NVIDIA RTX Spark Launch Countdown Begins for October 16
NVIDIA opened the official countdown for RTX Spark PCs shipping October 16, packing Blackwell and Grace silicon with up to 128 GB of unified memory.
Unified processor for Windows PCs pairs Blackwell and Grace silicon with up to 128 GB of memory to run local AI models starting October 16.
NVIDIA started the official countdown for the commercial debut of the NVIDIA RTX Spark superchip, a unified processor built to execute autonomous agents and machine learning models directly on desktop and laptop hardware. Retail availability begins Friday, October 16, 2026. The company is staging launch gatherings in Berlin, Beijing, Seoul, Istanbul, Taipei, and San Francisco, alongside an Instagram livestream announced on its official X account (@nvidia).
NVIDIA RTX Spark Architecture with MediaTek
The design bypasses the traditional split between x86 CPUs and discrete graphics cards. Inside a single package, the silicon brings together a Blackwell GPU carrying up to 6,144 CUDA cores and a 20-core Grace CPU. Both engines communicate across an NVLink-C2C interconnect and tap into up to 128 GB of unified LPDDR5X memory. NVIDIA rates compute capacity at up to 1 petaflop of FP4 inference throughput within an operational power window of 45 to 80 watts.
MediaTek co-developed the high-speed unified memory controller. This shared memory pool directly addresses the primary bottleneck facing portable computers under heavy neural network loads: narrow memory buses and insufficient video memory, which stall local inference behind sluggish PCIe transfers.
Local Agent Support and Windows Ecosystem
Native integration with the CUDA software stack on Windows 11 allows developers and teams to run models of up to 120 billion parameters locally, supporting context windows of up to 1 million tokens. The central pitch for technical users is autonomy: keeping multi-step agent workflows running locally around the clock without network delays or unpredictable cloud API invoices.
For display output and gaming workloads, the chip introduces DLSS 5 with neural rendering driven by 3D buffers, alongside NVIDIA Broadcast 2.3 and PAIR. Unlike earlier attempts to bring ARM silicon to Windows, NVIDIA confirmed that widely used anti-cheat suites, including Vanguard, RICOCHET, BattlEye, and Easy Anti-Cheat, will run on the platform from launch.
Partner Hardware, Pricing, and Caveats
Commercial systems will reach store shelves through established OEM partners. Microsoft confirmed the Surface Laptop Ultra with a starting price of $2,599. Machines from ASUS, Dell, HP, Lenovo, and MSI are also scheduled for October 16, with Acer and GIGABYTE lining up subsequent releases. At local launch venues, NVIDIA plans to raffle custom ASUS ProArt P14 laptops signed by CEO Jensen Huang. In Berlin, the gathering at XPERION Alexanderplatz will feature a green lighting installation on the Berlin TV Tower.
Despite NVIDIA’s published figures, thermal sustained behavior and actual battery draw during heavy local AI inference remain unverified by third parties. Manufacturer claims of all-day battery life will remain untested until independent review embargos lift. In international markets such as Brazil, local retail release dates and pricing remain unannounced, leaving claims of local store availability on October 16 as an unverified RUMOR.
Sources
- NVIDIA on X (@nvidia) - Official countdown announcement for RTX Spark
- NVIDIA on X (@nvidia) - Details on global launch events and partner hardware
- NVIDIA RTX Spark - Official product portal with hardware specifications
- NVIDIA Investor Relations - Joint press release from NVIDIA and Microsoft
- NVIDIA GeForce News - Overview of DLSS 5 and gaming ecosystem support
- MediaTek Personal Computing - Technical details on the unified memory controller
- Eurogamer Germany - Coverage of the Berlin launch event at XPERION