AMD plans for FSR4 to be fully AI-based — designed to improve quality and maximize power efficiency

Asus ROG Ally X
(Image credit: Tom's Hardware)

We spoke at length with AMD's Jack Huynh, senior vice president and general manager of the Computing and Graphics Business Group, at IFA 2024. Along with deprioritizing flagship gaming GPUs and announcing the intent to create a unified UDNA GPU architecture to take on Nvidia's CUDA, the final major topic that he talked about is FSR4, FidelityFX Super Resolution 4.0. What's particularly interesting is that FSR4 will move to being fully AI-based, and it has already been in development for nearly a year.

The comments came as Huynh was talking about handheld gaming and battery life, but based on AMD's history of open-source, architecture-agnostic algorithms, FSR4 should be for more than just handhelds. We've edited and cleaned up the text from our recording, and here's what was said.

Jack Huynh: On the handheld side, my number one priority is battery life. If you look at the ASUS ROG Ally or the Lenovo Legion Go, it’s just that the battery life is not there. I need multiple hours. I need to play a Wukong for three hours, not 60 minutes. This is where frame generation and interpolation [come in], so this is the FSR4 that we're adding.

Because FSR2 and FSR3 were analytical based generation. It was filter based. Now, we did that because we wanted something with a very fast time to market. What I told the team was, "Guys, that's not where the future is going." So we completely pivoted the team about 9-12 months ago to go AI based.

So now we're going AI-based frame generation, frame interpolation, and the idea is increased efficiency to maximize battery life. And then we could lock the frames per second, maybe it's 30 frames per second, or 35. My number one goal right now is to maximize battery life. I think that's the biggest complaint. I read the returns too from the retailer, where people want to be able to play these games. [End quote]

It's interesting that the initial context for FSR4 seems to completely ignore how it might be used on non-handheld devices. And even then, it's about battery life rather than performance. Does that mean FSR4 will specifically need some of the features like the NPU on the latest Strix Point processors? We don't know, and we've reached out to AMD for clarification. But we suspect that's not the case.

As noted above, AMD's history with FSR is to support a wide range of GPU solutions. Going AI-based doesn't inherently preclude the use of GPUs either, as any recent GPU from the past six years at least can run both FP16 and DP4a (INT8) instructions that overlap with things like tensor cores and NPU instructions.

Intel has already taken a similar approach with XeSS, where it can run in XMX mode on Arc GPUs that have the Xe Matrix eXtensions. And for GPUs that don't have XMX support — like everything from AMD and Nvidia, as well as older Intel graphics solutions plus the integrated Arc Graphics in Meteor Lake processors — XeSS can use DP4a instructions to do similar work. The total compute available tends to be lower with DP4a, so it's not the exact same model being used, but more recent versions like XeSS 1.3 have focused on improving the quality and performance of the DP4a code path.

Plenty of questions remain with FSR4, like when it will even become available. If it's already been in development for 9–12 months, it could be very nearly ready for release. But as we've seen with past upscaling solutions, including DLSS and XeSs alongside FSR 1/2/3, releasing the API merely represents the first step. Getting games to support the new API takes longer.
AMD also just updated its Fluid Motion Frames 2 (AFMF 2) technology, which provides a driver-level solution for frame generation. That's probably the near-term solution for handheld gaming devices looking to boost performance, though as with all framegen solutions that rely on interpolation, we note that it's more frame smoothing than a true performance boosting feature. 

Will that be enough, and will FSR4 usher in a truly competitive alternative to DLSS and XeSS that finally leverages AI? We'll find out when AMD releases it to the public.

Jarred Walton

Jarred Walton is a senior editor at Tom's Hardware focusing on everything GPU. He has been working as a tech journalist since 2004, writing for AnandTech, Maximum PC, and PC Gamer. From the first S3 Virge '3D decelerators' to today's GPUs, Jarred keeps up with all the latest graphics trends and is the one to ask about game performance.

