Optimizing Attention on GPUs by Exploiting GPU Architectural NUMA Effects
This addresses a critical efficiency problem for AI training and inference on next-generation GPUs, though it is incremental as it builds on existing attention algorithms with a scheduling optimization.
The paper tackled the performance bottleneck caused by non-uniform memory access (NUMA) in large-scale attention workloads on disaggregated GPUs, achieving up to 50% higher performance and 80-97% L2 cache hit rates on AMD's MI300X architecture.
The rise of disaggregated AI GPUs has exposed a critical bottleneck in large-scale attention workloads: non-uniform memory access (NUMA). As multi-chiplet designs become the norm for scaling compute capabilities, memory latency and bandwidth vary sharply across compute regions, undermining the performance of traditional GPU kernel scheduling strategies that assume uniform memory access. We identify how these NUMA effects distort locality in multi-head attention (MHA) and present Swizzled Head-first Mapping, a spatially-aware scheduling strategy that aligns attention heads with GPU NUMA domains to exploit intra-chiplet cache reuse. On AMD's MI300X architecture, our method achieves up to 50% higher performance over state-of-the-art attention algorithms using conventional scheduling techniques and sustains consistently high L2 cache hit rates of 80-97%. These results demonstrate that NUMA-aware scheduling is now fundamental to achieving full efficiency on next-generation disaggregated GPUs, offering a path forward for scalable AI training and inference.