I just bought a second RTX 3060 only to be completely underwhelmed by the increase in speed. I dont actually think its using both GPUs at all. The current Project im working on takes about 30 seconds using OPTIX and 35 using cuda at 100 samples and the time is essentially the same whether i have 1 gpu turned on or both. Ive tried rendering with each seperately, so i know that one of them isnt broken. Im just extreamly confused, is there something im missing here or is it really not that much more efficient to use Both. Everything ive looked into says it should be double the speed.
Have you tried rendering more samples or a heavier scene? 35 seconds is a very short amount of time to judge the performance margin, as most of that time may be taken by building the BVH, uploading the textures to the GPU, etc. The actual render time may be just a fracture of those 35 seconds and the GPUs may not have the time to stretch their legs properly. Try more samples, or a different, more difficult scene. Set it up so that the render takes 10 minutes.
2 Likes
Hi. Can you share screenshots of your settings at:?
- Edit menu → Preferences → System and Optix tab
- With your scene open, Porperties editor/panel to the right in Blender, Render tab (Render Engine, Device)
Also you share render times obtained with a single GPU enabled and with both GPUs enabled rendering the official sample scene titled “Car Demo” (BMW27_2.blend.zip)
For the car demo it renders duel gpu at 14 seconds and duel gpu in 8 seconds, so it seems like it is faster, i dont know why it isnt making a difference on my other render, its definatly a very large project. It takes probably half of the 30 seconds just to load the textures to the gpu, is there any way to speed that up. The other part is that it doesnt really seem to be speeding up my render viewport very much either.
Here are the screenshots:
Also is there anything i can do to optimize my settings to make it faster?
Your right, It does just need more samples. When i rendered using 1000 samples it did it in 1:30 on duel GPU and 2:20 on single gpu. Good to know. That being said is there anyway to make my render viewport more optimized, or optimize my Render time/settings in general?
For viewport speed I would use simplify (reduce texture size and subdivision count), pixel size at 2x, and perhaps reduce the amount of light bounces to 2.
Also, if you are using the denoiser, make sure the denoising starts at the end of the render. For instance, you have 16 samples for viewport, so set the denoiser to 16 too. Denoising while rendering tends to slow things down as well.
2 Likes
You disable Intel CPU which shows on the screenshot. It has been reported that hybrid CPU+GPU rendering is not optimal when you have very fast GPUs. I am not saying that what I mention is the cause of the problem, but it is recommended to do it.
You should give more details of the scene (textures, particle systems, number of vertices, etc).
Those 30s you mention before you can see the render started generally correspond to CPU tasks. There are some tricks depending on the scene. For animation in some scenes it may be useful to enable Render Tab → Performance → Final Render → Persistent Data
For Cycles Viewport you have Render Tab → Simplify menu (as you have already been recommended)
1 Like
Yes, good idea to disable the CPU. It seems blender 3.0 does not scale well while rendering with GPU and CPU at the same time. Just the GPUs should give an even faster result.
1 Like