Meet XeSS, Intel’s answer to DLSS – AI SuperSampling on Xe HPG

Meet XeSS, Intel's answer to DLSS - AI SuperSampling on  Xe HPG

Meet XeSS, Intel’s answer to DLSS – AI SuperSampling on Xe HPG

At their 2021 Architecture Day, Intel has revealed their XeSS technology; an XMX AI accelerated Super Sampling technology that can enable high performance, high fidelity gaming on their upcoming Xe HPG “Alchemist” graphics cards.

Intel’s XeSS technology utilises beep learning to synthesize images that are close to the quality of natively high-resolution rendering. Better still, this technology will not be Intel-exclusive, allowing graphics hardware from Intel’s competitors to benefit from XeSS. 

Why do we need Image upscaling/reconstruction technologies?

Rendering games at high resolutions/refresh rates is hard. Every bump in resolution requires a dramatic increase in GPU performance, so much so that game developers and hardware manufacturers are working on clever ways to deliver higher-resolution gameplay without the insane hardware requirements of native resolution rendering. 

As gaming continues to enter the 4K era and slowly move to ever-higher screen resolutions, reconstruction and AI enhancement techniques are quickly growing in popularity as a method for game developers and hardware manufacturers to deliver ultra-high resolution gameplay experiences. 

Nvidia has created Deep Learning Super Sampling (DLSS), AMD has FidelityFX Super Resolution (FSR), and Intel has XeSS (Xe Super Sampling). All of these solutions are designed to do the same thing, deliver higher resolution gaming experiences at a lower performance cost. 

There is an image quality/performance trade-off when gaming, and with XeSS, Intel wants to deliver the best of both worlds. 

Open Source plans? 

While Intel plans to release early iterations of XeSS to developers and engine makers for direct software integrations, the company has confirmed that they plan to open-source XeSS when the technology matures. 

Intel plans to release an XMX compatible SDK for XeSS this month and a more broadly compatible version of DP4a instruction support later this year. 

Meet XeSS, Intel's answer to DLSS - AI SuperSampling on  Xe HPG  

With XeSS, Intel can reconstruct images using a neural network to deliver high-performance gaming with high levels of image quality. In Intel Xe graphics cards, these calculations can be done using XMX AI instructions, while DP4a instructions can be used on competing hardware. Intel promises up to a 2x performance boost and comparable levels of image quality when using XeSS. 

Sadly, we were not given access to Intel’s high-resolution video demo for XeSS, so we cannot comment on the image quality of Intel’s upscaling solution. Intel has confirmed that they plan to release an SDK for XeSS later this month with an XMX implementation. A more broadly compatible version with DP4a instruction support will be launched later this year. 

Meet XeSS, Intel's answer to DLSS - AI SuperSampling on  Xe HPG  

Performance-wise, Intel claims that they can deliver gamers an up to 2x increase in framerate at 4K while achieving comparable to native levels of image quality. XeSS will perform best with hardware that supports XMX instructions, though DP4a compatible hardware will still be able to achieve significant performance boosts. 

Intel has stated that DP4a instructions are supported by a “broad set of hardware, ” including integrated graphics. 

Meet XeSS, Intel's answer to DLSS - AI SuperSampling on  Xe HPG 

You can join the discussion on Intel’s XeSS SuperSampling technology on the OC3D Forums. 

Meet XeSS, Intel's answer to DLSS - AI SuperSampling on  Xe HPG