SlipStream: Enabling HBF for GPU Systems by Mitigating the Latency and SRAM Overheads

Anish Saxena

Hritvik Taneja

Christos Kozyrakis Stanford

Aamer Jaleel

International Symposium on High-Performance Computer Architecture (HPCA), 2027


Abstract

SlipStream enables high-bandwidth flash for LLM inference through software-hardware co-designed prefetching. Profiled weight-page access streams and demand-paced hardware tracking hide flash latency while reducing tracking and SRAM overheads.