{"slug": "ai-upscaling-is-about-to-transform-android-gaming-forever", "title": "AI upscaling is about to transform Android gaming forever", "summary": "Arm has unveiled its new Mali G2-Ultra GPU architecture with an integrated NX neural accelerator, promising AI-assisted upscaling and frame generation for Android gaming. The G2-Ultra delivers 9-14% gaming performance gains over its predecessor, up to 18-24% better ray tracing, and can boost a demo from 15fps to 60fps at just over 2W power. Arm's NX accelerator supports INT8/INT16 operations and includes a motion engine, marking a significant step toward bringing desktop-class neural rendering to mobile devices.", "body_md": "Affiliate links on Android Authority may earn us a commission. [Learn more.](https://www.androidauthority.com/external-links/)\n\n# AI upscaling is about to transform Android gaming forever\n\nSep 7, 2026 — 10:00 PM ET\n\nMobile graphics have increasingly closed the performance gap with integrated desktop-class GPUs and have even kept pace with cutting-edge rendering technologies and features typically found in gaming consoles and PCs. Ray tracing is one of the most recent examples, bringing a trimmed-down, lightweight version of the powerhouse lighting tech to our pockets at a fraction of the power draw.\n\nBut in the meantime, desktop-class GPUs have raced ahead with neural network-assisted resolution upscaling and frame-rate generation, enabling 4K/120 fps gameplay without the traditional rendering effort. Well, mobile is catching up here, too.\n\nYour next smartphone could soon boast AI-assisted upscaling, if it features a newly unveiled Arm Mali NX GPU.\n\n## Meet the Arm Mali G2-Ultra\n\nBefore we get into the AI enhancements found in this year’s Mali graphics chips, we need to dive into the core architecture and portfolio change in the G2 series.\n\nIf you’re after the raw performance numbers, Arm lists quite a range of results. Gaming performance increases by 9% (Honkair Starrail) to 14% (Fortnite) when comparing the 14-core Mali G1-Ultra reference platform to the G2-Ultra equivalent (which also features a 1.53GHz versus 1.70GHz shader core clock difference). At the upper end, Arm expects we could see 17-20% performance improvements in the most demanding mobile graphics benchmarks.\n\nGeneration-on-generation improvements in general rasterization are reasonably modest, but that doesn’t mean things haven’t changed. The execution engines have been revamped to support much larger desktop-class shaders, with up to twice the registers per warp (now 128 instead of 64) and also more dynamic allocation, supporting 16 warp-wide steps from 16 to 128 registers. This allows advanced graphical shader technologies like Unreal 5’s Lumen and Nanite to run without register spilling and the associated performance drops from reaching out to DRAM. This marks Arm’s biggest change to Mali in seven generations.\n\nThe G2 architecture also sports Arm’s third-generation ray-tracing unit, which introduces a more compact triangle representation that uses shared edges to reduce data duplication and fit more geometry in the cache. Arm’s improvements this generation lower DRAM traffic by 13%, removing bottlenecks and improving energy efficiency.\n\nThe G2 series also supports hardware-accelerated opacity micromaps for the first time, making it far more feasible to implement desktop-quality scenery with ray tracing in the mobile form factor. Combined, the core reference design promises around 18-24% better ray-tracing performance than its predecessor, depending on the benchmark.\n\n## NX explained — AI inside the GPU\n\nOK, those performance metrics are all very welcome, but Arm’s biggest announcement is undoubtedly the introduction of native AI support via the new NX neural accelerator. NX is a tightly integrated, optional hardware accelerator that sits inside the GPU’s shader core to run neural graphics tasks while costing as little as 1W in power.\n\nThe tightly integrated neural accelerator supports INT8 and INT16 operations and includes its own motion engine processing block, which we’ll touch on later. The NX core also has its own clock frequency operating at up to 2x the main GPU clock using the same voltage, and shares memory with the rest of the GPU subsystem to ensure memory coherency.\n\nThe headline figures certainly sound impressive. NX can take Arm’s Neural Dawn demo from an unplayable 15fps to a sustained 60fps with a power budget of just over 2W. That’s a 4x FPS boost and efficiency improvement, paired with a 70% reduction in DRAM memory traffic compared to last year’s G1-Ultra graphics core.\n\nOf course, there’s a compromise here: hitting this performance level uses 2x Neural Super Sampling and Denoising (NSSD) upscaling from 540p to 1080p, which also implements ray-tracing denoise in the same pass. The next step uses Arm’s Neural Frame Rate Upscaling (NRFU) to double the frame rate from 30 fps to 60 fps. Arm claims the end result is a more detailed image, a higher frame rate, and lower power consumption. However, I’d be very interested in how the latency feels and what artifacts can be observed. Arm did not elaborate on this.