We’ve run into an issue with our packaged build. It seems like VRAM is spiking during level load, causing it to crash out on machines which don’t have enough shared VRAM to shoulder the spike. As an example on my machine (3070 desktop)
it uses 7.5/8gb dedicated VRAM and 0.2/16gb shared vram on the main menu. When I go to load another level, the shared VRAM usage spikes to 10/16gb during the load (10-15 seconds) and then after the level loads it drops back down to 0.2gb.
Any idea exactly what’s going on here? The levels we are loading into are very light. We’re talking half a dozen actors with 1 or 2k textures on them. What’s causing this massive spike?
It might have to do with texture / model decompression. Normally models / textures are packed to reduce their size but during loading they are uncompressed. To do this the engine might use temporary memory to achieve this. The epic staff would probably know deeper details.
You could try staggering the loading of models through async loading to fan out the spike over time instead of one large load call.
Perhaps enabling texture streaming for the actors could reduce the strain initially loading lower resolution textures and slowly streaming in the higher details.
texture streaming in the project is enabled and as I mentioned there are only 6 actors in the scene with only 1 and 2k textures between them. there is nothing heavy to load in the scene that is there. For some reason stat memory is claiming 7 gb of textures in memory and 3.5gb of static meshes in memory, despite nothing being in scene…