Full Breakdown
Nvidia DGX Spark Faces Performance Criticism from John Carmack
10/28/2025, 11:25:39 PM
Overview of Performance Issues
Nvidia's DGX Spark, a $4,000 mini PC platform powered by the Grace Blackwell GB10 superchip, has come under scrutiny for not meeting its advertised performance metrics. John Carmack, the former CTO of Oculus VR and founder of Keen Technologies, has raised significant concerns regarding the system's power draw and computational capabilities. Carmack's tests indicate that the DGX Spark maxes out at approximately 100 watts, which is less than half of its claimed 240-watt rating. He also reported that the system delivers around 60 TeraFLOPS of BF16 performance, significantly below expectations.
Technical Specifications and Limitations
The DGX Spark is designed to support large AI models with specifications that include a 20-core Arm-based Nvidia Grace CPU and a Blackwell GPU. Nvidia claims the system can achieve one petaflop of AI performance, but this figure relies on structured sparsity, a technique that optimizes performance by skipping zero-value computations. Carmack and other developers have pointed out that when this feature is disabled, the system's performance drops to about half of the advertised figures.
Independent tests corroborate Carmack's findings, with reports indicating that the DGX Spark struggles to reach its full power ceiling under various workloads. Users have also reported issues such as spontaneous reboots and thermal throttling, raising questions about the system's reliability during extended use.
Criticism from the Tech Community
Carmack's criticisms have resonated within the tech community, prompting responses from other developers. Awni Hannun, an AI researcher at Apple, echoed Carmack's sentiments, noting that his benchmarks also fell short of expectations. The growing chorus of dissatisfaction suggests that Nvidia may need to reassess its marketing claims and address the underlying issues affecting the DGX Spark's performance.
Official Responses and Future Considerations
As of now, Nvidia has not publicly addressed the concerns raised by Carmack or the user-reported issues. The lack of communication from the company has left many users and developers seeking clarity on whether the performance discrepancies are due to firmware limitations, thermal constraints, or other factors.
The DGX Spark's compact design, while appealing for its size, may be contributing to its thermal challenges. Some users have suggested that a larger chassis could improve cooling and performance, indicating a potential area for future product iterations.
Verbatim Quotes
- “DGX Spark appears to be maxing out at only 100 watts power draw, less than half of the rated 240 watts, and it only seems to be delivering about half the quoted performance (assuming 1 PF sparse FP4 = 125 TF dense BF16) .” — John Carmack, Founder of Keen Technologies
- “It gets quite hot even at this level, and I saw a report…” — John Carmack
- “Nvidia’s marketing even suggests it can run 20-billion-parameter models locally, a feat few discrete setups can manage, due to its Blackwell architecture.” — Tom's Hardware
Conclusion
The Nvidia DGX Spark's launch has been overshadowed by performance criticisms, particularly from influential figures like John Carmack. As users continue to report issues, Nvidia faces pressure to clarify its specifications and potentially rectify the performance shortcomings of its flagship AI development platform. The future of the DGX Spark may depend on the company's responsiveness to these concerns and its ability to deliver on its ambitious promises.