With contributions from
  • Pierce2623
    Ok so it’s AI based, will rdna3’s AI focused matrix extensions allow it ton on there without discrete matrix accelerators?
    Reply
  • edzieba
    Pierce2623 said:
    Ok so it’s AI based, will rdna3’s AI focused matrix extensions allow it ton on there without discrete matrix accelerators?
    Likely not without a performance impact comparable to the performance uplift from the reduced rendering resolution, making it pretty much moot for RDNA3 and below. Hence the focus on handheld gaming devices which contain AMD chips with dedicated matrix accelerator hardware. AMD have yet to announce consumer desktop cards with such hardware, so they will likely not promote FSR4 in that market until they have announced cards that could take advantage of it.
    Reply
  • coolitic
    What a waste of hardware and dev-time, all just to peg a crummy "AI" label w/ negligible real-world benefits.
    Reply
  • ET3D
    Pierce2623 said:
    Ok so it’s AI based, will rdna3’s AI focused matrix extensions allow it ton on there without discrete matrix accelerators?
    The focus on mobile makes me think that it's going to use the NPU for upscaling.
    Reply
  • Makaveli
    edzieba said:
    Likely not without a performance impact comparable to the performance uplift from the reduced rendering resolution, making it pretty much moot for RDNA3 and below. Hence the focus on handheld gaming devices which contain AMD chips with dedicated matrix accelerator hardware. AMD have yet to announce consumer desktop cards with such hardware, so they will likely not promote FSR4 in that market until they have announced cards that could take advantage of it.
    That hardware may very well be in RDNA 4 but we will have to wait and see.
    Reply
  • usertests
    coolitic said:
    What a waste of hardware and dev-time, all just to peg a crummy "AI" label w/ negligible real-world benefits.
    Not a waste if it works and catches them up to DLSS.
    Reply
  • mikeztm
    Pierce2623 said:
    Ok so it’s AI based, will rdna3’s AI focused matrix extensions allow it ton on there without discrete matrix accelerators?
    RDNA3 does not have any discrete matrix unit as you said, so it will not be useful there.
    RDNA3's WMMA extension is almost pointless since it does not have any execution unit support.
    Reply
  • thestryker
    While Microsoft has made it simple to implement all of the upscalers I do still hope AMD has an open implementation similar to Intel's. I absolutely think it's necessary for AMD to move beyond what they have been doing as there's a definite ceiling as to how high quality it can be.
    Reply
  • DS426
    What is with everyone's obsession with thinking NPU's are needed for AI work? NPU's just make AI workloads more efficient on low power / battery-life-important devices like laptops.

    Additionally, FSR4 would have the potential of more real-world impact on a handheld, even as the desktop gamer world does demand FSR that's more competitive with DLSS. As stated in the article, let it prefer AMD hardware first and then if it's a different vendor, fall back to DP4a and FP16 as I also agree remaining hardware vendor agnostic and mostly open source is important now and going forward.
    Reply
  • usertests
    DS426 said:
    What is with everyone's obsession with thinking NPU's are needed for AI work? NPU's just make AI workloads more efficient on low power / battery-life-important devices like laptops.

    Additionally, FSR4 would have the potential of more real-world impact on a handheld, even as the desktop gamer world does demand FSR that's more competitive with DLSS. As stated in the article, let it prefer AMD hardware first and then if it's a different vendor, fall back to DP4a and FP16 as I also agree remaining hardware vendor agnostic and mostly open source is important now and going forward.
    If their focus is on laptops and handhelds, the NPU is an untapped resource that would let them leave GPU resources to games. Especially the XDNA2 NPU which is relatively powerful. But we don't know how this is going to shake out. The only details we actually know are that FSR4 will use AI, and they are focusing on mobile, efficiency, battery life.

    DLSS does a lot better than FSR at low input resolutions, so an AI-based FSR4 focusing on mobile makes a lot of sense. Crappy 540p input + AI = 1080p magic. Rendering at 540p lowers power consumption of the APU, leading to longer battery life.
    Reply