Parallel Programming on an NVIDIA GPU

[Mark44]

This article is the first of a two-part series that presents two distinctly different approaches to parallel programming. In the two articles, I use different approaches to solve the same problem: finding the best-fitting line (or regression line) for a set of points.
The two different approaches to parallel programming presented in this and the following Insights article use these technologies:

Single-instruction multiple-thread (SIMT) programming is provided on the Nvidia® family of graphics processing units (GPUs). In SIMT programming, a single instruction is executed simultaneously on hundreds of microprocessors on a graphics card.
Single-instruction multiple data (SIMD) as provided on x64 processors from Intel® and AMD® (this article). In SIMD programming, a single instruction operates on wide registers that can contain vectors of numbers simultaneously.

The focus of this article is my attempt to exercise my computer’s Nvidia card using the GPU Computing Toolkit that Nvidia...

Continue reading...

 

[Mark44]

I'm very interested in hearing comments or questions about this article or the other article in this set -- https://www.physicsforums.com/threads/parallel-programming-on-a-cpu-with-avx-512.1014578/.

 

[H4UnT3R]

Thank you very much for this! I was just looking at AMD ZEN4 release and they have some support for AVX-512. But still, for AI is IMHO still better some CUDA, e.g. RTX 3060 with 3584 cores plus nVidia Rapids, which optimize their cards for max performance. We'll see when zen4 CPUs will be tested :)