AMD is using its new Ryzen AI Max+ PRO 495 “Gorgon Halo” systems to make an early case for high-memory local AI computing just as Nvidia prepares to bring RTX Spark PCs to market. The timing makes the competitive intent clear, but AMD’s new benchmarks are not a direct Gorgon Halo-versus-RTX Spark showdown. Instead, AMD is comparing its flagship 192GB Ryzen AI Halo system against Intel’s Core Ultra X9 388H platform.
That distinction is important.
AMD clearly wants Gorgon Halo positioned as an alternative to the incoming generation of high-end local AI PCs, including Nvidia’s RTX Spark platform. But the benchmark data AMD has published so far primarily tells us how the Ryzen AI Max+ PRO 495 performs against Intel’s latest Panther Lake-based Core Ultra platform under selected generative-AI workloads.
RTX Spark remains the larger competitive question — and one that will require independent testing once systems are broadly available.
AMD Gets Its Local AI Message Out Before RTX Spark Arrives
The timing of AMD’s benchmark release is difficult to ignore.
Nvidia has already announced RTX Spark as a new class of Windows PC built around local AI agents, creator workloads, and gaming. Systems are expected to begin arriving in October, putting Nvidia directly into the same premium local-compute market AMD is targeting with Gorgon Halo.
AMD, meanwhile, already has Ryzen AI Max+ PRO 495 systems entering the market.
That gives AMD an opportunity to establish a performance and memory-capacity narrative before RTX Spark systems become widely available for independent comparison.
However, AMD is not yet publishing a direct benchmark against RTX Spark. The company’s current comparison pits the Ryzen AI Max+ PRO 495 against Intel’s Core Ultra X9 388H.
That makes the data useful, but it should not be interpreted as proof that Gorgon Halo is faster than RTX Spark.
What AMD Actually Benchmarked
AMD’s published results focus on local generative-AI workloads running through ComfyUI.
The company tested a Ryzen AI Max+ PRO 495 system with 192GB of memory, with 128GB allocated as variable graphics memory, against an Intel Core Ultra X9 388H system equipped with 64GB of memory.
According to AMD, the Gorgon Halo system posted performance advantages across workloads including image generation, video generation, music generation, and 3D content creation.
Some of the published results include:
- Z Image Turbo INT8: up to 2.2x faster
- Krea 2 Turbo INT8: up to 2.3x faster
- Ideogram V4 INT8: up to 1.9x faster
- Flux 2 Klein 9B: up to 1.3x faster
- FastVideo H3: up to 2.1x faster
- LTX 2.5: up to 3.1x faster
- Ace Step XL 1.5 Turbo: up to 10.8x faster
- Stable Music 3: up to 3.4x faster
- Microsoft Trellis 2: up to 1.3x faster
AMD’s wider comparison reportedly ranges from approximately 1.1x to 32.2x depending on the workload.
The largest figures deserve context. Some workloads did not run on the Intel comparison system, and extreme outliers can reflect memory limits, software support, or optimization differences rather than a straightforward difference in raw silicon performance.
That is especially relevant here because AMD’s test platform carries three times as much system memory as the Intel configuration.
Memory Capacity Is Part of Gorgon Halo’s Real Advantage
For local AI workloads, memory capacity can matter as much as compute throughput.
The Ryzen AI Max+ PRO 495 supports configurations with up to 192GB of unified LPDDR5X memory, giving developers room to run much larger models locally than would typically fit on a conventional consumer GPU.
That makes Gorgon Halo particularly interesting for users working with local large language models, image-generation pipelines, multimodal systems, and AI agents that need large context windows or multiple models loaded simultaneously.
AMD’s flagship chip combines 16 Zen 5 CPU cores and 32 threads with Radeon 8065S graphics and an XDNA 2 NPU. AMD lists a configurable power range of 45W to 120W for the Ryzen AI Max+ PRO 495, depending on the system design.
The broader platform pitch is therefore not simply “AMD versus Nvidia graphics.”
It is increasingly about how much AI capability, memory, software support, and sustained compute can be packaged into a compact Windows or Linux system.
RTX Spark Changes the Competitive Picture
Nvidia’s RTX Spark platform approaches the same market from a different direction.
RTX Spark combines a Grace CPU with Blackwell RTX graphics in a single superchip designed for local AI agents, creator applications, and gaming.
Nvidia says higher-end configurations can include:
- Up to a 20-core Grace CPU
- Up to a 6,144-core Blackwell RTX GPU
- Up to 128GB of unified LPDDR5X memory
- Up to 1 petaflop of FP4 AI performance
- Fourth-generation ray-tracing cores
- Fifth-generation Tensor cores
- CUDA and the broader RTX software ecosystem
Nvidia is also positioning RTX Spark as a platform for persistent local AI agents running directly on a Windows PC.
