INFOHARDWARE LAUNCH·2026-10-20
NVIDIA RTX Spark laptops on sale: Grace plus Blackwell, up to 128GB memory
▼ WHAT HAPPENED
NVIDIA put RTX Spark laptops on sale globally on October 16, 2026, from ASUS, Dell, HP, Lenovo, Microsoft and MSI, through the makers' websites and major retailers. RTX Spark is a single superchip that pairs a Blackwell RTX GPU with a Grace CPU; MediaTek worked with NVIDIA on the custom CPU design. NVIDIA quotes up to 128GB of unified memory and up to 1 petaflop of FP4 AI performance.
NVIDIA's product page lists two laptop configurations, both named RTX Spark N1X. One has a 6,144-core GPU, a 20-core CPU and up to 128 GB of LPDDR5X. The other has a 5,120-core GPU and an 18-core CPU. Both are rated 45-80 W TDP. NVIDIA's pages disagree on the smaller one's memory: the product page says up to 64 GB, the full specifications page says up to 32 GB and lists 827 teraflops of AI performance for it. A desktop configuration (6,144-core GPU, 20-core CPU, 140 W, up to 128 GB) is also listed. The specifications page gives a 256-bit memory interface for all three, but no bandwidth figure.
The October 16 press release says starting at $2,599; NVIDIA's October 17 blog says starting at $2,399. The press release adds that pricing and configurations vary by region.
The CUDA Toolkit 13.4 Update 2 release notes give 616.41 or later as the Windows driver for the RTX Spark device. The press release lists the first PyTorch-CUDA build for Windows on Arm and the first native Conda installer, and says Windows Subsystem for Linux provides Linux tools. NVIDIA's October 15 blog says Nemotron 3.5 Bolt 74B-A9B uses native 2-bit quantization to fit in 24GB and has support in llama.cpp, Ollama, LM Studio, Unsloth and vLLM. The press release ties it to "supported RTX Spark configurations" without naming them.
▼ OPERATOR ANGLE
Before you buy, check the memory on the exact laptop SKU, not the RTX Spark headline. The 128GB ceiling applies to the 6,144-core configuration, and NVIDIA's own pages disagree on the ceiling for the 5,120-core one. Memory size limits which models you can load. NVIDIA publishes no bandwidth figure, so do not estimate tokens per second from the spec sheet; wait for independent measurements. We have not benchmarked these machines. Install the driver version the CUDA release notes name before you set up a CUDA toolchain. This is a Windows on Arm platform, so confirm that your runtime has a Windows on Arm build with CUDA support, or that it works with GPU access under WSL on this machine. Confirm the price at the retailer, since NVIDIA's two posts differ.
SOURCE: https://nvidianews.nvidia.com/news/nvidia-rtx-spark-pcs-available-today-a-new-beginning-for-windows-pcs[VENDOR-PRESS]
▼ ENTITIES REFERENCED
[pulse item] · runlocalai.co/pulse/nvidia-rtx-spark-laptops-on-sale-up-to-128gb-unified-memory