Intel XeSS 3 multi frame generation now rolling out to Arc GPUs

DragonSlayer101

Posts: 979   +14
Staff
TL;DR: Intel unveiled XeSS 3, its third-gen AI-powered upscaling technology last October at the Tech Tour event. Now, the company is beginning to roll it out through its latest Windows graphics driver, bringing support to Arc Alchemist and Battlemage GPUs, along with Meteor Lake, Lunar Lake, Arrow Lake-S, Arrow Lake-H, and the latest Panther Lake processors.

Included in driver package version 32.0.101.8425/32.0.101.8362 WHQL, XeSS 3 is Intel's new multi-frame generation (MFG) approach to upscaling. The idea is not new (nor are we convinced we like it) but it does remain ambitious: insert up to three AI-generated frames after every traditionally rendered frame, smoothing out animation without increasing the game's native rendering workload.

Intel says XeSS 3 relies on an optical flow network built from motion vectors and depth buffers to generate those extra frames. But unlike some competing solutions, it performs only a single optical flow calculation pass for each batch of AI-produced frames – a design choice Intel admits made development significantly more complex and time-consuming.

In concept, XeSS 3 sits in the same space as Nvidia's DLSS, which also uses AI techniques to boost frame rates in supported games while preserving image quality. The difference is in the implementation: while DLSS 4.x on RTX 50 series cards taps into hardware Flip Metering, Intel's XeSS 3 remains a fully software-driven solution.

AMD, for its part, has also entered the multi-frame interpolation race with FSR Redstone. But the company has been more cautious about deploying the feature broadly, even holding it back from premium GPUs like the Radeon RX 9070.

According to Ryzen chief VP David McAfee, AMD's hesitation comes down to tradeoffs – multi-frame interpolation can introduce additional latency and reduce responsiveness as the system waits for "fake" frames to be generated.

Intel previously confirmed that multi-frame generation can be managed through the Frame Generation Override option in its Graphics Software application for most games, though support ultimately depends on game developers. Users will be able to choose between 2×, 3×, or 4× modes, or simply leave it set to "Auto" and let the software decide the optimal setting.

Beyond XeSS 3, the driver update also addresses several bugs, including one that caused the Pragmata Sketchbook demo to crash under specific conditions on Arc B-series discrete GPUs and Core Ultra Series 2 CPUs with Arc integrated graphics. Intel also fixed incorrect Variable Refresh Rate reporting in the Display section of its graphics software.

Permalink to story:

 
I've watched a number of frame interpolation videos comparing the modes and while 2X FG looks pretty convincing and works well enough as long as you're getting over 120fps out the other end (60+ fps real frames), the 4x modes are full of weird artifacts. 4x FG is a feature in name only but there's no way I'd want to play any game in that distracting uncanny valley mess. And now Nvidia goes 6x with their most recent feature update.

4x+ FrameGen is a paper-only feature. Sure, AMD should probably deliver it so it looks like they're "keeping up" but it's a crap feature. 2x FG max. Maybe in a few years everyone will figure out 4xFG like DLSS and FSR did but it's useless until then.
 
Frame generation is good, when it works well, that is.
Future for gaming is upscaling and frame gen. Nvidia knows it. AMD knows it. Intel knows it. Game developers knows it.
People that complain, simply never tried good upscaling / frame gen. Implementation/tech matter.

Lets say you have a 240 Hz monitor.
You get 100-120 fps. Looks fine. Pretty smooth.
You enable DLSS FG, X2 mode, now you get 200-240 fps, and your monitor is now being peaked.

Result? Smoothness increased. Yes, it actually does. Vastly smoother looking and the input lag is miniscule and pretty much non-existing especially with DLSS FG/MFG.

It is a nobrainer for many users, especially in demanding AAA single player games. Some even like it for multiplayer, due to the smoothness increase *if base fps is already good*

DLSS FG is not simple frame interpolation like FSR FG or Loseless Scaling but many are gladly using these anyway.

People simply don't understand that Frame Generation is actually good, unless your base fps is crap to begin with. Then it won't be good. Also, that FG tech needs to be good as well, like DLSS FG/MFG. Nvidia is years ahead of most other FG tech, which are just frame interpolation like TVs are doing, melting frames together. Not the same.

Bottom line: Frame generation don't save crappy performance. It improves on good performance. I need at least 60 fps, but rather 80-100 fps, before I use FG and then it looks smoother for sure.

FG (X2) also lowers 1% lows by 100%, which is crucial for smoothness.

Just like people doubted upscaling, they will doubt frame gen. In some years, most will gladly use it. Many already are using it, and used it for years.
 
So, since it's software-based, does that mean any GPU can use it, not just Arc cards? Similar to how XeSS can be used on non-Arc cards?
 
Back