memory - What happens when VRAM is full? -

i want know current nvidia/amd implementation of handling vram resource allocation.

we know operating systems use swap/virtual memory when system ram full, equivalent of swap when comes vram? fall system ram or hard disk?

i thought falling system ram rational, experience video games lag horribly(1/20 of typical fps) when out of video memory space, made me doubt using system ram because think system ram not slow make game lag much.

in short know current implementations , biggest bottleneck causes game lag under out-of-memory situations.

1. the swapping done ram

   • if there enough ram swap to
   • swapping file unusable due slow speed see next bullet

2. the ram self not slow (still slower) buses connected are

   • while swapping system memory swap file
   • the memory swap occur when needed (change focus of application,open new file/table,...)
   • this not frequent
   • but if out of vram in trouble
   • because of gfx data used in each frame
   • this leads swapping per frame
   • so need copy large data blocks often
   • for example swaping 256mb 20fps leads
   • 256m x 2 x 20 = 10 gb/s read
   • 256m x 2 x 20 = 10 gb/s write
   • which 20gb/s bandwidth needed
   • of coarse depending on memory controller , architecture
   • you can read/write simultaneously points can close 10gb/s in total theoretically
   • but still huge number 256mb chunk of data
   • look here cache size estimation on system?
   • my setup @ time has memory write around 5gb/s
   • which near needed memory transfare rate needed such task