AWS Thinkbox Discussion Forums

Multiple workers or multiple concurrent tasks degrade performance

Hi. We have a new system. 64 core Threadripper 7985, 256gb ram. and 2x 4090 cards. It tears through renders, but if I spin up 2 workers, or even assign it more than 1 task (concurrently) it behaves like a 10 year old PC. It will kinda get thorugh the renders, but at 2x 4x 10x the time. So there is no benefit.

Now when its doing 1 render, it maybe uses 1%-2% GPU and about 5% CPU… it just seems like a waste of this machine when I can’t maximize the render performance.

for the dual workers I did split CPU and GPU affinity between the workers. But in the worker reports it shows each worker using both GPU.

What has me worried is we’re building an 8x GPU box and if this is the types of gains we’d see, its not worth it.

is there somethign I’m missing? I want to see 100% CPU and 100% GPU however I can get there.

forgot to mention this is C4D w/ Redshift renders

thanks
s

playing with the GPU affinity settings in the workers and the Job I was able to get different tasks to only use 1 GPU, but they only use the SAME GPU, no matter the settings. GPU 0. :frowning:

Privacy | Site terms | Cookie preferences