Microsoft Building AI Upscaler to Compete with DLSS, FidelityFX

News
By
published

Microsoft plans to build a DirectX-based solution.

Microsoft
(Image credit: Microsoft)

Microsoft has begun to hire software engineers to develop machine learning-based upscaling algorithms implemented using its DirectX/DirectML application programming interface (API), which is designed for Xbox and Windows systems. The technology will compete both against AMD's FidelityFX Super Resolution and Nvidia's DLSS. 

The software giant recently posted two job openings one for a Senior Software Engineer and another for a Principal Software Engineer for Graphics seeking graphics specialists who will work on the new technology in Redmond, Washington (via TechPowerUp). The openings were posted on June 18 and June 30, just days after AMD officially started to support its FidelityFX Super Resolution upscaling technology on Microsoft's Windows platform. Since the openings were published last month, it looks like the technology is in its early stages of development. 

There are two competing upscaling technologies for games these days: AMD's FidelityFX Super Resolution (FSR) and Nvidia's deep learning super sampling (DLSS). Both technologies have to be implemented by game developers, but AMD's FSR works on virtually any GPU (including those from Intel and Nvidia) and does not use any proprietary machine learning techniques, whereas Nvidia's DLSS requires the company's GPUs to work. For the sake of truth, it should be noted that Nvidia is not the only company offering a proprietary AI-based upscaling technology. Samsung has been offering one with its high-end televisions for a while. 

As it turns out, Microsoft wants to have the best of both worlds. It wants to use machine learning-based upscaling, implemented using its industry-standard DirectX API that is widely supported by hardware from different vendors.  

Creating universal ML-based upscaling method makes great sense for Microsoft, which has for years focused on providing a consistent gaming experience on Windows no matter which hardware a user has. For hardware developers, a Microsoft-developed technology means that they will know which features to support and how to support them in a bid to provide competitive performance on all of the best graphics cards and even those that don't quite make the cut.  

One of Microsoft's job descriptions reads: 

"Xbox is leveraging machine learning to make traditional rendering algorithms more efficient and to provide better alternatives. The Xbox graphics team is seeking an engineer who will implement machine learning algorithms in graphics software to delight millions of gamers. Work closely with partners to develop software for future machine learning hardware. Make a big impact in computer graphics and gaming."

See more GPUs News
Anton Shilov
Anton Shilov
Contributing Writer

Anton Shilov is a contributing writer at Tom’s Hardware. Over the past couple of decades, he has covered everything from CPUs and GPUs to supercomputers and from modern process technologies and latest fab tools to high-tech industry trends.

5 Comments Comment from the forums
  • hotaru.hino
    An API based solution would certainly be a welcome change. But there's the question of whether it gets added for free on games already using the API and we can enable something globally or it still has to be explicitly included in games.

    But if Microsoft is doing an API wide solution, that just leaves the question of how Khronos will respond. But I guess AMD could just give them FSR and be done with it.
    Reply
  • JarredWaltonGPU
    hotaru.hino said:
    An API based solution would certainly be a welcome change. But there's the question of whether it gets added for free on games already using the API and we can enable something globally or it still has to be explicitly included in games.

    But if Microsoft is doing an API wide solution, that just leaves the question of how Khronos will respond. But I guess AMD could just give them FSR and be done with it.
    It's almost certain to require per-game implementation. At the very least, you need the UI elements to enable/disable the upscaling. Having it outside of the games would drastically limit its capabilities. Only the game knows what things benefit (in performance) from upscaling, vs. the HUD overlays and text that invariably look worse when upscaled.
    Reply
  • Alvar "Miles" Udell
    New feature in DirectX 13, requiring Windows 11 and a GPU with DirectX 13 support, which no current GPUs can support no doubt...

    I'll stick with DLSS.
    Reply
  • deesider
    If integrated into DirectX, it could be used by Developers as simply part of the quality settings with requiring the user to specifically choose it. So just part of the spectrum from low to high quality. Although the same could be achieved with AMD's FSR without needing a new DirectX. It would really be no different to what PS4 and Xbox ONE were already doing to reach 1080p.
    Reply
  • renz496
    hotaru.hino said:
    An API based solution would certainly be a welcome change. But there's the question of whether it gets added for free on games already using the API and we can enable something globally or it still has to be explicitly included in games.

    But if Microsoft is doing an API wide solution, that just leaves the question of how Khronos will respond. But I guess AMD could just give them FSR and be done with it.

    since it involved AI i doubt i can be enabled just like that globally. also even if AI upscaling being standardized though common API such as Direct X or Vulkan it doesn't mean it will encourage developer to use the feature. just look what happen with DX12 mGPU. i dare t o bet developer would rather use existing non AI upscaler if using AI based upscaler would mean they need to spend more resource to train the data themselves.

    about khronos i doubt AMD can just give their FSR unless AMD intend to update FSR to use AI since this is AI based upscaling. and nvidia most likely try to influence khronos as well. for example khronos can develop a universal AI upscaling for vulkan but those feature also build to utilize their tensor core as well.
    Reply