GPU Targets Deep Learning Applications

The 16-nm FinFET-based Tesla P100 (Fig. 1) is the next-generation GPU from NVidia. It targets a range of applications including deep learning delivering 12 times the performance of the previous GPU generation, NVidia’s Maxwell architecture. The Tesla P100 incorporates many new features and technologies, including High Bandwidth Memory 2 (HBM2) and Chip-on-Wafer-on-Substrate (CoWoS) technology. The Tesla P100 consists of 15.3 billion transistors. HBM2 is the next generation of stacked, on-chip memory. AMD used HBM in its latest Radeon R9 series. NVidia also added unified memory support for CPUs and GPUs that is similar to AMD’s Heterogeneous System Architecture (HSA). This allows the CPU and GPU to use the same pointer structure simplifying CPU and GPU programs. The improved memory system deliver a 3x memory bandwidth performance boost. The Tesla P100…


Link to Full Article: GPU Targets Deep Learning Applications