Megacore: Megafast, Megacool?
summary
Claim: Erlang plus a pattern-driven refactoring tool and a Scale skeleton library produce practical parallelism with concrete results. Implementation creates 331 million lightweight threads (20 seconds on a 16-core box as a stress test), yields about 60x speedup on 64 cores for an evolutionary optimizer, adds ~100 lines and one day of work versus 353 lines and seven days for manual concurrency, maps parts to GPU via ledo, and shows linear C++/MPI scaling to 33,000 cores.
AI-written summary · about
chapters (7)
- 0:00Start
- 2:17Intel Xeon Phi
- 16:01A Better Way To Build a Wall
- 17:21How Not To Build a Wall
- 23:12Pattern Discovery Tool
- 44:35The Pattern Discovery Tool
- 44:59Dynamic Scalability
related talks
24mIndustrial Strength High Performance RISC V Processors For Energy Efficient Computing
11mAccelerated Software for a Post-Moore World
01h 53mMachine Learning for Manycore System Design and Optimization
01h 07mGeneric Parallel Programming
56mAutomatically Scalable Computation
43mHigh Speed Query Execution with Accelerators and C++
14mAchieving High-Performance the Functional Way - A Functional Pearl on Expressing High...
52mC++ Standard Parallelism