Intel Unveils XeSS 3 with Multi-Frame Generation: A Game-Changer in AI-Powered Graphics

Intel Claims You Won't Need an Nvidia GPU to Push Gaming Frame Rates Through the Roof

Nvidia previously claimed you needed its GPUs to achieve multi-frame generation. Intel says "nuh uh." Intel is trying to get into the good graces of gamers with its big Panther Lake chipset reveal. In that vein, Intel took more time to explain an update to its two-year-old AI upscaler, XeSS 2. No, we’re not getting XeSS 3. No, Intel hasn’t confirmed what the hell it's doing with future GPUs. Instead, like Nvidia before, Intel is getting into multi-frame generation to artificially push game frame rates higher. However, unlike Nvidia, Intel isn’t restricting players to its own chips. (Full disclosure: Intel invited me to its chipset fab in Phoenix, Ariz. Travel and lodging were paid by Intel, but Gizmodo did not guarantee any coverage as a condition of accepting the trip.) Multi-frame generation is a word that could create a typhoon of controversy among PC gaming circles with its mere utterance. At its simplest, it’s a process of shoving multiple AI-generated frames in between two frames rendered by the PC itself. Nvidia sold its latest RTX 50-series GPUs on the promise that multi-frame generation would push games well beyond their rasterization performance, or their ability to put a big number into your fps counter when playing games. Naysayers have been calling frame generation “fake frames,†and it's a moniker that’s not entirely wrong. Nvidia CEO Jensen Huang went full used car salesman at CES 2025 when he promised a lower-end RTX 5070 graphics card could beat the company’s last flagship GPU, the RTX 4090. In practice, frame generation makes sense for lower-end devices more than some high-end desktop PC that already has the graphics capabilities to push the utmost pixels. That’s where XeSS 2’s multi-frame gen capabilities could start to change the nature of how we think of lighter PCs orâ€"potentiallyâ€"supercharge handheld PCs like MSI’s Claw series. Frame generation technology relies on PCs having enough data to help stick the AI-made frames in between the ones rendered by the PC. Common wisdom states that most games want between 45 and 55 fps before you kick in frame generation. This helps eschew problems with odd visual glitches and artifacts, where you’ll find visual discrepancies that weren’t meant to be there. In a roundtable with reporters, Tom Petersen, Intel’s big cheese behind graphics and software, said he was fine using multi-frame gen at framerates as low as 30, which bumps it up at 4x to approximately 120 fps. Intel showed off the upcoming Painkiller reboot with a new performance monitor that can show frame rates before and after frame generation, akin to Steam’s updated performance overlay. With the quadrupling in frame generation (that means there are four AI frames between each rendered frame), the game was running at around 150 fps on a pre-production version of Panther Lake with 12 Xe3 GPU cores, on a reference platform running at 45W. Frame generation creates latency that discerning players can feel and see. This is due to the nature of the technology, since it has to know the next frame in a sequence to insert the AI-generated frames between them. Petersen said the latency could be more, depending on the game and framerate you’re playing at, but it will be down to users whether they opt in or not. “There are a whole bunch of technologies that have not been announced that could make that better, but right now you can see some just turn off frame gen,†Petersen said. He later referred to potential technology that would use AI to predict mouse clicks, which could potentially allow the game to artificially remove the sensation of latency without removing it altogether. In the future, there’s an opportunity for games to have predictive frame gen that kicks in only in certain moments where gamers won’t feel the impact of latency as much. He even went as far as to say frame rates closer to 30 would actually be better than doing frame gen at higher frame rates, as the latency isn’t as noticeable at lower frame rates as higher frame rates. That being said, I spoke to others at Intel who may still recommend at least 45 fps before enabling multi-frame generation. Your mileage may vary. I ran through the Painkiller demo at around 50 fps, which turned into around 200 fps on-screen playing at 1080p. The game is as dark and gray as an older Xbox 360 title when we all somehow thought Gears of War looked great. It was more difficult to spot issues with textures or geometry through all the grime. However, the fire effectsâ€"which previously posed problems for Intel's 2x frame gen