2024-11-07 at 1:55 PM UTC
Originally posted by Elbow
I kind of just assumed it wouldn't be used until VRAM is overloaded to the point that it would be preferable to nothing. Like, surely there is a break even point where that happens?
It depends, sometimes the driver or api will do something like swap, some stacks will just elect to crash on the assumption there’s no point in continuing. But if you need to wait on main memory access in a graphics pipeline it defeats a large part of the purpose of the parallelism of a gpu and it’s not just like things get slower, you end up approaching unaccelerated performance and you more or less might as well not have the gpu.
The following users say it would be alright if the author of this
post didn't die in a fire!
