Goldorak wrote:
The feedback wasn't conclusive in the poll, and much of the data was pre-V3. I think we now have a good sense about some of the areas we should addressing in addition to the planned subscription offering we are rolling out, including:
- rent to own option (as an alt. to pure rentals, but perhaps without continuous support/updates)
- allow more flexibility in network rendering configurations with single per account network rendering price point (for example some user want more than two nodes, e.g. 3+, even if it comes to same GPU count)
The second point above is something we are working on from a technical perspective in the licensing system . V3 does allow for an implementation of this, specifically in place for headless rendering across two nodes, but it is not tested yet, and can't provide an ETA until it is.
Still no news, basically no answers.
Goldorak wrote:Regarding the current 20 GPU limit, we have found that support issues go up way, way up as we enable more complexity in network rendering.
We ideally want users to keep network rendering to small node counts (even though we make less money on this model with boxed licenses). There is an outlier case where users spread network rendering across 20 nodes, with 1 GPU each, which is a challenge to support (and not how network rendering should be used on non-server boxes if at all possible).
When rendering still frames I may or may not need more than 20 gpus, but animations with thousands of frames is another story. And for animations, why would you use Octanes network render, or any other approach that doesn’t distribute the frames?
Am I supposed to lock up my workstation for batch rendering? No, you use a render manager.
In my case I use C4Ds Team Render Server with 20 dedicated render servers and can add 10 workstations during night/weekends, and with
any other render engine that’s not a problem. And even if I were allowed to use all my gpus the cost would be insane.
A little example:
If I buy licenses for the just released Cycles4D it would cost me about $1600 total for the main license and (2 are included) 28 slave licenses.
If I go with Thea render it would be only about $800 for the same setup.
What would it cost with Octane? A freakin’ $18000.
Eighteen thousand dollars
Goldorak wrote:It was much worse in V2 when losing a single GPU killed the whole render, which V3 addressed. If you are going invest in machines to use all 20 GPUs w/ V3, then consider 2x 4U servers w/ 8x GPUs in each node, which will likely max out the power of most offices and provide ideal stability. Our DTLA office is in a former Pacbell building, with huge power supply and an A/C server room; but even, so an 2x4U 16- GPU config is about the limit of what we can reliably power
How is what works in
your server room relevant?
Are you on 110 or 240 volt? Assuming you completely maxing your power draw it would mean something like 36 ampere if you’re an 110V but only 16 if you’re on 240. In my server room, which is nothing fancy, I have 4 separate 16A outlets and 4 10A and I’m on 240V so that’s 25 kW, enough for 135 GTX1080.
Goldorak wrote:overheating is an issue even in an AC server room
An AC can provide 500W of cooling or 10 kW or 4.7 kW or any other number, so that’s a pretty useless statement.