technologyâ€"seemed closer to how it would look without using upscaling or frame generation. Petersen wasn’t afraid to share his opinion regarding Nvidia’s multi-frame gen beholden to its latest RTX 50-series of GPUs. “Trash,†he called it under his breath. He’s half-joking, maybe, but Intel now sees itself as an underdog fighting a behemoth; it’s allowing itself to act scrappy. Still, what's more important is how well its Panther Lake chips can perform in gaming scenarios before you enable any AI options. Intel claims it has optimized power delivery on the CPU and GPU to reduce stuttering in games. XeSS 2's multi-frame gen isn’t the kind of technology that makes expensive hardware better. Instead, it offers more options for lower-end systems. XeSS 2 remains hardware agnostic, meaning it can run on any device, whether it has an Intel or AMD chip inside. XeSS 2 works on two different kernels for AMD and Intel devices, and on AMD-powered systems, it requires a compatibility layer that will make the upscaler work slightly worse than on Intel’s chips. Intel is now the only major U.S. chipmaker still developing hardware-agnostic AI upscaling for PC. At the same time, XeSS is still lagging behind the competition in terms of supported games. AMD has turned its attention to the latest version of FidelityFX Super Resolution, or FSR 4. The upgrade over FSR 3 and 3.5 has allowed upscaled games to look less muddy with sharper details in both the foreground and background. AMD relegated FSR 4 to its latest Radeon RX 9000 series GPUs, as in all those currently using RDNA 4 microarchitecture. AMD recently claimed its latest AI upscaling drivers are available in more than 85 titles. Nvidia’s latest version of its Deep Learning Super Sampling (DLSS 4) AI upscaler is currently available in more than 175 games, according to the company. In May, Intel said XeSS is available in more than 200 games, but if you’re regularly playing the latest titles, it can be a crapshoot which games support Intel’s upscaling and which don’t. Intel obviously hopes more developers sign on once they see what XeSS 2's multi-frame gen can do. XeSS multi-frame gen should be available to some games in the future, though players may need to use Intel software to enable it to override on games that are already compatible with Intel's upscaler.

Intel reveals XeSS 3 with Multi-Frame Generation - and unlike Nvidia's MFG, it works on older GPUs

This is a big advantage over Nvidia's MFG, and with Panther Lake chips due as 2026 arrives, there are exciting possibilities on the horizon Intel has taken the wraps off its next-gen XeSS tech to boost frame rates with its Arc (and Xe) GPUs, and Team Blue is following in Nvidia's footsteps with its own take on Multi-Frame Generation (MFG). VideoCardz reports that Intel announced XeSS 3 upscaling and XeSS-MFG, which facilitates multi-frame interpolation, meaning that the feature inserts multiple AI-generated frames in between the (real) frames rendered by the GPU. This artificially boosts the frame rate, of course, and XeSS-MFG will offer up to 4x frame generation (meaning three AI frames for every real one, to quadruple the frame rate). The welcome twist with Intel's MFG tech is that it's going to be backwards compatible with older Arc GPUs, which means the Arc Alchemist 1st-gen desktop graphics cards - and Xe2 integrated GPUs (and eventually Xe1, too). There was no release date revealed for XeSS 3 plus MFG, but it's expected to debut next year. To be more precise, XeSS 3 with MFG is expected to be part of Intel's grand plan to pep up mobile CPUs along with Panther Lake. Those Intel Core Ultra series 3 processors for laptops (and handhelds) will see "broad market availability starting January 2026", Intel also told us in a press release. So, Panther Lake with integrated Xe3 graphics is already expected to 'set a new standard' and provide over 50% faster graphics performance compared to Xe2 - a big leap, and one that can be made into an even bigger jump when you throw frame generation into the equation (with supported games, that is). In short, next year we could see some very compelling wallet-friendly gaming laptops packing Intel chips with seriously powerful integrated GPUs (not to mention fast thin-and-light gaming notebooks and peppy handhelds, too). Intel is also claiming a notable achievement in that it'll be the first GPU maker to allow previous generations of its graphics cards to use MFG - Nvidia is the only firm offering MFG right now, but the feature remains exclusive to its most recent RTX 5000 GPUs. AMD has frame generation, but has not entered the Multi-Frame Generation race just yet - though it's expected to do so with its next-gen take on FSR codenamed 'Redstone'.

