As of: 2026-10-16 12:39 CEST. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/v/micro-center-rtx-spark-benchmarks-ai-gaming/ # RTX Spark Benchmarks: AI, Gaming & More at Micro Center Micro Center, 15 October 2026, YouTube. 33,776 views as of 16 October 2026. Kind: Review. Watch: https://www.youtube.com/watch?v=Rqtdx4V0TkQ ## Why it is here Micro Center's launch-day benchmark session (1 h 2 min) on the 128 GB Microsoft Surface Laptop Ultra with RTX Spark. Per the description it compares local-AI speed with NVIDIA's DGX Spark across five models (generation speed, prompt processing, time to first token, GPU energy efficiency), Cyberpunk 2077 against RTX 5070 and 5070 Ti laptops, and DaVinci Resolve and Blender workflows. The retailer's takeaway: "competitive local AI performance, promising creative capabilities, and gaming results that depend on the game and settings". About 34k views on 2026-10-16. ## Description (written by Gemini from the video) **Summary** A Micro Center presenter provides an in-depth performance review and benchmark breakdown of the new NVIDIA RTX Spark platform running on a Microsoft Surface Laptop. He analyzes the platform’s specifications, compares its local AI inference and power efficiency against the desktop DGX Spark and a mobile workstation RTX Pro 6000, and tests live PC gaming and creative workflows in DaVinci Resolve and Blender. --- **What is shown** * **[01:15]** Overview of the tested Surface Laptop hardware architecture: 20-core Grace CPU (custom built with MediaTek, N1X), Blackwell RTX GPU (6,144 CUDA cores), and 128 GB LPDDR5X unified memory connected via NVLink C2C (600 GB/s bidirectional interconnect). * **[05:13]** Physical comparison between the RTX Spark Surface Laptop and the standalone desktop NVIDIA DGX Spark unit (DGX OS Linux, ConnectX-7 NIC). * **[12:45]** Local AI inference benchmark charts evaluating generation speed, prompt processing speed, time-to-first-token, and power efficiency across five models: GLM 4.7 Flash, Nemotron 3 Nano Omni 30B, Qwen3.5 122B A10B, Muse Glimmer 30B, and Qwen3.8 27B. * **[20:45]** Cyberpunk 2077 benchmark performance charts comparing RTX Spark against discrete laptop GPUs (RTX 5070 and RTX 5070 Ti) at 1080p, 1440p, and native 3270×2180. * **[27:57]** Live gaming tests on the Surface Laptop demonstrating real-time frame rates and settings adjustments in *Cyberpunk 2077* (Native/DLSS), *Doom: The Dark Ages* [40:00], *Helldivers 2* [41:18], and *Counter-Strike 2* [46:38]. * **[51:19]** Creative editing demonstration in DaVinci Resolve Studio 21 scrubbing 10-bit 4:2:2 HEVC footage, integrating an MCP server with Codex/Hermes for automated AI timeline transcriptions and rough cuts [54:20], followed by DLSS viewport denoising in Blender [59:30]. --- **Claims & numbers** * **Availability & configurations**: The presenter states RTX Spark laptops go on sale October 16 at Micro Center across multiple OEMs (ASUS, Microsoft Surface, HP, Lenovo, Dell), with memory configurations ranging from 24 GB, 32 GB, 48 GB, 64 GB, up to 128 GB. * **Architecture specs**: The tested unit features a 20-core Grace CPU, 6,144 Blackwell CUDA cores, 1 petaFLOP FP4 AI performance, 128 GB unified LPDDR5X memory, up to 300 GB/s memory bandwidth, and 600 GB/s NVLink C2C bandwidth between CPU and GPU. * **Power envelope**: Laptop RTX Spark systems operate within a 45 W to 80 W power envelope, compared to 140 W+ on the desktop DGX Spark and 300 W max on the mobile workstation RTX Pro 6000. * **AI inference benchmarks**: Across testing with 100-token and 10,000-token prompt runs (10 runs per test), the presenter claims RTX Spark delivers inference throughput and time-to-first-token comparable to the Linux DGX Spark unit, while using lower average GPU power (approx. 43 W to 48 W vs. 50 W to 73 W on DGX Spark). * **Gaming performance**: In *Cyberpunk 2077*, the presenter claims RTX Spark matches or closely tracks an RTX 5070 laptop GPU (8 GB GDDR7) across 1080p and 1440p settings when running via Microsoft’s Prism x86-to-Arm translation layer. --- **Notable quotes** * **[00:14]** *"TL;DR: 5070 mobile performance for gaming, and then you're looking at comparable inference performance to a DGX Spark..."* * **[48:56]** *"If you really want to maximize performance... honestly, I would just drop this to the lowest resolution... when it comes to competitive shooters, a lot of guys don't really care about the graphics, it's really just more about the frames."* * **[52:03]** *"I can say without a doubt, this is going to be an editing behemoth."* --- **Assessment** This is a hands-on review and retail overview from Micro Center demonstrating verified retail laptop hardware running live local AI benchmarks, game engines, and creative production software. The presenter is transparent about the machine's strengths and limitations, explicitly reminding viewers that it is designed as an AI and creative workstation rather than a dedicated high-end gaming laptop. _Described by gemini-3.8-flash on 2026-10-16 from the video's audio and frames._ ## Related - 2026-10-07: [Microsoft and Nvidia launch RTX Spark Windows PCs for local AI](https://postcutoff.com/e/2026-10-07-microsoft-windows-rtx-spark-event-mxc/)