Nvidia DGX Spark delivers half quoted performance for John Carmack
John Carmack reports performance issues with Nvidia’s DGX Spark AI system
John Carmack, ID Software founder and former CTO of Oculus VR, has been testing an Nvidia DGX Spark AI system. So far, he is not impressed by the performance the system has delivered. His system appears to be maxing out at 100 watts, which is below its 240W power rating. Furthermore, the system has achieved 60 TFLOPS of BF16 output, less than half of what Carmack expected.
Carmack also noted that his unit gets hot under his 100W loads and claims he saw a user report of “spontaneous rebooting” from another DGX Spark user. If this is true, Nvidia’s £3,699 (current UK pricing at Novatech) system doesn’t sound as premium as it should be.
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) . It gets quite hot even at this level, and I saw a report…
— John Carmack (@ID_AA_Carmack) October 27, 2025
Awni Hannun, an AI researcher at Apple, reported similar issues when microbenchmarking the DGX Spark. They also claimed that they “expected higher” benchmark results. It is worth noting that Nvidia’s benchmarks give their DGX Spark 1 PetaFLOPS of Sparse FP4 performance. With Sparsity, Nvidia can enable faster performance by skipping zero-value computations, enabling up to a 2x performance increase. Sparsity also lowers memory bandwidth usage and power consumption.
The “up to 2x” performance increase from sparsity does a lot of heavy lifting for Nvidia. Since not all AI workloads are sparse, they do not benefit from Nvidia’s performance boost for sparse workloads. For “dense” (not sparse) workloads, performance is halved. This seems to match the performance that Hannun and Carmack have uncovered.
That is what we are seeing. Seems about right if it is only drawing 100W instead of 200W+.
— John Carmack (@ID_AA_Carmack) October 27, 2025
Based on the thoughts of these tech experts, Nvidia’s DGX Spark system doesn’t appear to be meeting the expectations of users. Perhaps they are not considering Nvidia’s use of “sparsity” data in its performance specifications. Regardless, it’s not a good look when renowned developers are disappointed in your new hardware.
You can join the discussion on Nvidia’s DGX Spark failing to meet performance expectations for users on the OC3D Forums.
