Earlier this week, Intel announced Arc, the branding for the brand new gaming GPUs that can face off with Nvidia’s GeForce and AMD’s Radeon playing cards once they launch early in 2022. At the moment, Intel provided a few additional details on Arc, its underlying Xe-HPG structure, its graphics drivers, and the “XeSS” upscaling expertise that can work with each Intel’s GPUs in addition to GPUs from Nvidia and AMD.
The HPG in Xe-HPG stands for “high-performance gaming,” and it builds on the Xe-LP (“low-power”) graphics tech included in Intel’s 11th-generation Tiger Lake laptop processors and the low-end DG1 dedicated GPU. At a high-level, Xe-HPG will help the entire options in DirectX 12 Ultimate, identical to Nvidia’s RTX 2000- and 3000-series GPUs and AMD’s RDNA2-based Radeon RX 6000-series playing cards. This consists of, most notably, help for hardware-accelerated ray tracing, variable rate shading, and mesh shaders.
-
Intel is changing the “EU” as its base GPU {hardware} measurement with the “Xe-core,” a mixture of 16 vector engines, 16 matrix engines, some cache, and some different components.
-
4 Xe-cores mix with ray tracing and fixed-function {hardware} to kind a render slice…
Intel -
… and a render slice is mixed with extra cache and a reminiscence interface to make a GPU.
-
Alchemist chips will scale as much as eight render slices whole, although Intel wasn’t particular about what configurations it deliberate to deliver to market.
Intel
Since not less than 2006, Intel has talked about its GPU {hardware} by way of “execution items,” or EUs. The corporate is putting off that terminology for Xe-HPG, changing it as a substitute with the idea of the “Xe-core.” Every Xe-core consists of 16 vector engines and 16 matrix (or XMX) engines, in addition to L1 cache and another {hardware}. 4 Xe-cores mix with ray-tracing items and different fixed-function {hardware} to kind a “render slice,” which is the naked minimal any Xe-HPG GPU might want to operate (together with L2 cache and a reminiscence interface).
Alchemist-based chips can embrace “as much as” eight of those render slices, which means that we’ll see not less than a few totally different Arc GPUs with totally different ranges of computing energy. However Intel hasn’t disclosed any particular {hardware} configurations, and it additionally hasn’t gone into any element about clock speeds, the reminiscence interface, or the quantity or sort of RAM that the primary Arc GPUs will embrace. (Fast terminology reminder: Xe-HPG is the identify of the GPU microarchitecture, Alchemist is the codename for the chips based mostly on Xe-HPG, and Arc is the buyer branding you may see on retailer cabinets.)
Intel says that the Intel-specific XMX model of XeSS will present the most effective picture high quality and the shortest render time. However the DP4a model will get pleasure from engaged on some Nvidia and AMD GPUs, which can make it extra interesting to sport builders.
Intel
The XMX {hardware} in Xe-HPG GPUs will likely be key to Intel’s AI-accelerated upscaling expertise, additionally known as Xe Tremendous Sampling or XeSS. Like Nvidia’s DLSS, the concept is to upscale lower-resolution pictures with as little high quality loss as doable. Doing so will make it doable to play a sport at 4K resolutions with out requiring the graphical horsepower crucial to truly render a local 4K picture.
Intel says that you’re going to get the most effective picture high quality and efficiency out of XeSS in the event you’re benefiting from its XMX {hardware}, however there is a separate model of XeSS that makes use of DP4a directions widespread in fashionable GPUs from Nvidia and AMD as effectively. Intel did not name out particular GPU fashions, however DP4a directions are included way back to the Pascal structure in Nvidia’s still-popular GTX 1060 and 1050 collection playing cards. Nvidia would not truly help both of those fashions with DLSS. Built-in GPUs will have the ability to profit from XeSS as effectively. The SDK for the XMX model of XeSS will likely be obtainable this month, with the DP4a model following later this yr.
Intel talked about its upcoming GPU codenames earlier this week when it introduced the Arc branding. Alchemist is Xe-HPG, Battlemage will likely be Xe2-HPG, and Celestial will likely be Xe3-HPG. Druid’s structure would not have a reputation but—we would recommend Xe4-HPG.
Intel
Once they’re launched, Arc GPUs will use the very same driver bundle as Intel’s built-in GPUs, so enhancements made to learn one structure can doubtlessly profit all of them (Intel boasted that it had overhauled its graphics driver previously yr to enhance efficiency in CPU-bound video games operating on Xe-LP GPUs by between 15 and 80 %).
Lastly, the elephant within the room: Intel is not making Alchemist chips in its personal chip factories, not less than not now. The corporate introduced that it’ll use TSMC’s 6 nm course of for Alchemist chips, that means that Intel’s GPUs might want to battle for capability with AMD, Apple, and everybody else manufacturing cutting-edge chips at TSMC.
“The rationale [for using TSMC for Xe-HPG] is straightforward,” wrote Intel SVP Stuart Penn in an accompanying press release. “Simply as our designers use the suitable structure for the suitable workload, we additionally select the node that most closely fits that structure. At this cut-off date, these foundry nodes are the suitable alternative for our discrete graphics merchandise.”