That makes the eventual comparison with Gorgon Halo more interesting than a traditional CPU or GPU benchmark.
AMD brings more maximum memory capacity in current 192GB configurations, while Nvidia brings CUDA, TensorRT, RTX, and a software ecosystem that remains deeply entrenched across AI development and creative workloads.
This Is Really a Local AI Platform Battle
The most important comparison between Gorgon Halo and RTX Spark may not be a single benchmark chart.
For local AI users, several factors are likely to matter more:
- Maximum usable memory: Larger local models often depend first on whether they fit in memory at all.
- Inference performance: Tokens per second and generation throughput matter once the model is loaded.
- Software compatibility: CUDA remains a major advantage for Nvidia, while AMD continues expanding ROCm and Windows AI support.
- Power efficiency: These systems are intended to deliver workstation-class capabilities in compact form factors.
- Model support: Framework compatibility can determine whether theoretical hardware performance is usable in practice.
- Price: Early Gorgon Halo systems equipped with 192GB of memory are appearing well above conventional high-end desktop pricing.
That last point may become especially important.
Some of the first Ryzen AI Max+ PRO 495 systems are priced in the $6,000 to $7,000-plus range when configured with 192GB of memory and large storage capacities.
At that level, these machines are no longer casual enthusiast PCs. They begin competing with specialized workstations, developer systems, and small AI servers.
AMD’s Benchmark Numbers Need Context
AMD’s benchmark results are still meaningful.
They show that Gorgon Halo can perform strongly in memory-heavy generative-AI workloads, and the 192GB unified-memory configuration gives AMD a compelling specification for users trying to run large models locally.
But the comparison also highlights why AI benchmarking is becoming more complicated.
A system with 192GB of memory may complete a workload that a 64GB system cannot run efficiently — or cannot run at all. That is valuable to the buyer, but it does not necessarily mean the processor itself is 10x or 30x faster.
Memory capacity, framework optimization, precision mode, model quantization, driver maturity, and backend support can all dramatically affect the final result.
For that reason, AMD’s published figures should be treated as vendor benchmarks rather than a definitive ranking of local AI platforms.
ITD Insight
AMD’s real advantage may be less about claiming an early benchmark victory and more about getting 192GB local-AI systems into the market before RTX Spark becomes widely available. The numbers AMD has published are against Intel, not Nvidia, so the real Gorgon Halo-versus-RTX Spark fight has not happened yet. Once both platforms can be tested with the same models, memory allocations, software stacks, and power targets, we should get a much clearer picture of whether AMD’s capacity advantage can offset Nvidia’s CUDA and RTX ecosystem.
What Buyers Should Watch
For someone shopping for one of these systems, I would not make a decision based on AMD’s current benchmark slides alone.
A developer running large local language models may find the 192GB Gorgon Halo configurations extremely attractive simply because more models can remain entirely in local memory.
A user heavily dependent on CUDA, TensorRT, Nvidia-optimized AI frameworks, or RTX creator applications may find RTX Spark much easier to integrate into an existing workflow.
Homelab users should also pay close attention to networking, virtualization support, Linux compatibility, idle power consumption, memory allocation behavior, and whether these platforms can realistically replace a discrete-GPU workstation or small AI server.
Those details often matter more over several years of ownership than a launch-day generative-AI benchmark.
Bottom Line
AMD’s latest Gorgon Halo benchmarks are an effective reminder that Nvidia will not enter the premium local-AI PC market uncontested.
The Ryzen AI Max+ PRO 495 combines a powerful Zen 5 CPU, integrated Radeon graphics, and up to 192GB of unified memory in a platform specifically suited to large local AI workloads.
But AMD has not yet demonstrated that Gorgon Halo beats RTX Spark.
The benchmarks currently available compare Gorgon Halo primarily against Intel’s Core Ultra X9 388H, while Nvidia’s RTX Spark systems are only beginning to reach the market.
The real comparison will start once reviewers can run the same models and applications across both platforms under consistent conditions.
At that point, the question will not simply be which chip produces the biggest benchmark number. It will be whether AMD’s larger memory ceiling and increasingly capable local-AI stack can compete with Nvidia’s Blackwell hardware, CUDA ecosystem, and deeply established AI software support.
Sources: AMD Ryzen AI Halo, AMD Ryzen AI Max+ PRO 495 specifications, Tom’s Hardware, and Nvidia RTX Spark.