Nvidia’s pint-sized DGX Spark AI workstations just unofficially took a rite of passage for PC hardware: they can game.
Early community tests indicate it’s perfectly playable, just as PlayStation 3 titles emulate nicely—assuming you’re prepared to spend the eye-popping sum of money currently required to use an AI dev box to do so.
What’s Required To Run Cyberpunk 2077 On DGX Spark
Early owners, including someone who paid as much as $4,000 for a DGX Spark reference unit, got Cyberpunk 2077 running on their little refrigerator practice machines using a combination of Box64 (an x86-to-ARM translation layer) and some help from Box32 and Steam.
Based on information relayed by VideoCardz, the game reportedly delivered an average of about 50 fps when playing at 1080p with medium settings. That’s a respectable showing from something never intended to game, especially when you consider it’s converting x86 game code on an ARM platform and also translating graphics APIs through Proton.
The system’s Blackwell-class GPU is home to 6,144 CUDA cores—which should put it somewhere around an RTX 5070 class of product on paper—and is mated with a 20-core Grace ARM CPU in the GB10 superchip. With 128GB of LPDDR5X memory, there’s plenty of overhead for large assets and AI workloads. But the bottleneck continues to be the software stack. DLSS wasn’t usable on the test rig, and frame-producing features weren’t enabled either, revealing that drivers and API support still lean toward typical x86 Windows rigs.
PS3 and Xbox 360 emulation performance on DGX Spark
On YouTube, ETA Prime did a video on an MSI EdgeXpert AI system following the same DGX Spark design for roughly $3,000. They ran PS3 and Xbox 360-era games—such as Skate 3 and the original Forza Horizon—in RPCS3 and Xemu at solid frame rates of 30 fps. Emulation is often CPU-bound and favors strong single-threaded performance, paired with a good Vulkan driver: the Grace CPU and Blackwell GPU delivered in spades on that front, and the maturity of RPCS3’s Vulkan renderer continues to serve us well too.
It’s not the litmus test of this class of hardware, but it does kind of drive home the point: tuned translation layers and emulators combined with modern ARM devices that actually have powerful GPU subsystems can be a console-like experience. The RPCS3 project has been steadily expanding its compatibility into the thousands, and Xemu’s headway on Xbox emulation is a welcome addition to this trend, providing fans with a broadening and increasingly functional library to delve back into.
Why an AI Development Box Can Game at All
Don’t think of DGX Spark machines as living room entertainment but rather as small-scale AI development. But the same traits that make them powerful for model training—tons of GPU compute, high memory bandwidth, and fast ARM cores—in gaming aren’t a bad thing once you get the software out of the way. In other words, you are trading native ecosystem support for raw horsepower and bridging in the translator.
Translation overhead can be significant. Box64 does dynamic recompilation of x86_64 to ARM; Proton frequently translates DirectX calls to Vulkan. Every layer adds CPU and driver pressure, which is why 50 fps at 1080p medium in Cyberpunk stands out. An equivalent x86 PC with native drivers would probably post higher frame rates at the same settings—especially when using DLSS and frame generation—but that’s exactly what makes these ARM-first results exciting as a proof of concept.
The Cost and the Catch of Gaming on Nvidia DGX Spark
There’s no sugarcoating the economics. At $3,000 to $4,000, a DGX Spark setup is an extremely wasteful way to game. Less than that buys you a mainstream desktop with a midrange GeForce card that consumes far less power and provides better, more consistent game performance while supporting all driver features. Even handheld PCs of today can run many of these modern titles at 800p up to 1080p—all with less power draw and far less setup friction.
A bigger problem is software support.
Who will be demanding it?
If Nvidia goes this route, gaming on Spark would be easier to implement with the presence of official drivers and feature parity for ARM Linux and/or Windows on ARM. But that service would appeal to just a thin slice of what’s already a niche product, and at the same time Nvidia’s focus remains on datacenter deployments, workstation AI, and enterprise software stacks. In other words, don’t expect Spark to get the Game Ready driver treatment in the near future.
What This Says About ARM and the Future of PC Gaming
One thing the experiments do show is a more general trend: ARM-based systems are getting closer to being considered credible PC gaming devices, as they have strong GPUs, maturing translation layers, and emulators that target Vulkan more directly.
We’ve seen similar momentum in other ecosystems—Apple Silicon’s Rosetta 2 and Metal-based ports, as well as a whole litany of Windows on ARM efforts. As toolchains get better, going non-x86 becomes less an issue for certain classes of games.
At present, DGX Spark gaming is a mere novelty—impressive in showing what’s technically possible rather than something you’d go out of your way to do. But it’s a useful after-the-fact data point for developers and hardware makers: with the right software scaffolding, an ARM system plus an able GPU can power modern gameplay and accurate seventh-gen console emulation. However, if future ARM PCs actually incorporate this feature with native ports and full-featured drivers, the “can I run it” question could get a lot less thrilling—and a bit more mundane.