Multi-process usage #68
Replies: 1 comment 1 reply
-
|
You mean from the time point of view? Definitely from multiple processes (or using multiple threads + CUDA streams) - the usage of GPU is more efficient because computations and memory transfers can interleave. |
Beta Was this translation helpful? Give feedback.
-
|
This seems nice that I/O and computation can overlap, but what is the CPU doing in the mean time? When I write an endless loop, doing nothing more than gpujpeg_decoder_decode followed_ by gpujpeg_encode_encode, I see it can compress/decompress 75 times per second with my 1728*3072 images on my machine. But the CPU has also one core running 100% usage. I assume transfer is DMA, so what is the CPU doing? |
Beta Was this translation helpful? Give feedback.
-
What is more efficient: doing all images sequentially in one process, or doing multiple processes simultaneously their own set of images? Can multiple compressions overlap in time?
Beta Was this translation helpful? Give feedback.
All reactions