Concurrent HBM And Host Memory Access Improves LLM Inference Throughput (Georgia Tech, Nvidia, Stanford)
researchSemiconductor EngineeringSep 21, 2026
Researchers at Georgia Tech, Nvidia Research, and Stanford University published a technical paper titled “BOOST: Concurrent Access to Host Memory and HBM to Accelerate LLM Inference.” Abstract Excerpt: “This paper presents BOOST, the first runtime system th...
Impact: MediumConfidence: LowNovelty: Material
researchsemiconductor-engineering