Intel announces XeSS 3 with multi-frame generation, putting it ahead of AMD in the AI-powered graphics performance race

Though Intel would probably prefer its GPU sales to be ahead instead. At this year's Technology Tour event in Arizona, Intel not only launched its Panther Lake CPU architecture but it also announced an update to its AI-powered performance scaling package, and it now brings XeSS on par with Nvidia's DLSS, as Arc GPU users will soon get the benefit of multi-frame generation. Exactly how soon is yet to be confirmed, and it's not clear whether XeSS 3 will have any other notable additions or improvements. However, XeSS-MFG is definitely coming, and in games that already support XeSS-FG, you'll be able to enable 3x or 4x frame generation. This can be either in the game itself or via Intel's Graphics Software application. As it so happens, Intel has sneakily been hinting at multi-frame generation for several months, but it's only very recently that this was noticed. Hidden away in a config file was a single line titled 'Multi-Frame Generation (XeSS)', but I just assumed that it was either a fairly generic label or that Intel wouldn't have anything to show for a good while. How wrong I was on both counts. So just what exactly is XeSS-MFG? Well, Intel's frame generation system works very much like Nvidia's does. The GPU renders a frame as normal and then stores it in VRAM. Once the next successive frame has been finished, the two frames are then processed through an AI algorithm to generate an in-between frame. Once that's done (and it's much quicker to do than rendering), the previous frame is then scheduled for presentation on the monitor, followed by the generated frame and finally, the one that's just been rendered. From the perspective of the user, the end result is a higher rate of frames appearing on screen, giving the whole shebang a smoother appearance. Frame generation doesn't always make things feel smoother, though, because the actual rendered frames are delayed from appearing on screen in order to generate the additional one. And in the case of Intel's new XeSS-MFG, where one, two, or even free extra frames can be created, the delay can be big enough to be noticeable in terms of input lag. It all comes down to how quickly the 'normal' frames can be rendered. The faster that process is, the less noticeable the presentation delay will be, and the lower the input lag will be. Hence why frame generation is generally recommended to only be used with upscaling and when the base frame rate is above a certain level (eg, 60 fps). Just as with Nvidia's GeForce RTX 50-series, Intel doesn't use any kind of hardware optical flow system as part of the frame generation process. DLSS 4 MFG and XeSS-MFG both use an AI algorithm to handle all of that. Blackwell GPUs run that on Tensor cores, whereas Alchemist and Battlemage GPUs use XMX (Xe matrix extension) units. I got to see XeSS-MFG first-hand while over in Arizona, as Intel was running a demonstration of it, using an early version of the Painkiller reboot on a Panther Lake laptop. The 12 Xe core GPU in that PC ran the game at around 45 to 50 fps at 1080p native, and enabling 4x multi-frame generation pushed that to around 200 fps. And just like DLSS 4, it actually felt perfectly fine to use. I tried my best to look for visual artefacts, but nothing really stood out as being a problem. The same is true of input lag: I managed to get a quick check of native vs MFG, and I honestly couldn't feel any difference between the two. However, such short experiences under tightly controlled conditions hardly make for compelling evidence. That said, my experiences of using XeSS normally have always been very positive, so I'm confident that the brief blast with Painkiller and a Panther Lake iGPU bodes well for XeSS-MFG on a full-blown Arc A750 or B580. Compared to AMD's FSR and Nvidia's DLSS, XeSS is nowhere near as supported by game developers, which is a real shame considering how good it is. The reason why is simple: Intel's share of the discrete graphics card market is minuscule, so if developers only have enough time to properly implement one or two upscaling/frame generation packages, it makes sense to choose those that the majority of gamers are going to use. As things currently stand, XeSS-MFG is only for Intel Arc GPUs, and when I asked Intel if it had plans for extending it to other GPUs (just as it has done with XeSS upscaling), I just got a repetition of 'XeSS-MFG is only for Intel Arc GPUs'. I dare say that if Intel did make it widely accessible, it wouldn't do anything for its graphics card sales, but it would at least ensure all the hard work behind the development of XeSS would find its way to a bigger audience. Whatever happens in the future with XeSS-MFG, it now just leaves AMD as the only major GPU vendor without a multi-frame generation technology. It doesn't really matter whether one needs it or not -- it's about showing the world that you're at the cutting edge of graphics tech, and right now, it has to be said that AMD's FSR is currently in third place behind DLSS and XeSS.

