Nvidia has showcased RTX graphics running on ARM CPUs as a way of highlighting the potential performance non-x86 processors can offer to consumers — and to make the point that gaming does not require an x86 microprocessor. There are a few caveats to the presentation (we’ll address them), but make no mistake: This is as much a warning to Intel and AMD as it is an opportunity to crow about Nvidia’s graphics prowess.
Nvidia showed one game, Wolfenstein: Youngblood, and one tech demo (the Bistro, from the Open Research Content Archive (ORCA). The video Nvidia posted for GDC (embedded below) indicates that the testbed is running an RTX 3060 on a MediaTek Kompanio 1200, which is based on the MT8195 SoC. MediaTek calls these parts “chipsets,” but that word is typically used to refer to motherboard control logic, not an SoC.
MediaTek is generally known as a budget device developer, but the MT8195 looks like it has some heft to it. The SoC is built on TSMC’s 6nm process, which offers higher logic density compared with its 7nm node. TSMC built the 6nm node for companies that wanted to take advantage of its logic density but didn’t want to jump ship for 5nm and its new design rules. The CPU contains four Cortex-A78 cores and four Cortex-A55 cores with an AI processing unit, which MediaTek confusingly refers to as an “APU.” The chip uses quad-channel LPDDR4x 2133, but we don’t know how wide the channels are. This is a modest chip overall, however — the MT8195 only hits a maximum default clock speed of 2.2GHz on the Cortex-A78 cores. The SoC is typically used in Chromebooks.
We suspect that’s part of Nvidia’s point. The MT8195 is not the kind of CPU we’d expect to see in an actual ARM-powered gaming system. If it can run RTX games well when paired with just an RTX 3060, this implies good things about the products that could appear in the future.
Nvidia ported five specific capabilities to Linux: DLSS (presumably used above to improve performance), RTX Direct Illumination, RTX Global Illumination, Nvidia Real-Time Denoisers, and the RTX Memory Utility. DLSS is used to improve image quality by upscaling lower resolution output to higher resolutions, Direct Illumination supports dynamic lighting and global illumination is used to calculate ray bounces. Real-time denoising is self-explanatory and the RTX Memory Utility is said to optimize GPU memory usage.
The fact that the system is running Linux (Arch Linux) throws an interesting wrinkle into all of this. Windows on ARM support exists, but Nvidia chose not to demonstrate a system running Windows. Linux isn’t normally thought of as a gaming OS, but Valve has announced that its upcoming Steam Deck will ship with SteamOS and will run a number of Windows games through Proton, its fork of Wine. The Steam Deck relies on an AMD APU, not an Nvidia solution, but this is still an interesting demonstration of gaming performance
Nvidia isn’t going for broke with a major product announcement; GTC is a developer conference more than it is a consumer-facing event. The message here is unmistakable, if low-key: Neither x86 nor Windows are intrinsically required for high-performance gaming.
Nvidia has more reason than most companies to hammer that message. If the company’s attempt to buy ARM is approved by regulators, it will have an opportunity to make a play to become the simultaneous provider of leading-edge GPU and CPU solutions for gaming. Intel is making its own play for that title with its Xe HPG lineup of products and AMD already occupies that position if you’re a console gamer. But no single manufacturer has ever claimed a dominant market position in both CPUs and GPUs when it comes to supplying the PC gaming market. Showing off ray tracing in Linux is good for Linux as a gaming OS. Showing it on a MediaTek MT8195 is good for Nvidia. There are no plans to release this version of Youngblood commercially, according to the video, but this probably won’t be the last time we see an ARM CPU + Nvidia GPU sharing space in a laptop or desktop.
Now Read: