Key Takeaways
1. Nvidia has launched the DGX Spark, a compact AI workstation powered by the GB10 Superchip, designed for researchers and developers working with generative models.
2. The DGX Spark is priced at $4,000, which may discourage some potential buyers due to its high cost.
3. GMKtec’s EVO-X2 mini PC competes with the DGX Spark and has reportedly outperformed it in benchmark tests with various large language models.
4. The EVO-X2 offers better low-latency performance for real-time tasks, thanks to its AMD Ryzen AI Max+ chip and lower price of $2,199.
5. While Nvidia’s DGX Spark excels in high-throughput scenarios, the AMD Strix Halo provides better value for developers focused on efficient on-device AI and token workloads.
Nvidia has introduced the DGX Spark, marking its initial venture into compact AI systems. This workstation, about the size of a desktop, is driven by the company’s GB10 Superchip and can reach up to 1 PFLOP when operating at FP4. Nvidia brands it as a “personal AI supercomputer,” aimed at assisting researchers and developers engaged with extensive generative models.
Price Tag Dilemma
However, the price may deter some potential buyers. The DGX Spark comes with a hefty price of $4,000, leading many enthusiasts to think twice about its cost-effectiveness.
The Competition Emerges
In response, GMKtec, known for its high-performance mini PCs, has brought its own EVO-X2 into the ring, pitting it against the Nvidia DGX Spark and sharing the outcomes. The EVO-X2 is powered by AMD’s Ryzen AI Max+ 395 Strix Halo APU.
GMKtec claims that the Strix Halo outperformed the DGX Spark in benchmark tests with various open-source large language models like Llama 3.3 70B, Qwen3 Coder, GPT-OSS 20B, and Qwen3 0.6B. The AMD machine demonstrated superior performance in token generation rates and initial response times.
Performance Insights
According to GMKtec, the EVO-X2 mini PC shined in low-latency performance during real-time inference tasks, thanks to the Ryzen chip’s CPU + GPU + NPU configuration and the XDNA 2 AI engine. In contrast, Nvidia’s supercomputer focused more on sheer throughput.
The highest-end model of the EVO-X2 is priced at $2,199, which is about half the cost of the DGX Spark. Although GMKtec’s findings indicate that Nvidia’s setup is still the go-to hardware for large-model, high-throughput scenarios, AMD’s Strix Halo appears to provide better value for AI developers and enthusiasts dealing with on-device AI and efficient token workloads.
Source:
Link