Intel XeSS 3 MFG "Multi-Frame Gen" With Up To 4x Mode Unveiled, Coming To All Arc GPUs With XMX Cores, XeSS 2 Games Supported & Further Perf/Efficiency Optimizations For Arc

Intel is bringing MFG "Multi-Frame Generation" support to XeSS 3 across all Arc GPUs with XMX alongside various software updates. Intel's XeSS software has seen tremendous improvements since launch. Even at launch, the upscaling technology was comparable to NVIDIA's DLSS and vastly ahead of AMD's FSR. All three vendors have done an impeccable job in fine-tuning their upscaling algorithms through AI and Machine Learning. This year, we are once again seeing some major updates to upscaling technologies. NVIDIA was the first to offer the biggest upgrade in the form of MFG support in DLSS 4, which allows the latest RTX 40 GPUs to generate up to four frames, boosting FPS for smoother gameplay. Soon after NVIDIA, AMD made its own big announcement with FSR Redstone, which will bring another leap in upscaling and frame-gen quality through Machine Learning. Intel did launch XeSS 2 last year with Frame-Gen support and further tweaks to its image quality, but this year, XeSS 3 is launching with further improvements and support for a big and highly anticipated feature. So today, Intel is announcing its XeSS 3 upscaling technology with Multi-Frame Generation support, or MFG in short. This makes Intel the second graphics vendor to provide MFG support on its GPUs, with the first being NVIDIA, who have the tech already out and featured in multiple games. Like NVIDIA, Intel's MFG works by taking two prior rendered frames, generating an optical flow network using motion vectors and depth buffers, and then, they are going to generate up to 4 frames between the two interpolated frames. So, a short brief on how frame generation works. Two frames are rasterized by the GPU, and when those two come together, you get a generated frame. So you're looking into the future, and looking in the past, pulling those two frames together and generating an optical flow with those frames being synthesized interstitially. Then, Frame Pacing is used on the output so that those frames show up at the right time on the display. With MFG, Intel simply ups the output by generating 3 additional frames besides the one that was used as the input, so up to 4x the frame output at the user end. You can see that XeSS MFG adds three blobs instead of one to the Optical Flow, following the Super Res. Intel uses one optical flow calculation to interpolate three times, which produces great-looking frames & smoother gameplay. Intel showcased game demos, which presented two FPS metrics: an FPS-App (Average), which is what the application is running at, and an FPS-Presents (Average), which is the FPS that the user will be able to see. The two demos include Dying Light: The Beast and Painkiller; both titles will be getting the XeSS 3 MFG update around Panther Lake's launch. Now the games were running internally at 30-60 FPS, but with MFG enabled, the actual FPS is boosted to 130-240 FPS, providing a much smoother experience without the tearing and artifacts that you get with lower FPS. While the XeSS 3 MFG announcement is huge in itself, the other most important thing that Intel stated is that XeSS 3 with MFG will be supported across all Arc GPUs and products with built-in XMX hardware. As per Intel, XeSS MFG will be introduced with the Panther Lake, but is also aiming to add launch support for other Arc GPUs, such as the ones on the Core Ultra 200V "Xe2" and Arc B-Series dGPUs "Xe2". The older Xe1 GPUs will also receive XeSS 3 MFG support later on, and these include Core Ultra 200H and the Arc A-series dGPUs. This would make Intel the first graphics vendor to offer Multi-Frame Generation Support not only on its latest GPUs but also on older hardware. NVIDIA's MFG support is so far limited to RTX 50 GPUs, and AMD has not made any MFG announcements so far. On the software