Xe-HPG

Xe-HPG. It is a high-performance graphics microarchitecture developed by Intel for its dedicated GPUs, marking the company's major entry into the discrete gaming graphics card market. First implemented in the Intel Arc Alchemist series, it was fabricated on TSMC's N6 process node and represented a significant architectural departure from the earlier Xe-LP design used in integrated graphics.

Architecture and design

The Xe-HPG architecture was built from the ground up for scalable, high-throughput gaming performance. Its core organizing principle is the Render Slice, which groups four Vector Engines—the primary shader execution units derived from the Xe-LP's EUs—with dedicated ray tracing units, texture sampling hardware, and geometry pipelines. This modular design allowed Intel to configure different GPU dies by varying the number of Render Slices. The architecture also integrated a new AI acceleration unit called XMX (Xe Matrix Extensions), analogous to Nvidia's Tensor Cores, to accelerate XeSS upscaling and other deep learning tasks. Memory subsystem innovations included a high-bandwidth cache hierarchy and support for GDDR6 memory interfacing via a new memory controller.

Product lineup and specifications

The Xe-HPG architecture debuted in the Intel Arc Alchemist family of graphics cards, codenamed "DG2." The lineup spanned from the entry-level Arc A310 and Arc A380 to the performance-oriented Arc A750 and flagship Arc A770. Key GPU dies included the ACM-G10 used in the Arc A770 and Arc A750, featuring up to 32 Xe-Cores (the equivalent of 8 Render Slices) and 16GB of GDDR6 memory. The smaller ACM-G11 die powered the Arc A380 and other mobile parts. Specifications varied significantly across the stack, with core counts, clock speeds, and memory bus widths scaled to target different segments of the GPU market.

Performance and features

In terms of gaming performance, Xe-HPG-based cards like the Arc A770 competed closely with contemporaries such as Nvidia's GeForce RTX 3060 and AMD's Radeon RX 6600 XT in DirectX 12 and Vulkan titles. A key software feature was XeSS, an AI-driven spatial upscaling technology that leveraged the XMX units for performance enhancement. The architecture also provided hardware-accelerated ray tracing support via dedicated Ray Tracing Units within each Render Slice. Full support for modern APIs like DirectX 12 Ultimate was included, encompassing features such as mesh shaders and variable rate shading. Early performance was notably stronger in newer APIs compared to older ones like DirectX 11.

Manufacturing and process technology

Unlike Intel's prior Xe-LP graphics, which were built on the company's own Intel 7 process, Xe-HPG was manufactured externally by TSMC using its advanced N6 (6nm) node. This decision marked a strategic shift for Intel, leveraging TSMC's high-volume manufacturing expertise and established process technology to ensure competitive transistor density and power efficiency for its first major discrete gaming GPUs. The use of an external foundry was a significant aspect of Intel's broader IDM 2.0 strategy during this period.

Market positioning and competition

Xe-HPG and the Intel Arc Alchemist cards were positioned as a challenger to the established duopoly of Nvidia's GeForce RTX 30 series and AMD's Radeon RX 6000 series. Intel targeted the mainstream and enthusiast segments, with the Arc A750 and Arc A770 aimed at 1080p and 1440p gaming. Market success was challenged by intense competition, initial driver maturity issues, and a complex global launch strategy. The architecture served as a crucial learning platform for Intel, paving the way for its successor, Xe2-HPG, and establishing a foundation for future generations of Intel discrete graphics.

Category:Intel microarchitectures Category:Graphics microarchitectures Category:2022 in computing