building a 1THz render farm for blender users. Would it be used?

Hello Blender friends,

So a friend of mine is wanting to put together a render farm for blender users to use but before he puts it together we were wondering if it would even be used.
the cost would be in the range of 0.02/GHz/hr

were estimating the cost to keep the farm up and running would cost Him about 13,000 a month to run

so depending on the demand for it would depend on where we start hardware wise. low density or high density and GHz

Please let me know as we are building this for all blender users. :slight_smile:

There’s a reasonable bit of competition. I can think of five similar farms just off the top of my head and I don’t even use them very often. But that price is nice if you can live up to it.

if we can maintain 80% - 100% client usage then we would. That is the main reason I am asking. We don’t want to be spending such a large amount per month if the usage is under 80% per month

There are already a few renderfarms supporting Blender (or even are built for Blender) so you’d need something to really stand out from the rest to get good business.

The pricing sounds good from a user-perspective so I’d probably try it out but if you want people to choose your renderfarm over everyone else you need either very good marketing or an exceptionally good service.

Start by changing that pointless billing metric.

0.02/GHz/hr, what cpus are you using? what architectures? how much ram is available, is there gpu acceleration? etc.etc.etc.

I would imagine some good marketing and an easy user interface along with a good price would win a few people over. I mean you will have to compete with something like this: https://render.st/

So it would have to be something of similar capability or offer something unique. Maybe Octane rendering as well? Maybe develop a plugin for your render farm so users don’t need to go to the website to upload a file to render. Get the plugin to alert when the file is ready for pick up? That sort of stuff.

very insightful thank you.
My Friend and I have discussed what kind of front end we would have…
we have access to the resources and are working on the best way to implement it.
research research research :wink:

It’s really noble and positive but I am very afraid… As mentioned there is a lot of competition and I am very doubtful of scenarios where you really need a farm for average Blender user/customer - i may be completely wrong. As an example though I just had to render 7500 frames of architecture and it took 14 hours with full GI, flicker free with a few GPUs (below 10 sec per frame @ 720p). We considered a farm but GPUs saved the day and drawbacks of farm would be immense file size over net (issue of its own) + multiple rerenders to eliminate errors that adds up to cost.
In other words I think a small studio who needs to react to rapid deadlines will have the resource to get these few compute devices while a freelancer may not have the funds(/be reluctant) or does not usually deal with such huge animation works requiring a farm. Exceptions in need already have competition.

I think if you wish to offer render farm services you need to transition into it gradually. The main work is about creating super elegant automation (minimal hassle maintaining, perhaps some smart bootup/power save); flexibility for maximum audience including support for various other programs (licensing will be painful) as well as best possible user experience (interface).

Yup, we have ~24 GTX5xx cards available to our artists for rendering on. we are able to usually output 4000-10000 frames a night (depending on the scene) with this setup.

do you run into any restrictions when relying only on GPU?

Yeh the 1.5GB memory limit… we manage fine though…