What if I use more than 20 GPUs Tweet

[coilbook]

Hi We would like to use 30-40 GPUs at the same time to render the scene by means of buying more slave licenses. Can we do it or will we get an error?

We've been waiting for octane cloud rendering for many years and it is still not part of octane. Can octane 3.1 have a limit raised to 30-40 GPUs?

Thanks

[Goldorak]

ORC is part of Octane and has been for months, so I suggest you give it a shot if you haven't yet looked into it. The payment page will soon be public (and based entirely on OB/hr) once the new OB bench is released with Pascal card support in the coming weeks. Until then you can use ORC testdrive - PM FuzzyBro and he can get you set up.

I do think ORC is going to be cost effective probably right around the point where 20 (high end) GPUs aren't enough. That being said, you can always do multi-frame job renders on an infinite unmber of GPUs locally, and that is how ORC works for max efficiency. We will definitely re-examine the network render limit after see how headless rendering, AMD cards, and 3.1 affect stability.

We have Enterprise tier of Octane out in the wild for validated GPU boxes (like VCA) with Linux and 10GE for max stability. You can talk to our BD team ([email protected]) if that is something you might be interested in.

[coilbook]

ORC is part of Octane and has been for months, so I suggest you give it a shot if you haven't yet looked into it. The payment page will soon be public (and based entirely on OB/hr) once the new OB bench is released with Pascal card support in the coming weeks. Until then you can use ORC testdrive - PM FuzzyBro and he can get you set up.

I do think ORC is going to be cost effective probably right around the point where 20 (high end) GPUs aren't enough. That being said, you can always do multi-frame job renders on an infinite unmber of GPUs locally, and that is how ORC works for max efficiency. We will definitely re-examine the network render limit after see how headless rendering, AMD cards, and 3.1 affect stability.

We have Enterprise tier of Octane out in the wild for validated GPU boxes (like VCA) with Linux and 10GE for max stability. You can talk to our BD team ([email protected]) if that is something you might be interested in.

thanks I did not know it is out. is it integrated into 3ds max plugin where i can just purchase and enable cloud GPUs or i have to send out orbx files ?

Also 20 GPUs is total or if my master has 3 GPUs slaves can still have 20? Total 23? or 17 Thanks

[rappet]

(moved my question as additional lower in this thread

[Goldorak]

ORC is part of Octane and has been for months, so I suggest you give it a shot if you haven't yet looked into it. The payment page will soon be public (and based entirely on OB/hr) once the new OB bench is released with Pascal card support in the coming weeks. Until then you can use ORC testdrive - PM FuzzyBro and he can get you set up.

I do think ORC is going to be cost effective probably right around the point where 20 (high end) GPUs aren't enough. That being said, you can always do multi-frame job renders on an infinite unmber of GPUs locally, and that is how ORC works for max efficiency. We will definitely re-examine the network render limit after see how headless rendering, AMD cards, and 3.1 affect stability.

We have Enterprise tier of Octane out in the wild for validated GPU boxes (like VCA) with Linux and 10GE for max stability. You can talk to our BD team ([email protected]) if that is something you might be interested in.

thanks I did not know it is out. is it integrated into 3ds max plugin where i can just purchase and enable cloud GPUs or i have to send out orbx files ?

Also 20 GPUs is total or if my master has 3 GPUs slaves can still have 20? Total 23? or 17 Thanks

20 total at least for the normal release. Just a data point: we are in a pac bell building with nearly unlimited power and decent cooling. We still strugglw to get more than 20x 1080Tis working reliably, before we go to ORC.

ORC has been built in to Octane core since 3.0 and ORC supports delata uploads, so you don't have to export and send an entire ORBX or even save to anything to file. You will send almost no data to cloud for simple scene graph/node value changes. We haven't tested delta sync exhaustively in 3.06 w/ ORC yet, but will do so in the coming weeks.

In 3.07 we will be finalizing the rende job nodes and they will drive complex offline multi-part render tasks (baking for example) on both ORC and local nodes.

[coilbook]

just enabled 21st video card and it renders fine total 21 cards (4 master cards and 17 cards on slaves) not sure how limit is 20

[Goldorak]

just enabled 21st video card and it renders fine total 21 cards (4 master cards and 17 cards on slaves) not sure how limit is 20

I am glad (and surprised) this is working for you. I am not sure how stable a configuraiton that squeezes this many devices will be in the long term, but it will depend on node/GPU power density as well as device#. Pleease keep me posted.

[rappet]

just enabled 21st video card and it renders fine total 21 cards (4 master cards and 17 cards on slaves) not sure how limit is 20

I am glad (and surprised) this is working for you. I am not sure how stable a configuraiton that squeezes this many devices will be in the long term, but it will depend on node/GPU power density as well as device#. Pleease keep me posted.

So, the 20 limit is not a restricted limit, but a support limit which above is not garanteed to be working... that might be good news for those who want to strecth the possibilties.
I am not a fan of limits, speciaaly not hardware limits for using software, so I am happy with that although I thinnk I might not reach the 20 gpu anyway.

Is there any limit on number of slaves or does it have any influence on the performance by having more slaves?
I.e. do more slaves slow down the total by having to exchange information from master to more more slaves at the same time?

cheers

[mbutler2]

Actually it looks like a hard ceiling to me. Won't let me use more than 20 GPUs.

I've got 1 C4D master and 3 slaves, with 24 GPUs total. As soon as I enable enough GPUs to reach 21, it shuts down one of my slaves completely. Reducing the total number of GPUs to well below 20. (Third slave shows up as enabled in Network Preferences, but in Octane window it isn't listed in total number of slaves, and its GPUs aren't doing anything.)

But my plan was to have two masters anyway. So I can work on one machine while the other is the master for the render machines. Then for overnight renders, two masters with one slave each.

Octane licenses are expensive. With Cycles 4D I can do all of this for $400, and Cycles 4D is pretty awesome. But with Octane it's about 4X $400 and 2x $250 = $2100.

(C4D R18, Octane 3.06.2, Win10. Cards are 1080TIs, 1070s and 980TIs)

[coilbook]

same here. Mine worked with 21 and after I made some changes slave disappeared until I went back to 20 again
I know you guys want to sell more licenses but you should take into consideration one can have 1 license with 19 cards connected through amfeltec but some people buy 5-7 license and have each PC hold 2-3 video cards. So the once that have more license should be allowed to connect more slave cards