A5022388993- Low-power high-performance VLSI design
- Parallel Computing and Optimization Techniques
- Semiconductor materials and devices
- Stochastic Gradient Optimization Techniques
Meta (United States)
2023
Meta has traditionally relied on using CPU-based servers for running inference workloads, specifically Deep Learning Recommendation Models (DLRM), but the increasing compute and memory requirements of these models have pushed company towards specialized solutions such as GPUs or other hardware accelerators. This paper describes company's effort in constructing its first silicon designed recommendation systems; it accelerator architecture platform design, software stack enabling optimizing...
This fourth-generation processor combines two enhanced third-generation cores using an advanced 90nm dual-V/sub t/ dual-gate-oxide technology. Hardware additions feature expanded caches and inclusion of a 2 MB level cache 3 tag. The chip operates at 1.8GHz while dissipating <100W 1.1V.