Nick Lewis’s reported charts show higher token-generation rates for a 64GB Lenovo Yoga RTX Spark than for the DGX Spark in each displayed model comparison. That result applies to the systems and runs in his tests; NVIDIA’s specifications point to a separate choice about operating system, form factor and multi-system deployment.
What the reported runs show
The charts compare Llama 3.1 8B, GPT-OSS 20B and Qwen3 30B-A3B. They include results for Ollama and llama.cpp, along with a separate test labeled for an 8K-token context. The RTX Spark bars are higher in the displayed model comparisons, though the gaps vary. Lewis summarized the overall difference as about 10% more LLM output; that is his summary of these tests, not a result for every RTX Spark configuration or workload.
The practical takeaway is narrow but useful: the tested Lenovo Yoga led the tested DGX Spark in those plotted comparisons. The charts do not establish a general performance ranking across other models, configurations or tasks.
Treat a token-generation lead as a result for the tested workload, then choose a platform for the way you plan to run and scale your AI work.
Windows 11 laptops or a DGX OS desktop?
NVIDIA lists RTX Spark in Windows 11 laptops and desktops. Its laptop specifications give two CPU/GPU configurations, with memory ceilings of up to 128GB or 64GB; its desktop configuration is listed with up to 128GB. Laptop TDP is 45–80W, while the desktop is listed at 140W. NVIDIA says CUDA runs natively on RTX Spark. (NVIDIA RTX Spark specifications)
DGX Spark is a compact desktop system running NVIDIA DGX OS. NVIDIA lists 128GB of coherent unified memory, 273 GB/s of memory bandwidth and a 140W TDP for its GB10 system. Its specifications also describe a preinstalled NVIDIA AI software stack. (NVIDIA DGX Spark specifications)
| Platform detail | RTX Spark | DGX Spark |
| Form factor | Laptops and desktop PCs | Compact desktop system |
| Operating system | Windows 11 | NVIDIA DGX OS |
| Listed memory ceiling | Up to 128GB on top laptop and desktop configurations | 128GB unified memory |
| Listed power | Laptop TDP: 45–80W; desktop TDP: 140W | GB10 TDP: 140W |
| AI software | CUDA runs natively | NVIDIA AI software stack preinstalled |
The operating-system choice is a concrete workflow distinction: RTX Spark is the Windows platform, while DGX Spark is the DGX OS desktop. For background on NVIDIA’s broader home-AI push, see NeoTeo’s earlier coverage of DGX Spark.
DGX Spark’s multi-system option
NVIDIA says ConnectX networking can link up to four DGX Spark systems for models of up to 700 billion parameters in its stated multi-system scenario. That gives DGX Spark a defined scaling path for deployments that need more than one system; NVIDIA’s RTX Spark page positions laptops and desktops but does not describe a matching multi-system configuration.
A separate English-language Linus Tech Tips review examines an ASUS ProArt P16 H7607 with RTX Spark N1X, including AI, gaming, productivity, battery and thermal testing. It is a review of an RTX Spark laptop, not a direct DGX Spark comparison.
