You are viewing the course site for a past offering of this course. The current offering may be found here.
Lecture 23: How GPUs work (48)
kevkang

For more information on how CUDA is different from SSE/AVX vectorization: http://bitmagic.io/bmcudasse2.html

arjunsrinivasan1997

Could someone explain what exactly SIMT means and how it differs from the ideas we have been talking about?

kingdish

In addition to the number of processors, is there any other major differences among different generations of GPUs? Specifically, what are the differences among various GPU architectures (Turing, Maxwell, etc.)?

JerryJiehanWang

https://devblogs.microsoft.com/directx/announcing-microsoft-directx-raytracing/ Here's some information about how does real time ray tracing work in microsoft directx

yzyz

SIMT stands for "single instruction, multiple threads". From my understanding this is SIMD with multithreading, which allows for things like branches.

eliot1019

Newer RTX series of GPUs (after the 1000 series) also contain RT cores which are used for real time ray tracing. This article from nvidia goes over it on a high level, pretty cool how we can understand most of it now: https://www.nvidia.com/en-us/geforce/news/geforce-gtx-ray-tracing-coming-soon/

GKohavi

Nvidia also has tensor cores for deep learning acceleration. It would be interesting to see how these cores differ/specialize than the normal cuda cores.

kingdish

Here’s a comparison of ray tracing abilities between RTX series and GTX series. We can see that even RTX2060 can far exceed GTX1080Ti in terms of raytracing.

You must be enrolled in the course to comment