Jump to content

Recommended Posts

Hi everybody,

my company wants to realize a sort of internal "render farm" because we need a lot of computing power.

In the next week,we will ask our IT manager, but this guy always tries to sell us expensive stuff also when are not needed, so I would like to arrive prepared with some ideas in my mind.

We use Redshift as a render engine.
Following the question:
1- Does redshift support multi GPU (more than 3 of course)?
2- Does C4D support multi GPU (more than 3 of course)?
3- How C4D and Redshift do communicate with this server/rack?

4- The server/rack how should be composed? (at the moment I can only image a big computer with a big motherboard and 10/15 GPUs plugged in )


At the moment we have 2 workstation; one with 2 Titan RTX, the other have 3 Geforce 2080. Not so bad, but my company wants to invest....so...


At the moment we don't have a budget but the investment could be of about 10/15K, and every 1/2 months add 1 extra GPU

Is there someone experienced in this?
I don't have a big knowledge in IT, someone could give me some "macro" info?


Thanks in advance!

Link to post

Redshift just like Octane does support multi-GPUs.

- Benchmarks for Redshift going all the way to 8 Tesla V100 for now, Source:



- Octane have more benchmark data with people going crazy with 16 Tesla V100 or Quadro 6000, Source:


- Both Render engines do scale kind of similarly as you add more GPUs to the mix, source:


- Here is a x16 GPUs rack system that might look like what you would end up with (or not).




Link to post

Thank you Voidium and sorry for my late reply, has been a busy period!!

I read that Redshift doesn't support more than 8 GPUs so 1 rack should be enough ( https://www.asus.com/Commercial-Servers-Workstations/ESC4000_G3S/ something similar) 


Anyway my main question is how my computer and this server do communicate?! by team render? this means that I will have to pay a Cinema4D and Redshift subscription only to be used as team render!?

And not least how I decide the other component?! CPU type, n. of CPU,  RAM, PSU, etc


I would give my chief an idea of cost for this internal render farm (excluding the GPUs), is there some site or specialist that could help me?!

Link to post

Hi Ennebie,

I maintain my company's internal render farm, and have just moved into using Redshift within the last year.  Here's my findings:



Yes you will need 1 Redshift license for each computer rendering, and up to 8 GPU's per computer. 



The biggest factor I kept in mind when trying to max out the GPU capacity for a Redshift renderer was the amount of PCIe lanes available on the CPU and motherboard.  Also, note the number of PCIe slots on the motherboard, as this will dictate how many GPU's you can physically use.  Without getting into a server type of setup, you'll be limited to probably 4 PCIe slots on a board at max, and at most 64 lanes on the CPU.  The best setup I've found for a dedicated renderer is a Threadripper CPU (64 lanes) paired with a Gigabyte AUROS motherboard (4 slots, 2 at 16x, and 2 at 8x speed).  


Rule of thumb for RAM, you'll want to shoot for 2x the amount of DRAM on the GPU.   EG 4 2070 supers = 32 gb of DRAM, so get 64gb of RAM.  4 2080ti would be 44gb of DRAM, but 64gb of RAM would still likely be adequate.  PSU - bigger the better.  I'm using a 1300 watt to power dual 2080ti hybrid cards, and a 1600 watt for 4 2070 supers.  If you want 4 2080ti's you'll probaly need a second PSU.  General electronics advice is to never exceed more than 80% of your power capacity.  This will allot for any power spikes and not shutdown your system.


Here's the GPU-centric build I put together for my studio:

https://pcpartpicker.com/user/ericsmall24/saved/w4sykL - but with 4 GPUs, not just 2


And here's my workstation build:



Data communication:

We use Qube! render manager to manage our renders.  We have 2 of those GPU rigs built, but add the onsite workstations to the farm after hours.  All said and done, we end up with 7 computers capable of rendering Redshift (with ~60 additional for CPU rendering ).  Qube! uses a central computer (called a supervisor) to tell all the rendering computers (workers) what to render.  All c4d files, texture files, cached data, etc, must be accessible to the supervisor and workers, hosted on a separate server or NAS.  Qube! is nice, but can be very complex if you aren't familiar with networked rendering.  It also requires c4d commandline licenses.


You can use Team Render, which comes bundled with C4D.  I believe you get 1 Team Render server license, which you can use if you have multiple animators and need to juggle animations.  If it is just you, then you don't necessarily need the server, you can do it all within C4D.  Team Render is quite limited, and historically I found it to be unreliable, but I have not used it since 2016 or thereabouts.  I am not sure how MAXON currently does the licensing for a computer that will be rendering only.  


One relatively obscure point of consideration: if you're going to connect lots of computers to the farm, Windows Home limits to 10 concurrent connections, Windows Pro limits to 20.  Meaning if you're going to have more than 10 or 20 computers rendering, you'll need something other than Windows Home or Pro (respectively).  Neither Linux nor Mac limit the number of concurrent connections.


Rereading my post, this is looking like info-overload.  Let me know if you have more questions.



Link to post

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Recently Browsing   0 members

    No registered users viewing this page.

  • Create New...