side, Intel's Graphics Software Application will be getting two new options for users to configure. The first of these options is the XeSS Frame Generation Override, which lets users set the MFG mode between 2x, 3x, and 4x. Intel also hints at MFG expanding further with additional generated frames in the future, kinda like what NVIDIA has also hinted at with up to 16x MFG modes. Another great thing about XeSS 3 MFG is that the API to this remains unchanged, so all games that support XeSS 2 will support XeSS 3 when Panther Lake is released. Currently, there are close to 50 games that support XeSS 2, so we can expect a good range of games with XeSS 3 MFG when the technology hits shelves. The other option is Shared GPU/NPU Memory Override, which is similar to AMD's shared memory option that allows the iGPU to share system memory. Users will be able to allocate a portion of the system RAM to the GPU/NPU, allowing faster performance and also enabling better game support in titles that require more than 8 GB VRAM. Precompiling Shaders For Faster Loading Times & Reduced Stuttering Recently, we talked about how Microsoft's DirectX "AgilitySDK" added a new feature called Advanced Shader Delivery. Intel is one of the GPU vendors that is working with Microsoft on enabling this tech & starting Panther Lake, the company will enable Precompiled Shader Distribution, which helps reduce the load/launch times in games, reduces stuttering on 1st launch, and offers an automatic shader cache update process. So how Intel achieves this is by having a cloud that looks at games and precompiles the shaders ahead of launch, and then puts them back again in the cloud. So Intel's Graphics Software App will look at the games you play and then automatically download pre-compiled shaders for a list of games so that you don't have to go through the cumbersome process of pre-compiling shaders for each game. This option can be disabled and enabled manually, so you're free to enjoy a stuttery gaming experience by keeping it disabled if you want to (/s). The feature will also push updates as shaders, game, or driver changes over time. PresentMon Gets Upgrade To Support MFG PresentMon is also getting an upgrade to feature support for MFG (Multi-Frame Generation). As noted above, the application will now present three different FPS metrics, separating the naturally rendered frames from generated frames. Besides this, the software will also get Animation error telemetry and percentiles for all metrics. Intel is also enabling better optimizations aimed at gaming power and performance across its low-power SoCs. Previously, Intel's Micro algorithms were optimizing power delivery with a lack of app awareness, so they didn't have an idea if the app was a game or not. This led to lower performance and stuttering. The company addressed most of these issues through a new feature called Intelligent Bias Control V2, and we first saw its implementation in Q2 2025 for the MSI Claw (Lunar Lake), which delivered substantial gains, up to 10% in average FPS and up to 25% in percentile FPS. This is done by looking at a couple of GPU heuristics within the graphics driver, such as whether the app is a game or not, the GPU utilization, and the driver then makes hints to the GPU. Those hints are then passed to Microsoft (OS) to adjust the scheduling and use of the cores. All of this reduces stutter and improves performance as seen in the chart below: All Lunar Lake SoCs are taking advantage of this feature, but with Panther Lake, Intel is already working on Intelligent Bias Control v3. Now, in addition to the GPU Heuristics and the application environment, there's a PID controller, which is a smoother algorithm for shifting between CPU and GPU power, and it now does E-Core first scheduling. This process provides the GPU with more power headroom and delivers a smoother experience and optimized power delivery compared to the previous-gen algorithm.