At the 2024 International Tech Expo (ITT), Intel unveiled the Xe2 architecture, the successor to its Xe “Alchemist” GPUs. Xe2 promises significant performance improvements, new features, and better power efficiency for both integrated and discrete graphics solutions. We’re particularly interested in what’s in store for the upcoming Intel Arc Battlemage discrete GPUs, and we see this as a step in the right direction.
We have all the key info for what fans and potential users can expect from this new architecture, as well as what is coming up next for Intel and Xe2. So without without further ado, let’s get started.
A new foundation for graphics
Intel has streamlined its naming scheme for Xe2. While internal chips will retain codenames, for consumers, it’ll simply be Xe2. The architecture focuses on achieving higher utilization, improved work distribution, and reduced software overhead compared to Xe. The 2nd generation Xe cores, or Xe2 cores, feature repartitioned resources for increased efficiency. Key improvements include:
- 8 512-bit Vector Engines for various operations
- 8 2048-bit XMX Engines, now supporting more data types and running significantly faster
- Support for 64b atomic operations
- Larger shared L1 cache
Improved rendering with redesigned render slices
Xe2 utilizes scalable render slices that can be stacked for increased performance. These slices offer several enhancements:
- New Geometry Engine with tripled vertex fetch throughput and mesh shading performance
- Larger L1 cache for out-of-order sampling
- Doubled throughput for sampling without filtering
- New HiZ unit with more cache and early culling capabilities
- Dual Pixel Backends with faster blending and larger color cache
Xe2 incorporates a larger Ray Tracing Unit with more traversal pipelines, box intersections, and triangle intersections compared to Xe. This translates to improved ray tracing performance.
Lunar Lake first in line for Xe2 GPU
The first chip to feature Xe2 graphics is Lunar Lake, an integrated solution for mobile SoCs. Here’s a breakdown of the Xe2 configuration in Lunar Lake:
- 8 Xe2 Cores
- 64 Vector Engines
- 2 Geometry Pipelines
- 8 Samplers
- 4 Pixel Backends
- 8 Ray Tracing Units
Benchmarks by Intel show a 50% performance increase for Lunar Lake’s Xe2 GPU compared to Meteor Lake’s Xe GPU, with lower power consumption at the same performance level. Lunar Lake also boasts an improved Display Engine with support for the latest display standards like HDMI 2.1, DisplayPort 2.1, and eDP 1.5. The new eDP 1.5 with Panel Replay promises smoother playback and better power efficiency.
The Media Engine in Lunar Lake features a dedicated side cache, reducing traffic to system memory and improving power efficiency, especially for encoding workloads. The engine also supports the latest video codecs like VVC, offering significant bitrate reduction compared to AV1.
What’s next
Intel assures that the Windows GPU software stack is ready for Xe2, with optimizations across all major APIs and frameworks. The company plans to reveal more details about its upcoming discrete Xe2-based Arc Battlemage lineup later in 2024. The Xe2 architecture represents a significant leap forward for Intel graphics. With its focus on performance, efficiency, and new features, Xe2 promises to power a new generation of powerful and versatile graphics solutions for both integrated and discrete applications.
