Design of a GPU with Heterogeneous Cores for Graphics

TL;DR

The paper addresses the intra-frame heterogeneity of graphics workloads by proposing KHEPRI, a heterogeneous GPU that pairs compute-specialized and memory-specialized shader cores with a locality- and affinity-aware tile scheduler. By predicting per-tile memory intensity from frame-to-frame coherence and preserving texture locality, KHEPRI maps tiles to the most suitable core type while maintaining cache locality. The approach yields an average 9.2% performance gain, 7.3% higher FPS, and 4.8% lower GPU energy compared to a homogeneous baseline, without hardware overhead. This work demonstrates a practical path to improved mobile GPU efficiency by exploiting within-frame workload diversity through targeted core specialization and intelligent scheduling.

Abstract

Heterogeneous architectures can deliver higher performance and energy efficiency than symmetric counterparts by using multiple architectures tuned to different types of workloads. While previous works focused on CPUs, this work extends the concept of heterogeneity to GPUs by proposing KHEPRI, a heterogeneous GPU architecture for graphics applications. Scenes in graphics applications showcase diversity, as they consist of many objects with varying levels of complexity. As a result, computational intensity and memory bandwidth requirements differ significantly across different regions of each scene. To address this variability, our proposal includes two types of cores: cores optimized for high ILP (compute-specialized) and cores that tolerate a higher number of simultaneously outstanding cache misses (memory-specialized). A key component of the proposed architecture is a novel work scheduler that dynamically assigns each part of a frame (i.e., a tile) to the most suitable core. Designing this scheduler is particularly challenging, as it must preserve data locality; otherwise, the benefits of heterogeneity may be offset by the penalty of additional cache misses. Additionally, the scheduler requires knowledge of each tile's characteristics before rendering it. For this purpose, KHEPRI leverages frame-to-frame coherence to predict the behavior of each tile based on that of the corresponding tile in the previous frame. Evaluations across a wide range of commercial animated graphics applications show that, compared to a traditional homogeneous GPU, KHEPRI achieves an average performance improvement of 9.2%, a throughput increase (frames per second) of 7.3%, and a total GPU energy reduction of 4.8%. Importantly, these benefits are achieved without any hardware overhead.

Figures (10)

• Figure 1: The Graphics Pipeline of a TBR GPU.
• Figure 2: Shader core microarchitecture.
• Figure 3: Ratio of memory requests to arithmetic instructions during the execution of a frame of Beach Buggy Racing (BBR).
• Figure 4: Architecture of KHEPRI. Each Raster Unit has its own private resources and specialized cores. Cores shaded in blue are compute-specialized, while cores shaded in red are memory-specialized.
• Figure 5: Beach Buggy Racing (BBR) game-bbr rendered frame with corresponding core type assignment maps: (b) memory-intensity-only tile assignment, (c) affinity- and locality-aware tile assignment. Tiles shaded in blue are assigned to the compute-specialized cores, while tiles shaded in red are assigned to the memory-specialized cores.