NVIDIA DGX Spark and NVIDIA RTX Spark share a name, a Grace-plus-Blackwell design and a 128 GB unified-memory ceiling, but they are built for different people. DGX Spark is a Linux desk system on the GB10 superchip, designed for AI developers who want NVIDIA's data-centre software stack, continuous operation and the option to cluster. RTX Spark is a Windows 11 platform on the N1X superchip, mostly sold as laptops, designed for people who want large local AI models inside an everyday Windows PC. Developers building for NVIDIA infrastructure will usually prefer DGX Spark; Windows users who need AI on the machine they already work on will find RTX Spark more practical. At SourceIT the NVIDIA DGX Spark is S$8,800 and the RTX Spark MSI Prestige N16 Flip AI+ runs from S$4,974.08 to S$10,854.12 (listed 28 Oct 2026).
DGX Spark vs RTX Spark at a glance
| Specification | NVIDIA DGX Spark (GB10) | NVIDIA RTX Spark (N1X) |
|---|---|---|
| Form factor | 150 x 150 x 50.5 mm desk system | Laptops at launch; small desktops such as MSI's EdgeMesa N AI+ announced |
| Superchip | GB10: 20-core Arm CPU + Blackwell GPU | N1X: 20-core (or 18-core) Grace CPU + 6,144-core (or 5,120-core) Blackwell RTX GPU + NPU |
| Unified memory | 128 GB LPDDR5X, 273 GB/s | Up to 128 GB LPDDR5X, 256-bit; bandwidth not published by NVIDIA |
| AI compute | Up to 1 PFLOP FP4 (with sparsity) | Up to 1 PFLOPS FP4 (6,144-core part); 827 TFLOPS (5,120-core part) |
| Power | 240 W external supply; GB10 rated 140 W | 45 to 80 W in laptops; 140 W desktop configuration |
| Operating system | NVIDIA DGX OS (Ubuntu-based Linux) | Windows 11 on Arm |
| NVIDIA model-size claim | "AI models up to 200 billion parameters" | "Run models up to 120B parameters with 1M context" |
| Clustering | ConnectX-7, 2x 200 GbE QSFP; 2 to 4 units | None published |
| SourceIT price | S$8,800 (4 TB) | S$4,974.08 (32 GB) to S$10,854.12 (128 GB), MSI Prestige N16 Flip AI+ |
From NVIDIA's DGX Spark and RTX Spark product pages and MSI's EdgeMesa N AI+ page, checked October 2026. Reviewers estimate RTX Spark bandwidth at about 300 GB/s from its LPDDR5X-9400 memory (ServeTheHome, StorageReview, October 2026); NVIDIA has not confirmed a figure.
Linux or Windows: why does the operating system matter so much?
Because the AI software world was built on Linux. The serving tools teams use in production, such as vLLM, SGLang and TensorRT-LLM, along with Docker GPU containers and multi-node tools, target Linux first. DGX Spark runs DGX OS, the same software family as NVIDIA's DGX servers, so a model served on your desk behaves like the one you will deploy later.
RTX Spark brings much of that stack to Windows for the first time. NVIDIA ships native CUDA, PyTorch, Python and Conda for Windows on Arm and says 97% of the top 15,000 Python packages are supported. But some tools still assume Linux: vLLM's own documentation says it "does not support Windows natively" and points to WSL, and TensorRT-LLM's installation guide covers Linux only. On RTX Spark you get Windows plus a Linux subsystem; on DGX Spark you get Linux natively. Our follow-up, Is Windows Finally Ready to Challenge Linux for Local AI?, goes framework by framework.
Is the unified memory really the same?
The ceiling is the same; how you use it differs. On DGX Spark, the GPU addresses the 128 GB pool directly and the system is dedicated to AI. On RTX Spark, Windows and your applications share the same memory, and reviewers found the GPU allocation capped by default: Notebookcheck saw 64 GB available to the GPU on a 128 GB ASUS ProArt P16, and StorageReview found a 64/64 split on a 128 GB Surface Laptop Ultra. The split can be adjusted, for example in MyASUS, but it means a 128 GB RTX Spark laptop is not automatically a 128 GB AI machine out of the box. For a model of 60 GB or more, check and set the allocation first.
