AMD Rushes Out New Benchmarks to Compete with Nvidia’s RTX Spark

AMD has released performance data for its new Gorgon Halo AI processors just days before Nvidia is expected to announce its RTX Spark lineup. The timing is strategic: as the AI PC market heats up, AMD wants shoppers to understand what its latest high-end chips can deliver. The first Gorgon Halo devices are already available, with top-tier models starting around $7,000.

Gorgon Halo represents a refresh of AMD’s previous-generation Strix Halo platform, designed specifically for running large language models and generative AI applications locally on your device. While the core architecture remains largely unchanged, AMD has made two meaningful upgrades: boosting unified memory capacity to 192GB (up from 128GB) and increasing CPU clock speeds by 100 MHz on the flagship Ryzen AI Max+ Pro 495.

Performance Gains in AI Model Execution

professional workstation setup AI model
Photo by Shahram Anhari

AMD’s benchmarking focused on generative AI workloads using ComfyUI software across multiple popular models. The company compared its top-spec Ryzen AI Max+ Pro 495 against Intel’s Core Ultra X9 388H, claiming performance advantages ranging from 1.1x to 32.2x depending on the model tested. However, the most dramatic results appear to be statistical outliers, so realistic performance gains fall in a more modest range for everyday use.

When running the GLM 5.3 Flash model with 320 billion parameters, AMD achieved peak throughput of 20 tokens per second using mixed-precision quantization techniques. The Qwen 3.8 Flash Next model, a multimodal AI system, reached up to 42 tokens per second. These figures demonstrate solid capability, though performance naturally degrades as context length increases, which is critical for real-world applications where long conversations or document analysis occur.

Memory Capacity: A Key Advantage Over RTX Spark

The biggest differentiator between Gorgon Halo and the incoming RTX Spark devices is memory capacity. AMD’s new chips support up to 192GB of unified memory, while Nvidia’s RTX Spark tops out at 128GB. Higher memory capacity matters because it lets you run larger AI models entirely on your local device without cloud dependencies, though there’s a tradeoff: larger models typically run slower than smaller, optimized versions.

This memory advantage doesn’t necessarily translate to faster performance. Earlier testing of the previous Ryzen AI Max+ 395 showed it consistently underperformed compared to Nvidia’s DGX Spark across both response time and token throughput metrics. The extra memory allows broader compatibility with different models, not necessarily better speed.

Market Context: Agentic PCs Are Still Niche

AMD claims to have shipped over half a million agentic PCs to date, a category that includes both Strix and Gorgon Halo devices. While this sounds impressive, it underscores how limited this market segment remains compared to mainstream consumer laptops. Initially, AMD had claimed shipping tens of millions of AI PCs before clarifying that most of those were traditional AI-accelerated devices, not the specialized agentic category.

The competition between Gorgon Halo and RTX Spark represents a narrower battle within the broader AI PC trend. Both platforms target power users and professionals who need to run sophisticated models locally, not casual consumers shopping for better laptop performance.

Pricing and Availability

performance benchmark testing metrics
Photo by Stephen Dawson

The Minisforum MS-S1 Max-P495, one of the first Gorgon Halo devices on sale, carries a starting price around $7,000 for top configurations. This entry-level positioning for Gorgon Halo is steep, and additional device options from other manufacturers will likely push prices higher. These aren’t machines for budget-conscious buyers.

What This Means for Shoppers

If you’re considering a high-end AI PC for running large language models locally, Gorgon Halo offers proven performance and maximum memory flexibility. The extra 64GB of memory beyond RTX Spark’s maximum is valuable for specific use cases involving extremely large models. However, speed-focused buyers should recognize that AMD’s processors historically trail Nvidia’s offerings in token throughput and response time.

The timing of AMD’s benchmark release suggests the company is serious about competing with RTX Spark, even if Gorgon Halo is a relatively modest generation-over-generation refresh. For shoppers, the real deciding factors will be specific model compatibility needs, available device options, pricing across the full range of configurations, and software optimization for your intended AI applications. Wait for RTX Spark’s official launch details and independent testing before making a final decision.