Interactive GPU programming with ClojureCUDA & ClojureCL
summary
Interactive REPL-driven GPU bindings for Clojure let users compile C kernels on the fly and target CUDA or OpenCL via ClojureCUDA/ClojureCL, integrate with the Neanderthal linear-algebra stack and a Bayesian library (Byadera). Microbenchmarks show idiomatic Clojure dot-product at 14 ms, Neanderthal CPU at ~4 μs and GPU at ~26 μs for small vectors, while a Bayesian MCMC run hits 61M samples in ~270 ms on an older AMD GPU (≈120 ms on newer NVIDIA), claiming ~1000× speedups versus Stan for full-analysis workloads.
AI-written summary · about
chapters (8)
- 0:00Start
- 1:18Data Processing
- 8:25Hello World
- 8:55Map and Reduce
- 11:39Matrix Multiplication
- 17:09Transfer Data from the Main Memory to the Gpu Memory
- 33:51Donations
- 39:23Dynamic Parallelism
related talks
25mMore Clojure Less Complication
26mThe Feynman Machine: AI architecture in Clojure/Neanderthal
31mScientific Computing with Clojure
34mPredictably Fast Clojure
38mScientific Clojure, a bird's eye view
23mDressed up performance in Clojure
40mBuilding Machine Learning Models with Clojure and Cortex
38m"Emmy: Moldable Physics and Lispy Microworlds"