How Far Can You Game on an RTX Spark Laptop? DLSS 5 Hair, and One Tester's Impression: Not a PC for Games

At NVIDIA's technical briefing, NBA 2K27 was set up to play with a controller ¹. With DLSS 5 neural rendering and frame generation switched on, the demo was pitched with the line "the texture of the players' hair is on a completely different level," according to the Game*Spark reporter ¹. Hands-on, the reporter found the picture quite beautiful and felt the frame rate was running above 60fps by eye ¹.

Game*Spark's report looks at an RTX Spark PC from a gamer's point of view, from DLSS 5 visuals to local AI (the post's headline, translated from Japanese) ². This article picks out what a person who plays games would care about. Benchmark numbers live in another article, so here it is about scenes and honest opinions.

DLSS 5 Makes the Picture a Little Darker

In ITmedia PC USER's side-by-side screenshots, turning on the 3D-guided neural rendering in NBA 2K27 makes the image look a little darker ³. The reason, it says, is that shadow rendering is being filled in ³.

As Game*Spark explains it, the engine renders a frame and the AI uses it as a guide to paint in how light falls and what materials look like, more realistically. Frame generation also switches automatically between 2x, 4x and 6x depending on the scene ⁴.

Support news came up too. According to Game*Spark, RTX Spark supports DLSS 5 from launch, and after NBA 2K27, Hogwarts Legacy is planned to get DLSS 5 support on November 5 ⁵. Familiar GeForce features such as DLSS, Reflex and G-SYNC can be used as they are, it adds ⁶.

The question every gamer asks next: this is Arm, so will my games run?

Can Arm Play Games? How It Works, and the Anti-Cheat Wall

According to Game*Spark, games built for x86 run on the Arm CPU through Windows 11's Prism ⁷. In NVIDIA's technical explanation as relayed by Game*Spark, only the CPU work is translated, and GPU rendering runs as is, so games that lean heavily on the GPU are affected less ⁷. NVIDIA also said, as reported there, that 97 popular PC games run well, either natively or through Prism ⁸.

Anti-cheat is the wall. As Game*Spark relays NVIDIA's explanation, anti-cheat that runs at the kernel level has to be rebuilt for Arm, and "not everything will be ready right away" ⁹. Game*Spark writes that NVIDIA's official blog lists Counter-Strike 2, Fortnite, Marvel Rivals and ARC Raiders, among others, as playable with their anti-cheat ¹⁰. Among the main titles planned next are the Call of Duty series, League of Legends and Valorant ¹¹.

Gizmodo's review of a Lenovo model, on the other hand, reports that Call of Duty crashed before the launcher ¹². That is about the laptop reviewed, at the time of that review.

There are native examples too. Game*Spark says the Marvel Rivals demo ran natively on Arm without Prism emulation, and its developer NetEase prepared anti-cheat for Arm ¹³. ITmedia PC USER reports that the latest version uses Arm64EC, so the main parts of the program, anti-cheat included, run natively on the Arm CPU ¹⁴.

Game*Spark's advice is simple. Titles are not guaranteed to work, so check individually that the game you want to play runs ¹⁵. The reporter also sees the situation of "it's Arm, so you can't play" as changing ¹⁶.

For Modders: the Remix Story

Game*Spark reports that RTX Remix, the modding tool that adds ray tracing to older games, gained DLSS 5 support on October 23 ¹⁷. At the briefing, a Fallout 4 Remix mod was used to compare DLSS 5 on and off ¹⁷.

For Minecraft fans: an RTX Remix mod kit for Minecraft: Java Edition has been released in beta. It can add path tracing and DLSS, and it works with existing Java Edition maps as they are ¹⁸. Being able to use existing maps as they are is tempting.

A Thin Gaming Laptop, or Not a PC for Games?

Opinions split here. In ITmedia PC USER's ProArt P14 review, the benchmarks showed higher gaming performance than the GeForce RTX 5060 Laptop GPU ¹⁹. It says RTX Spark laptops can also serve as "thin" gaming laptops ²⁰.

