Energy Efficiency Boost in the AI-Infused POWER10 Processor
Authors:
- 1 June 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 29-42
Abstract
We present the novel micro-architectural features, supported by an innovative and novel pre-silicon methodology in the design of POWER10. The resulting projected energy efficiency boost over POWER9 is 2.6x at core level (for SPECint) and up to 3x at socket level. In addition, a new feature supporting inline AI acceleration was added to the POWER ISA and incorporated into the POWER10 processor core design. The resulting boost in SIMD/AI socket performance is projected to be up to 10x for FP32 and 21x for INT8 models of ResNet-50 and BERT-Large. In this paper, we describe the novel methodology deployed and used not only to obtain these efficiency boosts for traditional workloads, but also to infuse AI/ML/HPC capability directly into the POWER10 core.Keywords
This publication has 39 references indexed in Scilit:
- Robust power management in the IBM z13IBM Journal of Research and Development, 2015
- Efficient PVT independent abstraction of large IP blocks for hierarchical power analysis2013
- Accurate Fine-Grained Processor Power Proxies2012
- Counter-Based Power Modeling Methods: Top-Down vs. Bottom-UpThe Computer Journal, 2012
- Pinpointing Representative Portions of Large Intel® Itanium® Programs with Dynamic Instrumentation2005
- Picking statistically valid and early simulation points2004
- Optimizing pipelines for power and performance2003
- EDA in IBM: past, present, and futureIEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2000
- Power-aware microarchitecture: design and modeling challenges for next-generation microprocessorsIEEE Micro, 2000
- Basic Linear Algebra Subprograms for Fortran UsageACM Transactions on Mathematical Software, 1979