Is Functional HPC the Key to Low Carbon Computing?
summary
Computing energy could contribute nearly 40% of CO2 by 2040 so functional HPC aims to cut waste by raising utilization and producing fewer churned devices, but it requires many concrete techniques: a core SAC-style language with n-dimensional homogeneous arrays, a specialization oracle to avoid code explosion, with-loop fusion plus scalarization and diffusion for eliminating intermediates, and bimodal reference counting to avoid locks on parallel threads. Implementations reach competitive sequential and multicore performance, GPUs force different kernel and transfer trade-offs, and widespread adoption needs years of cross-project comparison.
AI-written summary · about
chapters (29)
- 0:00Introduction
- 2:50What is Low Carbon Computing
- 3:06Energy Consumption
- 4:34Emissions Gap
- 5:28Renewables
- 6:49Computing Emissions
- 8:05Hardware Needs to Change
- 9:38Functional HPC
- 14:48Choices
- 17:52No Higher Order Functions
- 20:52Homogeneous Nesting
- 27:28Generality
- 28:29Intersection Types
- 29:40The Beast
- 31:42Pandoras Box
- 32:07Specialization Oracle
- 34:08Conclusion
- 35:35Index Vectors
- 36:58With Loop
- 38:50Transformations
- 41:39Diffusion
- 42:46Generating Code
- 43:46MultiThreading
- 47:48Temporal Locality
- 50:53Multicore
- 53:19Reference Count
- 54:32Architectures
- 55:09Conclusions
- 59:39The scary type
related talks
21mHPC Green Computing: Relationships Between Throughput and Latency
42mBuilding a green supercomputer in the cloud
37mHigh Performance Computing: The Modern Approach
58mBuilding a green supercomputer in the cloud
47mEnergy Proportional Edge Computing Infrastructure
32mFighting Climate Change by Building Sustainable Software
46mMaking Software & Data Architectures Sustainable
32mImproving High Throughput Computing’s Energy Efficiency - a Measurement-Based Case Study