12 gpus limitation is a limitation

Fri Sep 18, 2015 7:27 am

Ok,
now I have 16 gpus and 4 licenses.
I want
I need
I pretend to enable all gpus during network render.

That's all folks!

Fri Sep 18, 2015 8:19 am

That amount of power.... is to much for a human to handle.

Fri Sep 18, 2015 9:13 am

Ok,
now I have 12 gpus and 3 licenses.
I want
I need
I pretend to enable more gpus during network render.
i wish

That's all folks!

Sat Sep 19, 2015 7:53 am

Vray distribute render has not CPUs limits.

Sat Sep 19, 2015 8:20 am

gabrielefx wrote:Vray distribute render has not CPUs limits.

We will revisit the GPU limit with 3.0.

Sat Sep 19, 2015 2:33 pm

Goldorak wrote:
gabrielefx wrote:Vray distribute render has not CPUs limits.
We will revisit the GPU limit with 3.0.

that sounds promising..I've see more & more Guys bumping into this..-would be nice to be limited by licences not, by some sort of number (but I guess You made it for some reason).

Sat Sep 19, 2015 5:44 pm

They made it for $$$

Sat Sep 19, 2015 11:04 pm

glimpse wrote:
Goldorak wrote:
gabrielefx wrote:Vray distribute render has not CPUs limits.
We will revisit the GPU limit with 3.0.
that sounds promising..I've see more & more Guys bumping into this..-would be nice to be limited by licences not, by some sort of number (but I guess You made it for some reason).

Octane Cloud Edition (which powers ORC) can run smoothly on 100s of GPUs. But that is on 10GE/ 56 GB inifiniband, in a datacenter where we have uniform and validated GPU blade servers, 24/7 cooling and SLAs which make supporting network rendering at this scale straightforward.

Local network rendering has to work on 1 Gbit LAN + every possible combination of consumer hardware configs users may have. Sending huge film buffers+render pass layers over LAN is therefore a more ambitions support proposition, even on the 3-4 nodes typical users are likely to be capped to (much less the 12 they could theoretically run on). This has always been a critical path in 2.x, where being disconnected form a network slave for any reason means losing those samples forever.

In Octane 3, the filmbuffer system was rebuilt so that when a network node is lost, the samples are not (you can also save film buffers to disk and resume the render later). Network rendering is still being put back together in Octane 3 at the moment, but once that is done, we can reevaluate how this changes things. We will have OpenCL (GPU and CPU) to consider in mixed render configurations as well.

Sun Sep 20, 2015 8:19 am

Designing a good cache system can helps distributed render with many gpus.
Also...
On every workstation I installed a Logitech keyboard with an lcd display. (510 model)
With the MSI Afterburn app I check vram and gpus use, temperature, fans speed. Every machine has its own fan speed curve.
I would like to have an app that monitors all these things through my tablet.
I would like to turn on-off the gpus, allocate them to tasks, checks renderized frames, check temperatures and so on...
I don't use Backburner anymore because using moving proxies I optimized the scene loading, on each frame the loading task is reduced to minimum.
I want an option where if I plug-unplug gpus to the network Octane recognizes automatically the new gpus and add them to the current task.
I want dynamic allocation of Octane tasks....but I imagine that your data center has all these controls through an html interface...

regards