Was wondering of someone can share some details on pros and cons of these two options.
Running Redshift and GPU rendering on stations with 4 GPUs.
So far going with 2 concurrent tasks and using 2 GPU per frame works fine.
But looking into option to have two slave instances ruining, each assigned 2 GPUs and rendering like that instead.
I was thinking that for example in case of single slave and concurrent tasks, if 1 task is finished but another one is rendering last frame on an job, slave will have to wait till second task is done before taking up another task from another job, no the same one.
Figuring that two slaves running even if 1 slave is still rendering last task from 1 job, second slave should be able not to wait but take up another task from different job.

Does it make sense?

Correct! The key difference is how you explained in that multiple Slaves allows you to render potentially different jobs at the same time via different GPU slots. Multiple Slaves comes with added complexity for you as the user in configuring it, such as pools, groups, etc and indeed for your end-users to understand what they are doing; whilst a single Slave is a simpler configuration setup for admins and your end-users to understand, but does let you on occasion if you prefer, dial up the GPU per task variable via concurrent tasks. You should test both methods and see which fits your business model best.

Having some issues starting multiple slaves.

I’ve turned it on in Manage User Groups.
The thing is server that have repository installed on got option to run new slave instance, but other slaves don’t have that option and still shows just run slave or run slave by name and that is it.
Am I missing something?

I think this docs page would help to explain Multiple Slave usage and guidelines for use:
docs.thinkboxsoftware.com/produc … laves.html

The only other bit of info I might be able to provide is that when modifying user groups its best to re-open the Monitor so it can refresh that data.

If you go into super-user mode does it show the options? That’s the override for all features, so it’ll help find out if it’s an issue with the user groups and we can work from there.

it worked starting from monitor and then right click on slave and start new instance. option is missing on launchers on slaves them self but, again, it was ok from inside monitor so got that sorted out.
Now just to see if GPU affinity works as planned
Thank you

You need to restart Launcher for new settings to be visible. Also, I know you mentioned Redshift, but via which 3D app are you expecting to use GPU affinity with? I assume it is Maya, XSI or C4D? (Currently working on some ‘improved’ 3ds Max magic)

Maya, XSI and also looking into Houdini if it is supported?
Concurrent tasks with GPU selection works on Houdini now right? Is GPU affinity and multiple slaves a bit different?

Yup, Houdini is supported as well as RedLine. Looks like we need to add “concurrent task support” (GPUsPerTask) to Houdini. Currently, we just allow you to configure each of your Slaves with an explicit GPU override:
docs.thinkboxsoftware.com/produc … u-affinity

Maybe see how concurrent GPU task support works in Maya, XSI, etc first and if it turns out to be the winner for you, we can investigate if its possible to add to Houdini as well. (I need to check with another developer if there is a reason why it might not be possible).

Here’s a nice developer blog post we wrote up last year on Concurrent vs. Multiple Slave which you might find useful:
deadline.thinkboxsoftware.com/fe … -rendering

I think that someone here mentioned that it was added to Houdini already?
Concurrent tasks and GPU selection in submit to deadline for Houdini ?
Same as in XSI or maya