prior Ryzen AI Max+ 395, the outlet found it universally underperformed the DGX Spark in time-to-first-token and tokens-per-second across three models. That suggests the extra memory will let users run larger models locally, but not necessarily faster.
AMD also said it has shipped "over half a million" agentic PCs, after initially claiming "10s of millions" of AI PCs, which the company clarified referred to laptops. First Gorgon Halo devices are on sale now, with top configurations around $7,099. The RTX Spark launch is expected at Microsoft's event on Wednesday, October 7.