AMD shared benchmarks for its RTX Spark AI accelerator, positioned as a competitor to Nvidia's offerings in data center AI compute.

AMD has announced AI performance results for its new Gorgon Halo Ryzen AI Max+ Pro 495 CPU, highlighting the processor's distinctive 192GB memory capacity—50% more than a single Nvidia RTX Spark or AMD Strix Halo chip. According to AMD, this memory advantage enables machines equipped with the CPU to run 300-billion-parameter AI models.

The announcement arrives with deliberate timing, just days before Nvidia introduces the RTX Spark AI chip at Microsoft's October 7 event. The Ryzen AI Max+ Pro 400 series largely represents a refresh of the previous Strix Halo Ryzen AI Max+ Pro 300 series, with enhanced RAM addressability and a 100MHz higher clock speed, positioning it as AMD's direct response to the RTX Spark.

Since the RTX Spark has yet to demonstrate its full capabilities, AMD instead benchmarked against an Intel Core Ultra X9 388H—a somewhat disadvantageous comparison given the Intel chip's consumer market positioning and its 64GB maximum memory ceiling.

AMD claimed its Ryzen AI Max+ Pro 495 delivered between 1.1x and 32.2x faster AI performance than the Intel processor. The company reported up to 20 tokens per second on GLM 5.3 Flash, a 320-billion-parameter open-weight model regarded as among the best available (though only 18 billion parameters are active per token). AMD also achieved up to 42 tokens per second on Qwen 3.8 Flash, a 125-billion-parameter model with 5 billion active parameters per token.

Beyond supporting larger models, the substantial memory capacity enables running multiple models simultaneously—for instance, deploying a small, fast model for routine tasks while keeping a larger model in reserve for complex queries.

While the 192GB memory pool represents an advantage over the RTX Spark's 128GB, memory alone does not determine AI performance. Compute speed and memory bandwidth are equally critical. Higher GPU compute performance (FLOPS/TOPS) enables an AI system to process larger prompts faster, reducing latency before response generation. It also accelerates image generation by reducing rendering time. However, once text generation begins, speed becomes constrained by memory bandwidth.

The Ryzen AI Max+ Pro 495 pairs its memory advantage with a modest bandwidth improvement: 273GB/s, a 6.5% increase from Strix Halo's 256GB/s. Otherwise, the processor largely mirrors its predecessor, the Ryzen AI Max+ Pro 395, featuring 16 Zen 5 CPU cores and 40 RDNA 3.5 GPU compute units, with a 100MHz boost to CPU and GPU clock frequencies.

Gorgon Halo's strength lies in handling large AI models, with AMD claiming support for models up to 300 billion parameters. Whether this translates to an advantage over the RTX Spark remains uncertain. According to Tom's Hardware, the prior-generation Strix Halo achieved 56 tokens per second on GPT-OSS 120B with 4-bit quantization, compared to 64 tokens per second for Nvidia's DGX Spark. Given the DGX Spark's similar architecture to the forthcoming RTX Spark, the new Gorgon Halo may struggle to match Nvidia's performance on some AI tasks.

Gorgon Halo-equipped systems, such as the Minisforum MS-S1 Max-P495, will carry a $7,399 price tag. While steep, this reflects the substantial on-board memory. For portable options, AMD partners will release Ryzen AI Max+ Pro 400-equipped laptops, including the HP ZBook Ultra G3a at $7,449.