Same memory ceiling, different job: DGX Spark gives all of it to AI; RTX Spark shares it with your working day.
Which models can each run?
NVIDIA's own claims differ: up to 200 billion parameters for DGX Spark, up to 120 billion for RTX Spark. Both assume heavily compressed weights, and NVIDIA's DGX Spark datasheet ties its figure to FP4 support. In practice both run the popular open models in the 20B to 120B range, including GPT-OSS-120B, which NVIDIA and llama.cpp have benchmarked on DGX Spark and which Techtesters ran on the MSI Prestige N16 Flip AI+. DGX Spark adds two things RTX Spark cannot: it pools memory with a second unit for models up to about 400B, and it is rated by NVIDIA for fine-tuning models up to 70B. What the 120B and 200B figures actually mean is explained in What NVIDIA's 120B and 200B AI Model Claims Really Mean.
Who is each one for?
- DGX Spark suits AI developers, data scientists and research groups who work on Linux, run models for hours at a time, share one model server across a team, fine-tune, or may cluster two units later.
- RTX Spark suits Windows users who want large models on the computer they already use: consultants and presales staff demonstrating AI on site, creators using Adobe tools with AI features, developers in Windows-only organisations, and students or lecturers who need one machine for everything.
- Neither suits a whole department hitting one endpoint at once. That is a GB300-class job.
If you game as well, RTX Spark is the only one of the two that can; see Gaming, Coding or AI: Which One Actually Makes Sense?
What about the price gap?
At SourceIT on 28 Oct 2026, the NVIDIA DGX Spark 4 TB is S$8,800 and the ASUS Ascent GX10, which uses the same GB10 chip, starts at S$8,100 with 1 TB. The RTX Spark MSI Prestige N16 Flip AI+ 128 GB is S$10,854.12, with the 64 GB version at S$6,331.28 and the 32 GB at S$4,974.08. The laptop premium buys a 16-inch OLED touchscreen, battery, keyboard and Windows 11 Pro. If you need a dedicated AI box and already have a laptop, DGX Spark is the better value per gigabyte of AI memory. If the AI has to live inside your everyday computer, RTX Spark saves buying two machines.
RTX Spark desktops are also arriving. MSI's EdgeMesa N AI+ lists the full 6,144-core N1X, up to 128 GB of memory, a 140 W rating, Windows 11 Pro and 10 GbE, but no ConnectX clustering port; it is not yet listed at SourceIT, so ask us about availability.
Buying in Singapore
SourceIT supplies the NVIDIA DGX Spark, the ASUS Ascent GX10 and other GB10 systems, and all three MSI Prestige N16 Flip AI+ RTX Spark configurations, with local warranty and a GST invoice. Browse the NVIDIA DGX Spark collection, the DGX Spark Singapore page and MSI NVIDIA RTX Spark AI PCs, or read the RTX Spark laptop guide and RTX Spark laptop vs DGX Spark vs ASUS GX10. For a quotation, use Request for Quotation on the product page, call (65) 6978 3502, email Sales@sourceit.com.sg or use our contact page.
Frequently asked questions
What is the difference between DGX Spark and RTX Spark? DGX Spark is a Linux desk system on NVIDIA's GB10 superchip with ConnectX-7 clustering. RTX Spark is a Windows 11 on Arm platform on the N1X superchip, sold mainly as laptops. Both offer up to 128 GB of unified memory.
Is RTX Spark a DGX Spark in a laptop? No. It uses a different chip, the N1X, at 45 to 80 W in laptops, runs Windows rather than DGX OS, and has no published clustering capability.
Why would a developer prefer DGX Spark? It runs NVIDIA's Linux AI stack natively, including tools such as vLLM and TensorRT-LLM that are Linux-first, runs continuously at desktop power, and can cluster with a second unit.
Can a 128 GB RTX Spark laptop use all its memory for AI? Not by default. Reviewers found the GPU allocation capped at about 64 GB on 128 GB models, adjustable in vendor software. Windows and applications share the same memory.
How much do they cost in Singapore? At SourceIT on 28 Oct 2026, the NVIDIA DGX Spark is S$8,800 and the RTX Spark MSI Prestige N16 Flip AI+ ranges from S$4,974.08 to S$10,854.12.
SourceIT Approved
Researched and verified by the SourceIT team for Singapore buyers.
