AMD’s Strix Halo Mini PC: The Affordable Alternative to NVIDIA’s $4,000 DGX Spark Supercomputer

NVIDIA made waves in the tech industry with its release of the DGX Spark, a compact system crafted specifically for artificial intelligence (AI) workloads. Concurrently, AMD has been making significant strides with its APU series, particularly the Strix Halo APU, which reportedly outmatches NVIDIA’s GB10 chipset in various AI performance metrics. This emerging competition raises questions about efficiency and value in the high-performance computing space.

NVIDIA’s DGX Spark vs. AMD’s Strix Halo: Price-to-Performance Insights

The DGX Spark stands out as NVIDIA’s inaugural offering in the realm of compact systems aimed at AI applications, featuring the cutting-edge GB10 custom chip. Despite its state-of-the-art capabilities, many potential consumers have expressed concerns over its hefty price tag, estimated at around $4, 000, which significantly limits its appeal. In contrast, GMKtec, a notable manufacturer of mini-PCs, presents a compelling alternative: the EVO-X2, equipped with AMD’s Strix Halo APU, available for nearly half that price.

A detailed exploded view of NVIDIA's GB10 Superchip showcasing components like the Blackwell GPU with '1 PFLOP FP4 AI Compute, ' Grace CPU featuring '20 Arm Cores, ' high-bandwidth unified memory with '128 GB Low-Power DDR5X, ' and connectivity interfaces including Wi-Fi, Bluetooth, and USB. — DGX Spark | Image Credits: NVIDIA

In a recent blog post, GMKtec put the DGX Spark to the test against their EVO-X2 mini-PC. This comparison highlighted the Strix Halo APU’s ability to outperform the NVIDIA solution in several key areas, such as token generation speeds and response times. The tests employed a variety of open-source models, including Llama 3.3 70B, Qwen3 Coder, GPT-OSS 20B, and Qwen3 0.6B, yielding impressive results:

Test Model	Metric	EVO – X2	NVIDIA GB10	Winner
Call 3.3 70B	Generation Speed (tok/sec)	4.9	4.67	AMD
First Token Response Time (s)	0.86	0.53	NVIDIA
Qwen3 Coder	Generation Speed (tok/sec)	35.13	38.03	NVIDIA
First Token Response Time (s)	0.13	0.42	AMD
GPT-OSS 20B	Generation Speed (tok/sec)	64.69	60.33	AMD
First Token Response Time (s)	0.19	0.44	AMD
Qwen3 0.6B Model	Generation Speed (tok/sec)	163.78	174.29	NVIDIA
First Token Response Time (s)	0.02	0.03	AMD

According to GMKtec’s evaluations, the Ryzen Al Max+ 395 processor found in the Strix Halo APU excels with broader parameter models, showing a distinct advantage in first token response times due to the effective integration of CPU, GPU, and NPU architectures. The XDNA 2 engine enhances AI processing, yielding lower latency in outputs.

Conversely, NVIDIA’s strengths emerge in scenarios favoring throughput rather than memory latency. The DGX Spark is particularly well-suited for high-throughput configurations involving large models, offering impressive performance thanks to the GB10 Superchip’s capability of achieving PFLOPS at FP4. However, for applications emphasizing low-latency response—a crucial aspect for real-time inference workloads—the AMD platform presents a comparable alternative at a significantly reduced cost.

An AMD Ryzen AI MAX Series chip with visible branding is displayed in front of a geometric, glowing background. — Image Credits: AMD

Further solidifying this perspective, GMKtec’s EVO-X2 mini-PC is priced at $2, 199 for a top-end configuration (128 GB RAM and 2 TB storage), contrasted with the $4, 000 price tag of the DGX Spark—making the cost-to-performance ratio between the Strix Halo and GB10 markedly appealing. For businesses looking to implement localized AI models without straining their budgets, the EVO-X2 emerges as a viable, budget-friendly workstation option.

Source&Images