NVIDIA said on Oct. 16 that RTX Spark laptops were available worldwide through ASUS, Dell, HP, Lenovo, Microsoft and MSI.
Shared memory can give local models more room
RTX Spark combines a Blackwell RTX graphics processor and a Grace CPU in one superchip. NVIDIA lists up to 128 GB of unified LPDDR5X memory for the laptop platform. Because the CPU and GPU draw from a shared pool, a high-memory configuration can leave more room for local model weights and the memory used to handle a conversation’s context than a laptop with a smaller graphics-memory limit.
That total is not reserved for AI. Microsoft says the Surface Laptop Ultra’s 128 GB is dynamically shared between the CPU and GPU; the amount addressable by the GPU is lower and depends on configuration and workload. Windows and other applications also need memory. Buyers therefore need to check the exact system configuration, rather than treating “128 GB unified” as 128 GB available to a model.
Performance lacks a reproducible test, and prices differ
NVIDIA’s product page advertises models with up to 120 billion parameters and a one-million-token context, alongside up to one petaflop of FP4 AI performance. FP4 is a four-bit number format used for AI calculations. Microsoft describes the petaflop figure as a theoretical peak using sparsity. The launch materials do not report a repeatable test of a 120-billion-parameter model with its precision, laptop configuration and measured token speed specified, so it should be read as a vendor capability claim, not a demonstrated workload.
There is also a practical buying detail to verify: NVIDIA’s Oct. 16 release lists a $2,599 starting price, while its Oct. 17 launch blog says $2,399. The pages do not explain the difference. NVIDIA says configurations and regional prices vary, so buyers should confirm the model and price with the seller.
Related coverage: NVIDIA’s 24GB Bolt Model Has a Text-Only Memory Budget.
Illustrative image credit: “Laptop on a desk,” by Radek Grzybowski, CC0 1.0, via Wikimedia Commons.
