MikhbarMIKHBAR
Computing

AMD Shares Gorgon Halo Benchmarks Ahead of RTX Spark

AMD has released fresh AI performance benchmarks for its flagship Gorgon Halo processor, the Ryzen AI Max+ Pro 495, positioning the chip ahead of Nvidia's expected RTX Spark device debut.

AMD Shares Gorgon Halo Benchmarks Ahead of RTX Spark

Gorgon Halo Benchmarks Surface Ahead of Microsoft Event

Ahead of an expected Nvidia RTX Spark launch later this week, AMD has shared performance benchmarks for its flagship Gorgon Halo chip, the Ryzen AI Max+ Pro 495. As reported by [Tom's Hardware](https://www.tomshardware.com/pc-components/cpus/amd-attempts-to-get-ahead-of-expected-rtx-spark-launch-with-gorgon-halo-ai-benchmarks-company-says-it-has-shipped-over-half-a-million-agentic-pcs-to-date), the release follows a tease from Nvidia and arrives just days before a scheduled event from Microsoft.

The extra insight arrives a matter of days after the first Gorgon Halo devices launched, with some systems priced upwards of $7,099. Devices such as the Minisforum MS-S1 Max-P495 are available for sale now, featuring top-line configurations that sit around $7,000, though broader availability is expected to bring a wider range of pricing.

AMD Strix Halo Ryzen AI Max
(Image credit: AMD) · Source: Tom's Hardware

Performance Comparisons Against Intel Core Ultra

Without official performance results for the RTX Spark outside of questionable Geekbench leaks, AMD bypassed Nvidia for its latest comparisons. Instead, the company compared the Ryzen AI Max+ Pro 495 directly against Intel's Core Ultra X9 388H.

AMD utilized ComfyUI to measure generative AI performance, averaging multiple runs across various models to compare total throughput. The testing pitted AMD's top-spec 192GB configuration of the Ryzen AI Max+ Pro 495 against a system running the Core Ultra X9 388H with 64GB of memory.

A hand holding the Ryzen 7 9850X3D.
(Image credit: Tom's Hardware) · Source: Tom's Hardware

Generative AI Throughput and Model Capacities

In benchmark testing involving the GLM 5.3 Flash model featuring 320 billion parameters, AMD recorded a peak throughput of 20 tokens per second utilizing Unsloth's UD-IQ4_XS mixed-quantization format. While GLM 5.3 Flash carries a large total parameter count, only 18 billion parameters are active for each token.

Additionally, AMD detailed performance figures for the Qwen 3.8 Flash Next multimodal mixture-of-experts model. With dynamic 4-bit quantization and multi-token prediction enabled, the Ryzen AI Max+ Pro 495 achieved up to 42 tokens per second.

AMD agentic PC presentation.
(Image credit: AMD) · Source: Tom's Hardware

Unified Memory Advantages and Hardware Context

Gorgon Halo serves largely as a refresh of the previous-generation Strix Halo range, retaining identical core counts and microarchitectures while introducing a 100 MHz boost clock increase for the 495 and expanding unified memory support up to 192GB.

By comparison, RTX Spark devices and similar configurations top out at 128GB of unified memory. While higher memory capacity enables local execution of larger models, reviewers note that increased capacity does not automatically translate to faster processing speeds.

AMD agentic PC presentation.
(Image credit: AMD) · Source: Tom's Hardware

Agentic PC Shipments and Market Position

During a press prebriefing, AMD offered insight into its market presence, initially boasting of shipping tens of millions of AI PCs before clarifying that it has shipped over half a million agentic PCs to date. These figures presumably account for Strix and Gorgon Halo devices as the category expands to rival upcoming hardware drops.

Sources

  • Tom's HardwareAMD attempts to get ahead of expected RTX Spark launch with Gorgon Halo benchmarks

Continue chronologically

Related entity coverage