What happens when VRAM is full?
I want to know the current nvidia / AMD implementation for managing VRAM resource allocation.
We already know that operating systems use swap / virtual memory when system RAM is full, but what is the equivalent of swap when it comes to VRAM? Will they go back to system RAM or hard drive?
I thought reverting to system RAM was reasonable, but due to my experience, video games suffer a lot (1/20 of typical FPS) when they are out of the video memory area, which made me doubt they were using system RAM. I think the system RAM is not that slow to lag so far behind the game.
In short, I would like to know what the current implementations are, and what is the biggest bottleneck that keeps the game behind out of memory issues.
source to share
-
replacement is actually done in RAM
if there is enough RAM for the exchange. File sharing is not possible due to low speed, see next paragraph
-
The RAM itself is not that slow (still slower), but the buses connected to it
when replacing system memory with a paging file, memory is replaced when necessary (changing application focus, opening a new file / table, etc.), this does not happen so often, but if you do not have VRAM, then you have problems, because which is usually the most gfx data is used in every frame.
This results in swaps per frame, so you need to copy very large chunks of data very often, for example swapping 256 MB at 20 frames per second results in:
256M x 2 x 20 = 10 GB/s read 256M x 2 x 20 = 10 GB/s write
this equates to 20 GB / s bandwidth required for rough work, depending on the memory controller and architecture. You can read / write concurrently up to some point to theoretically get an approximate speed of up to 10GB / s, but still that's a huge number for just a 256MB chunk. see data here:
My setup at the time only had about 5GB / s writing to memory, which is nowhere near the required memory transfer rate required for such a task.
source to share