The IT Home page is stamped 2026/10/20 15:32:58. It is a hands-on, not an Nvidia release note. The RTX Spark superchip is described as announced at Computex 2026. What this page adds is that RTX Spark laptops are now opening for order, and that the Lenovo YOGA Pro 15 with 128GB of unified memory costs 42,999 yuan. Lenovo, ASUS, MSI, Dell, and Microsoft have all released laptops with the chip. The sell-out line says “some products” and “instantly,” with no unit count.
[1]
The page lists two N1X chips that share an architecture and a memory subsystem. The 5120 Core has an 18-core CPU, 40 SMs, and 5,120 CUDA cores. The 6144 Core has a 20-core CPU, 48 SMs, and 6,144 CUDA cores. The package is described as TSMC 3nm, with the CPU die and the GPU die linked by NVLink-C2C at up to 600GB/s. Memory is LPDDR5X, up to 9400Mbps, with a peak bandwidth of 300GB/s, and unified memory written as high as 128GB. The tested machine is called the top configuration, 128GB, and the page says NVIDIA App can raise dedicated graphics memory to 96GB. The page does not say whether that top configuration is the 5120 or the 6144.
[1]The local-model leaderboard scores are not printed. MLPerf Client was run on Qwen3.6-35B-A3B and Qwen3.8-27B, and the page only says the result can be called excellent. There is no score. UL Procyon’s AI Computer Vision 2.0 is described as handled without strain, also with no score. Another paragraph says that, with a very large KV cache, 7B and 8B int4 dense models can fully reach 50 tokens per second. That is the reviewer’s summary of the small models, not a score for those two Qwen models, and not a table of one row per model. The times that are printed are in Comfy Desktop: FLUX.2 DEV NVFP4 Base is 41.9 seconds, Turbo is 16.6 seconds, and LTX-2.5 NVFP4 is 71.15 seconds. The video condition is written as 8 steps, 720p, 120fps. Local image and video generation is written as under 60W. All of that is IT Home’s own test.
[1]Graphics scores are printed, and they are not the missing large-model scores. Cinebench 2026 is 543 single-core and 7,118 multi-core. The page compares the multi-core number with 7,105 for an Apple M5 Pro in a 14-inch MacBook Pro. V-Ray is 2,418 on GPU-CUDA and 2,890 on GPU-RTX. The three Blender Cycles scenes are Monster 1,983, Junkshop 1,533, and Classroom 1,407. The gap between battery and plugged-in scores is written as held within 5%. The ending says the AI tests at this stage are still limited and do not cover every setting. Game frame rates appear later. This edition does not use them to judge local models.
[1]