Intel shares Alchemist graphics card and Xe HPC architecture information
At the Intel Architecture Day 2021, Intel disclosed a lot of information, including Alder Lake, Sapphire Rapids, Ponte Vecchio, Xe HPC GPU, etc. Many people’s eyes are on the upcoming Alchemist graphics card (DG2), which is the first product of Intel’s new high-performance graphics product brand, Intel Arc. This graphics card will support hardware-based ray tracing and artificial intelligence-driven super sampling (XeSS), providing full support for DirectX 12 Ultimate.For a long time in the past, Intel has been limited to integrated graphics. After Raja Koduri defected from AMD to Intel, after four years of polishing, he is about to help Intel complete its ambition to return to the independent gaming graphics market. If it succeeds in the end, then the senior vice president, chief architect, and general manager of the architecture, graphics, and software department of Intel will be the top contributor. At this Architecture Day event, Intel shared many details about the Xe HPG architecture.
For a long time, Intel’s GPU basic module is EU, which will become a new Xe core (Xe Core) on the Xe HPG architecture, which contains 16 vector engines and 16 matrix engines (Xe Matrix eXtension, XMX), high-speed cache, and share internal video memory, support DirectX Raytracing (DXR) and Vulkan Ray Tracing unit. The new generation GPU will be manufactured using TSMC’s 6nm process, which is an optimized and upgraded version of the 7nm process.
What interests many people is Intel’s Deep Learning Super Sampling Technology (XeSS), which uses Alchemist’s built-in XMX AI acceleration. XeSS relies on DP4a instructions to provide AI-based super sampling on various hardware including integrated graphics. Many game developers have begun to use XeSS, Intel will also provide the initial version of the XMX SDK to independent software vendors (ISV) in the near future, and the DP4a version will be launched later this year.
Intel did not disclose too many details of XeSS technology this time, and it is generally speculated that it will be closer to Nvidia’s DLSS technology. Earlier people discovered that Anton Kaplanyan, who was responsible for ray tracing and deep learning super-sampling technology research during Nvidia’s tenure, had moved to Intel and served as vice president of graphics research and development. This expert in the graphics field laid the foundation for the future development of NVIDIA’s DLSS technology.
Intel said that after optimization, the Xe HPG architecture will have 1.5 times the performance and performance per watt than the previous Xe LP architecture. At the same time, Intel works closely with developers and strives to be consistent with industry standards in design. Drivers also cover integration in a unified code base. Discrete graphics card, the driver component has been re-architected, especially in the memory manager and compiler, which has increased the throughput of computationally intensive games by 15% and shortened the game loading time by 25%.
For many players, after experiencing a long-term supply shortage of graphics cards and substantial premiums, new choices may have new hopes. Since the introduction of HD Graphics in the Clarkdale era in 2010, Intel has been steadily increasing in the graphics field for more than ten years. At least in terms of market share, Intel has long been the world’s largest GPU supplier. After testing the waters of the Xe LP architecture, the Xe HPG architecture will have an impact on Nvidia and AMD.
In the future, the Xe HPG architecture will continue to be optimized, the second generation is Battlemage, the third generation is Celestial, and the next is the new Xe architecture Druid.
