High-performance computing on gamer PCs
summary
Implements reservoir computing with a fly-eye architecture to invert an ill-conditioned convolution using Python drivers and CUDA-accelerated linear algebra (PyCUDA/CUV and small C kernels). Combines particle-swarm optimization plus genetic crossover to search reservoirs, runs on a homemade GPU cluster (master + six workers, GTX 285/580/590 cards, up to seven GPUs in one box, 64 PCIe lanes), uses literate programming (noweb) and reproducible workflows; Cythonization yields a 11s→0.12s speedup (~100×). Temperature peaks ~90°C, 20 TB long-term storage, system works.
AI-written summary · about
chapters (12)
- 0:00Start
- 6:03Deconvolution model: Reservoir computing
- 7:17Matrix approache necessary conditions to solve the problem
- 14:58Finding the minimum
- 18:20Fly eye reservoir final layer structure
- 26:08Reproducible computational research
- 26:54The software
- 31:03The System
- 32:22The HPU4Science cluster
- 33:42The GPU cards
- 34:43The motherboard: Gigabyte GA-X58A
- 37:17Software components for non- computational cluster activity
related talks
41mLuxuries, necessities, and the challenges that remain
30mCUDA in Your Python: Effective Parallel Programming on the GPU
31m"Optimizing Scientific Python with C++, CUDA, and Serverless Compute"
34mAccelerating GPU Algorithms in pure Python
30mCuPy: A NumPy-compatible Library for GPU
35mCUDA in Python- A New Era for GPU Acceleration
02h 28mGPU Accelerated Python
40mScaling Python - Bank Edition