I’m using rtx2080 with 8gb’s memory on windows 10, but I can only run style.py when batch size is 1, when I make it to 2 or a bit more, it won’t start and just stay at the cudnn version(maybe but I didn’t remember clearly), when it comes to about 8-12, the memory is not enough. Also, when batch size is 1 and it’s running, I see that the gpu usage is only about 5% and it runs quite slowly. When batch size is higher, there is only a high peak of about 100% then the gpu doesn’t work. There’s an interesting thing that when I run two at a time, the usage jumps frequently from 90% or higher to about 40%. The cuda version is 11.7 and tensorflow-gpu verison is 2.9.0. Sorry that I forget the python version and cudnn version.
I’m using rtx2080 with 8gb’s memory on windows 10, but I can only run style.py when batch size is 1, when I make it to 2 or a bit more, it won’t start and just stay at the cudnn version(maybe but I didn’t remember clearly), when it comes to about 8-12, the memory is not enough. Also, when batch size is 1 and it’s running, I see that the gpu usage is only about 5% and it runs quite slowly. When batch size is higher, there is only a high peak of about 100% then the gpu doesn’t work. There’s an interesting thing that when I run two at a time, the usage jumps frequently from 90% or higher to about 40%. The cuda version is 11.7 and tensorflow-gpu verison is 2.9.0. Sorry that I forget the python version and cudnn version.