Shimizu, who tried the MSI Prestige N16 Flip AI+ in a YouTube video, says otherwise (in Japanese). Shimizu did try games, but the result was lower than a laptop with an RTX 5070 ²¹. Many games and apps are not yet supported on Windows on Arm, so Shimizu's personal impression is that this is not a PC for games ²². When Shimizu tried to measure game frame rates with Afterburner, it did not always work ²³. Games run, Shimizu says, but they do not yet seem to deliver their full performance ²⁴.

Note that the comparisons are against different machines. ITmedia compared against the RTX 5060 Laptop GPU ¹⁹, while Shimizu compared against a laptop with an RTX 5070 ²¹, and the machines and the ways of testing differ. It is best to read each as a result for that machine and that person's tests.

Game*Spark also adds that if you choose purely on game performance, a gaming laptop with a stronger GPU at the same price is an option ²⁵. For game benchmark numbers, see How far do games and 3D go on RTX Spark laptops?.

Go Easy on Games, Try Local AI Everywhere

What makes Shimizu's video interesting is what comes next. Shimizu introduces themself as someone who had never done local AI ²⁶. The basic setup was installing LM Studio and ComfyUI, then adding models from there ²⁷. Shimizu says ChatGPT and Claude helped with the setup, and that Claude also did the operations in this video automatically ²⁸.

The video starts by generating an image of Iekei ramen from text ²⁹. In the resulting video, Shimizu says, the ramen-eating looked quite realistic ³⁰. Games may not have shown the machine's full performance, but it made a ramen video.

A few numbers, kept light. In the configuration shown in the video, the memory that can be assigned to AI is up to 96GB ³¹. A 120B-class model ran in a little over 60GB in its quantized version, which Shimizu calls surprising ³². Generating a single image took about 2.2 seconds for the fastest models and about 38.6 seconds for the slowest ³³. gpt-oss, on the other hand, took especially long to load, about 160 seconds ³⁴. These are Shimizu's results on this machine; the text models were quantized versions.

Images and videos get repeated until a good one comes out. With pay-per-use billing the cost could become unlimited, Shimizu says, but locally, if you have the hardware, you can basically repeat for the cost of electricity ³⁵.

Shimizu's conclusion is that, for now, it is very much a PC for AI, and one that should appeal to people who want to try many things or who want to carry around an environment like DGX Spark ³⁶. The honest wish is for it to grow in games and creative work too, with hopes that more compatible software will arrive ³⁷.

For Streamers and Clip Makers

The briefing demos also had something for people who stream or cut clips. ITmedia PC USER reports that Hermes, running on the device, operated DaVinci Resolve as instructed and exported a vertical video for social media ³⁸. Another demo combined LTX-2.5 and RTX Video to upscale a video generated at Full HD and 30fps to 4K and 120fps, exported as HDR (HLG) ³⁹.

ITmedia introduced the demos with a headline that sets cloud AI costs against an electricity bill (translated from Japanese) ⁴⁰. The chart shown at the briefing said a monthly token load costing 1,200 dollars in the cloud would cost 20 dollars of electricity a month on an RTX Spark PC ⁴¹. That is the claim of the briefing's chart.

A word on battery as well. According to Game*Spark, at around the same 60W of power consumption, performance is about 2x that of a Blackwell-generation laptop GPU and CPU pairing, and a presenter said it keeps about 98% of its AC-power performance on battery ⁴². This is NVIDIA's side of the story.

If You're Choosing

Check, from the official sources, whether the games you want to play and the tools you want to use are supported before you buy.

Shimizu used an MSI model, and the machine ITmedia benchmarked is an ASUS model.

If you are looking for RTX Spark laptops, start here.

A basketball game like the one in the demo makes you want a controller in your hands.

For competitive games like Marvel Rivals, a headset that carries footsteps and callouts makes it more fun.

Read Next

For game benchmark numbers, see How far do games and 3D go on RTX Spark laptops?. For the whole picture of the briefing, see How NVIDIA talked about RTX Spark at the Tokyo briefing. For the people making games, see Anime and games: the creators using AI are showing up.