Doe's NVlink work for VfR?

Hi,

I tested the memory usage and NVlink, but the rendering is stopped with the message:

“Cannot allocate memory for the progressive frame buffers. To reduce memory usage, reduce the number of render elements.”

Before at the process window was shown:

"
Total memory for device 0 : 11264 MB, free 8376 MB

Building optimized CUDA kernel (0) for comp cap 7.5 for device 1…

PTX file generated with CUDA Toolkit v7.5 for CUDA compute capability 2.0. CUDA runtime version 7050.

Optimized CUDA kernel assembled successfully

Total memory for device 1 : 11264 MB, free 8376 MB

Maximum kernel work group size: 1024

Maximum kernel work group size: 1024

Querying NVLINK …

NVLINK detected between 2 devices

Device 0 is using NVLINK with id 0

Device 1 is using NVLINK with id 0
"

Why there is no 22GB limit used?

-Micha

This message “Cannot allocate memory for the progressive frame buffers. To reduce memory usage, reduce the number of render elements.” is for CPU memory, not GPU.

Otherwise NVLINK is used, your mem limit is close to 22GB (doesn’t scale linearly completely).

Best,
Blago.

This doesn’t seem like memory related crash.
best to send scene to support

Best,
Blago

@Micha , could you please provide some additional info about the unhandled exception error?
- Are you able to reproduce it reliably?
- Does it occur only when attempting to render a specific scene?
- Is it reproducible after restoring the default render settings and switching back to the GPU engine (note that Environment, Camera and Render Output settings will be reset as well)?

I have tasked my colleagues from our Support department to reach out to you for further assistance with the GPU memory issue.
Regarding the unhandled exception error, if you manage to reproduce it again, please share the info requested in my previous post.

Kind regards,
Peter

Thank you for tasking your colleagues. I don’t wrote anything new about the error because I suppose so the problem is caused by I have still 11GB available VRAM only. What I did - I filled the VRAM per Rhino below the 11GB limit and used Photoshop to fill the VRAM until the limit was reached - this caused a driver crash and the error. It shows, that there is no shared VRAM above 11GB available, not for the system, not for VfR. I’m very curious to get the shared memory working so that more than 11GB are available. Maybe the latest driver 430.86 is the problem.

Driver isn’t the problem. Same available VRAM per NVlink and non NVlink, only the VRAM usage is the 100% same in NVlink mode, but no bigger scenes can be rendered.