What happened in Seoul
NVIDIA held a press briefing for RTX Spark at the JW Marriott Hotel Seoul on November 6, according to Financial News. Attendees could try features across AI development, content creation and gaming. Jeff Yen, NVIDIA's technical marketing director for Asia-Pacific, led the presentation.
Inven Global says RTX Spark went on sale in South Korea on October 16. That matches NVIDIA's global launch announcement, which says the laptops became available that day from ASUS, Dell, HP, Lenovo, Microsoft and MSI. NVIDIA says pricing and configurations vary by region.
The Seoul briefing was a regional showcase of a product that was already on sale. It was not a new launch.
What RTX Spark is
NVIDIA describes RTX Spark as a platform for Windows PCs, not a discrete graphics card. The core is a superchip that pairs a Blackwell RTX GPU with a Grace CPU. NVIDIA says it offers up to 128 GB of unified memory and up to 1 petaflop of FP4 AI performance. Both numbers are ceilings, not what every laptop ships with.
Inven adds a few details from the presentation:
- The CPU was co-designed with MediaTek and has up to 20 cores.
- The CPU and GPU share data over NVLink-C2C.
- The unified memory is meant to ease the capacity limits of running large AI models on ordinary PCs.
Microsoft's own materials fit this picture. Its Surface Laptop Ultra page says the machine runs Windows 11 on an Arm-based NVIDIA RTX Spark platform, and that unified memory is configurable up to 128 GB. That Arm detail matters for app compatibility, which I cover below.
The local AI pitch
The headline claim is that AI agents can run on the PC itself instead of in the cloud. Yen listed the usual advantages of local AI: privacy, less dependence on a network connection, faster responses and lower cloud API costs.
"Local" is not the same as "always offline" or "private by default." Inven reports that NVIDIA also showed NeMo Switchyard, which sends simpler tasks to local models and hands harder ones to cloud models. In practice this is a hybrid setup, and Microsoft describes the same approach. Its Surface announcement says Microsoft Dev Box adds cloud capacity when needed.
For IT teams, the question is which tasks stay on the device and which leave it. That decides how much of the privacy benefit you actually get.
The demos
The Seoul demos covered three areas.
AI agents
- Chief of Staff handles schedules and tasks.
- Microduck assists with development and robotics work.
Content creation
- Video Producer Agent assists with AI-based video production.
- Music Performance Reimagined showed AI in creative work.
- NVIDIA says more than 130 creative applications support RTX acceleration. That is a company figure.
Gaming
- NBA 2K27 and AION 2 were the featured titles.
- Per Inven, NBA 2K27 was the DLSS 5 example. AION 2, an Unreal Engine 5 game, was shown with DLSS 4.5. The two are not interchangeable.
Inven states a limit plainly. Journalists did not operate the software or measure performance. NVIDIA staff ran the features in preconfigured environments. So these were demonstrations, not independent tests.
DLSS 5 and the gaming numbers
Inven says DLSS 5's 3D-Guided Neural Rendering uses AI to improve lighting and materials, including skin and hair, rather than only upscaling or generating frames. NVIDIA's launch release says a new AI model for DLSS 5 is 2x faster. I can't tie that sentence to a verified source, so treat the release wording as NVIDIA's claim. The same release says the model boosts DLSS 5 frame rates by up to 1.4x.
Yen's slides, as Inven describes them, showed NBA 2K27 going from about 65 FPS to about 90 FPS after the October model update. That is roughly 1.4x, which is consistent with NVIDIA's "up to" figure. It is still vendor data, and the settings weren't given.
Inven also relays NVIDIA's claim that 97 of the 100 most popular games at launch are playable on RTX Spark. Treat that as a snapshot, not a guarantee. Anti-cheat software and Arm-based Windows compatibility are the usual trouble spots, so check your specific titles.
Hardware, battery and price
Financial News lists the hands-on laptops as the Microsoft Surface Laptop Ultra, ASUS ProArt P16, MSI Prestige N16 Flip AI+ and Lenovo Yoga Pro 9n. Inven drops the "AI+" from MSI's name. The two outlets don't share a single SKU naming convention, and neither gives full configurations.
Inven's other reported details:
- The laptop superchip runs at a 45–80 W TDP.
- A desktop Mini PC was on display, with a planned November release.
- Yen shared an internal NVIDIA result that the laptops keep about 98% of plugged-in performance on battery. No test protocol was published.
- Inven's impressions of light weight and low heat were labeled as on-site observations, not measurements.
Price came up in the Q&A. Yen said no one would call RTX Spark inexpensive and pointed to early adopters, AI developers and creators as the first buyers. An NVIDIA representative added that gamers focused only on gaming might do better with a conventional high-performance gaming laptop.
NVIDIA's global announcement says laptops start at $2,599, and PCWorld's review of the Surface Laptop Ultra lists a $2,599.99 base price. The review unit PCWorld tested cost $3,699.99. PCWorld also mentions a $7,000 ASUS ProArt P16 with 128 GB of unified memory. Microsoft's Surface RTX Spark Dev Box is available for pre-order for $5,999 (MSRP), exclusively on Microsoft.com in the U.S. None of these are Korean prices, which the two reports don't give.
Independent testing so far
Independent testing exists, though not from the Seoul event. PCWorld says it spent a month testing several RTX Spark laptops. In its Surface Laptop Ultra review, the 20 Arm CPU cores delivered excellent multicore performance, while single-core performance was considerably less impressive. It also saw roughly 10 hours in office-work testing, which it called respectable but not revolutionary. That is a useful counterweight to NVIDIA's battery messaging. The 10-hour figure is one reviewer's office-work result, and it doesn't speak to the 98% claim, which is about performance retention.
Microsoft's own comparison against the MacBook Pro, based on testing it commissioned from Principled Technologies, is also vendor-sponsored. It used preproduction units.
Practical takeaways
- Check app compatibility. These are Windows 11 on Arm machines. Many popular apps run natively on Arm, but verify any specialized software, drivers or anti-cheat tools you depend on.
- Match memory to the model. The 128 GB figure is the top configuration. Most buyers will see lower tiers, and the model sizes you can run depend on that choice.
- Understand what "local" means. Ask which agent tasks run on the device and which go to the cloud.
- Be careful with the gaming claims. DLSS 5 gains are NVIDIA's slide figures for one title.
- Verify promotions. Inven mentions a launch offer of two months of Perplexity Pro for select new buyers. NVIDIA's release also mentions a limited-time promotion for new Perplexity users. Check eligibility before counting on it.
The bottom line
The Seoul briefing presented RTX Spark as a premium Windows platform that combines shared memory, RTX graphics and local AI. NVIDIA's own staff conceded the price is high and that pure gamers may be better served elsewhere. The agent demos were staged, and the performance figures came from NVIDIA. Independent reviews so far describe a capable and fairly ordinary Windows laptop with strong multicore results. Whether it justifies the premium depends on whether your workload actually uses the memory and the local AI stack.