"Making Elm Talk to Your Personal Supercomputer"

on YouTube ↗32m

" summary

Combines Elm front end with Futhark-generated GPU code via OpenCL and a Python bridge and demonstrates real speedups and practical workarounds: Futhark compiles to OpenCL/pyOpenCL and runs heavy linear algebra on the GPU, achieving ~90x speed-up on 1600×1600 matrix multiplication (153 ms vs 14 s C, 48 s JS). Shows a concrete pitfall — a nonassociative reducer gives wrong sums unless rewritten as map then reduce — and ships a Rube-Goldberg pipeline: Elm→HTTP→Python→Futhark→GPU that produces working PCA and heat-flow simulations.

AI-written summary · about

" chapters (21)

1. 0:00Intro
2. 0:11Elm + Futhark
3. 1:45How to get Elm to talk to a supercomputer
4. 2:43Creator of Futhark
5. 3:07OUTLINE
6. 4:49II. What is Futhark?
7. 5:47Code Snippet
8. 6:56Compiler is smart - How?
9. 7:31Futhark vs NVIDIA & AMD
10. 8:43HotSpot Benchmark
11. 9:48III. A Real World Problems
12. 10:24(b) Classify fossils
13. 11:55Data on 10,000 fossil bones
14. 15:03IV. Matrix Multiplication
15. 19:09Benchmarks
16. 22:23Futhark in Five Minutes .... (vector x Matrix)
17. 24:05Futhark is for standard humans!
18. 24:47VI. A Pitfall in Parallel Programming
19. 26:32Futhark makes assumptions
20. 27:27The bad reducer
21. 28:05VII. Communication

" related talks

• 36mBuilding a Graphical IDE in Elm (...)Claudia Doppioslash · Erlang User Conference 2016
• 19mFuthark: A High-Performance Purely Functional Array LanguageTroels Henriksen · Lambda Days 2017
• 44mFunctional Array Funhouse Intensive - Part 5 of 5Aaron W Hsu · LambdaConf 2017
• 09m0-60 in 15 Minutes: Building a Realtime App With Elm and HorizonAbadi Kurniawaan · Elm Conf 2016
• 42mClojure is Not Afraid of the GPUDragan Djuric · EuroClojure 2016
• 47mFunctional Array Funhouse Intensive - Part 4 of 5Aaron W Hsu · LambdaConf 2017
• 41mHow to Elm-ify Your MLOscar Boykin · Scale By the Bay 2017
• 02h 50mFunctional Array Funhouse Intensive - Part 3 of 5Aaron W Hsu · LambdaConf 2017