By GraphicCardPrices. Verified . This is a launch explainer, not a hands-on review.
NVIDIA has announced availability for RTX Spark laptops, with configurations offering up to 128GB of unified memory. For someone shopping for local AI hardware, that raises a useful question: should you buy a new computer or upgrade the graphics card in the desktop you already own? NVIDIA’s launch announcement
Start with the model and workload you want to run. Then compare the usable memory, software support and total cost of each option. The biggest memory number alone cannot tell you which purchase makes sense.
What is launching?
RTX Spark combines an NVIDIA Grace CPU and Blackwell RTX GPU in Windows laptops and compact desktops. The announced laptop date is ; compact desktops are scheduled for . Microsoft separately confirms the date for Surface Laptop Ultra. These are announced dates, not a guarantee that every configuration will be in stock. Microsoft’s announcement
NVIDIA’s update also describes launch software for local AI and development. Check the particular app and version you intend to use instead of assuming that a platform announcement confirms your whole workflow. October 13 software update
128GB unified memory is not 128GB dedicated VRAM
A desktop graphics card has its own VRAM. RTX Spark’s CPU and integrated GPU share physical memory, but NVIDIA documents three logical regions: a GPU-reserved region, shared CPU/GPU memory and CPU-only memory. The GPU-reserved region also comes from system DRAM. NVIDIA’s memory architecture guide
That means the advertised system capacity is not a promise that an equally large model will fit. Windows sets a GPU memory budget, other applications consume memory, and NVIDIA warns against allocating the entire available budget. CUDA memory budgeting documentation
Model weights are only part of the requirement. Longer context also increases memory use. Use our VRAM estimator as a planning aid, then verify your chosen model, quantization and context length in the intended runtime. Ollama’s context-length guidance
Compare the money you actually need to spend
For a desktop upgrade, add the graphics card price to any necessary power-supply, cooling or case changes. Confirm that the card fits and that the power connections are suitable.
For an RTX Spark purchase, price the exact complete computer, including its memory and storage configuration. “Up to 128GB” describes a maximum, not the specification of every machine.
Our GPU price listings cover graphics cards, not complete PCs. Dividing a laptop price by its unified-memory capacity and comparing that figure with a graphics card’s price per GB would mix different purchases.
Which option belongs on your shortlist?
- Already have a suitable desktop? Start by checking whether a compatible card can meet your workload’s measured memory needs.
- Need a new laptop anyway? An RTX Spark configuration is worth evaluating alongside the cost of replacing your current computer.
- Need more memory than the cards you are considering provide? Investigate unified-memory systems, while checking the actual GPU budget and software compatibility.
- Need a speed comparison? Look for tests using the same model, quantization, context and runtime. Memory capacity alone does not establish tokens per second.
The practical next step is to choose one representative workload before choosing hardware. Our local AI GPU buying guide helps narrow desktop-card options; use the resulting shortlist to compare an upgrade with the exact RTX Spark system you would buy.