Page MenuHome

Multiple CUDA gpu are not being identified or used correctly
Open, NormalPublic


OS: Windows 10 Pro 64bit, April Update
Blender: 2.79b f4dc9f9d68b
GPU: 2 x EVGA GTX1080 Hybrid (not FTW Hybrid, just the base Hybrid)

I have both monitors hooked up to one card. When I launch blender and configure GPU rendering it shows two GPUs, but both say "(Display)" next to them. When I render, I often get the error "illegal address in cuctxsynchronize line 1372". I increase the TdrDelay and it improves things. But the more complex the render, the longer that value has to go up, or if I'm overclocking.

Now, If I select only one of my GPUs to use for rendering, let's call it the "top one" in the list of GPUs, and I render, GPU0 in my task manager shows high usage and GPU1 shows 0%. If I go back into blender and deselect the "top one" and select the "bottom one" and run the render, the same exact result happens in task manager. GPU0 shows high usage and GPU1 shows 0%.

This happens for every blend file, simple or complex. All I have to do is click 'render'

I've tried this with and without SLI enabled, as well as with and without the SLI bridge installed.



Event Timeline

I'm not sure why you can't do much, seems like blender isn't interacting with the cuda libraries to detect gpus. Can you explain more?

Just an FYI, the same problem is happening in 2.8 beta