It could still use the GPU then even if the memory was a lot slower.
If there was a way to get it to fall back to using system memory instead of VRAM and still render on the GPU if it runs out, that might work similar to what you're talking about. So your expensive 3080 is no better than integrated graphics for that demanding, memory intensive render. It's rendered 100% on the CPU from that point on. If you set it to render on the GPU, it will start rendering on the GPU but if it exceeds the VRAM on your card at any point then the GPU is no longer used at all. But with rendering it's generally all or nothing. It wouldn't be so bad if there was an option to do as much as possible on the GPU until you run out of memory and then do the rest on the CPU. Other times your scene would not fit within the memory on the 3080 so you'd either have to waste maybe 20 minutes or so per image optimizing and then add that to your render time, or render on your CPU, in which case you're looking at potentially hours for render time instead of minutes.
So using a comparison of a 3080 vs a 3090 for instance, some of the time they would have pretty similar render times. So not only is this time not spent rendering, it's time spent actively making your scene look worse just so it fits into memory. The alternative is spending a significant amount of time optimizing your scene so it takes less memory by scaling back texture quality, removing objects, removing light sources, etc. A GPU that doesn't have enough memory is almost entirely useless for 3D rendering if the app has an option to render on the GPU instead of the CPU. But if it doesn't fit then you're comparing a render that took a few minutes to render on a GPU to a render that took several hours to render on a CPU.
If whatever you're trying to render fits within the VRAM on your card, then performance scales pretty similar to gaming benchmarks. I don't think it's really something you could see very well in a benchmark.
0 kommentar(er)