\n\nUnlike a generic upscaler (such as TAA), Arm’s Neural Super Sampling is trained on real-game data, allowing the neural network to recover fine detail and data with just a quarter of the traditional input data. You’ll probably be familiar with this idea from NVIDIA’s DLSS, AMD’s latest FSR, and even Intel’s XeSS 2. [Qualcomm’s next-gen mobile GPU](https://www.androidauthority.com/qualcomm-snapdragon-8-elite-gen-6-gpu-3705379/) follows a similar idea. Arm claims its approach improves temporal stability, reducing jitter and ghosting. The use of a history buffer, no doubt, helps here as well, and there’s also an integrated anti-aliasing that removes the need for a separate pass. NSSD (NSS with denoise) is an enhanced version of the technique that denoises ray-tracing results as part of the upscaling pipeline.\n\nNeural Frame Rate Upscaling, meanwhile, delivers up to 2x the input frame rate, but its implementation is arguably more complex to avoid introducing distracting visual artifacts. It takes depth and motion vectors from two pairs of frames, and hardware accelerates “optical flow” estimation beyond the provided in-game vectors, feeding all of this into the neural network to produce a new intermediate frame. Depth and motion frames are usually provided by the game engine, but the G2 architecture now includes a motion engine that detects motion direction between frames, using a technique similar to video encoding engines.\n\nArm’s three neural models are in various stages of development. NSS was published last year, and NRFU has just been released. Both are available for developers via Hugging Face.NSSD is in early access but will be playable in 2026. All three will require a GPU with NX capabilities to make use of, so they aren’t backward compatible with last year’s Mali G1 series.\n\nArm is aiming to make AI-supported graphics as viable as possible on its platform by supporting Vulkan ML for a wide range of applications and allowing for the retention and per-game shipping of its own models. Time will tell if that’s enough to attract an even wider range of partners to start using AI-upscaling and ray-tracing in their next-gen mobile games.\n\n## The makings of a next-gen mobile processor\n\nPutting this all together, there’s an even wider range of Mali configuration options than ever before, and it’s all getting rather overly complicated. As before, the Ultra range offers the highest core count, between 10 and 24 shader cores, to scale all the way up to laptop-class performance. The Ultra branding is appended to the NX brand, as it must contain at least one NX core, and there’s also an opinion to include ray-tracing support, though this doesn’t change the naming scheme. The G2-Premium range is a mid-tier mobile option with six to nine shader cores, optional NX branding if at least one such core is present, and optional ray tracing. Finally, the budget “Pro” segment caters to five cores and below, again with optional AI acceleration.\n\nAll in all, there are five possible configuration groups: the G2-Ultra NX, G2-Premium NX, G2-Premium, G2-Pro NX, and G2-Pro, with some variation within these brackets. Confused? Me too. In any case, we’re likely looking at around 12-core versions for premium smartphone platforms, no doubt with roughly half as many NX cores and ray-tracing support.\n\nWhile bringing “fake frames” to mobile will undoubtedly be a controversial topic in some circles, I’ll admit I’m hesitant to see developers rely on AI techniques to compensate for a lack of traditional optimization and simply rush products to market. Still, mobile perhaps stands to benefit far more from AI-assisted graphics rendering than desktop and console players, owing to our more conservative power budgets for high-end rendering.\n\nIf Arm’s claims that titles look just as good but run more smoothly, and our batteries can game for longer, turn out to be true, then I’m all for it.\n\nThank you for being part of our community. Read our [Comment Policy](https://www.androidauthority.com/android-authority-comment-policy/) before posting.", "url": "https://wpnews.pro/news/ai-upscaling-is-about-to-transform-android-gaming-forever", "canonical_source": "https://www.androidauthority.com/arm-mali-g2-ultra-deep-dive-3706351/", "published_at": "2026-09-08 02:00:53+00:00", "updated_at": "2026-09-08 02:30:12.469337+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-chips", "ai-infrastructure", "generative-ai"], "entities": ["Arm", "Mali G2-Ultra", "NX", "G1-Ultra", "Unreal 5", "Lumen", "Nanite", "Android Authority"], "alternates": {"html": "https://wpnews.pro/news/ai-upscaling-is-about-to-transform-android-gaming-forever", "markdown": "https://wpnews.pro/news/ai-upscaling-is-about-to-transform-android-gaming-forever.md", "text": "https://wpnews.pro/news/ai-upscaling-is-about-to-transform-android-gaming-forever.txt", "jsonld": "https://wpnews.pro/news/ai-upscaling-is-about-to-transform-android-gaming-forever.jsonld"}}