News for hackers. Stuff that matters.

A comment on CUDA Grep. Parent comment

Re:CUDA Grep by cma on Thursday June 27, 2019 @01:08PM (#20296354)
Some GPUs have integrated memory with the CPU and don't need to do the transfer, it could still speed things up there couldn't it?

Also, they didn't ignore the key detail, they mention it: "This is without taking into account the overhead of transferring data to the GPU."

[ Parent | on HN ]
Re:CUDA Grep by alfalfasprout on Thursday June 27, 2019 @01:09PM (#20296368)
Yeah, but integrated GPUs wouldn't be usable with CUDA.
[ Parent | on HN ]
Re:CUDA Grep by Athas on Thursday June 27, 2019 @01:36PM (#20296663)
I have not been able to inspect the code (dead links), but from the description of their approach, they do not depend on anything particularly fancy from CUDA. An OpenCL implementation seems just as feasible.
[ Parent | on HN ]
Re:CUDA Grep by cma on Thursday June 27, 2019 @01:19PM (#20296468)
Nvidia makes several with integrated memory. Nintendo Switch is based on one etc.
[ Parent | on HN ]
Re:CUDA Grep by justinmeiners on Thursday June 27, 2019 @07:51PM (#20300114)
Perhaps on PCs, but plenty of mobile and consoles use a shared memory architecture.
[ Parent | on HN ]
Re:CUDA Grep by shaklee3 on Friday June 28, 2019 @12:29AM (#20301478)
The tegra products are all integrated.
[ Parent | on HN ]
Re:CUDA Grep by bkase on Thursday June 27, 2019 @01:16PM (#20296449)
Oh good, I'm glad we did mention this!
[ Parent | on HN ]