Intel Graphics has shared its Q3-2023 update for the Arc GPU Graphics drivers, set for an upcoming launch.
This underlines Intel's efforts in GPU performance optimization and introduces a new performance metric termed "GPU Busy."
Furthermore, the company has emphasized reducing frame-times and has devised a new approach to gauge the GPU's role in frame generation. This methodology empowers users to determine if their system's bottleneck is the CPU or the GPU during diverse operations. Originally, Intel's Arc "Alchemist" GPU architecture was tailored for DirectX 12 and Vulkan, depending on API translation for DirectX 9 games. In prior Spring driver updates, Intel documented a 43% performance improvement in DirectX 9 games through per-game optimization. The current release furthers these enhancements, providing roughly a 19% performance uptick in DirectX 11 titles at 1080p. It's worth noting that these increments are based on tests with an A750 GPU and a Core i5-13400F processor, while averaging about 12% on the i9-13900K processor. This variation stems from the specialized adjustments made to Intel's DirectX 11 driver.
The "GPU Busy" metric has been developed to break down the GPU's processing duration for API requests from the CPU. Although game rendering involves both the CPU and GPU, Intel's metric highlights the GPU-exclusive processing time for each API request. This data is pivotal for grasping how the updated drivers can trim overall frame-time, facilitating more fluid gameplay. A frame-time graph analysis of "Overwatch 2", comparing the launch drivers and the recent update, displays marked enhancements. There's a noticeable decrease in frame-time fluctuations, improved GPU Busy time consistency, and more fluid performance, especially when evaluated on a Core i5-13400F-operated system equipped with an A750 GPU pre and post-update.
With the inception of the GPU Busy metric, Intel has introduced a separate overlay application, PresentMon. This utility allows users to capture and scrutinize GPU Busy metrics, assisting in evaluating frame-time optimization.
