Average Render Nodes - How Many?

[jujubee]

I'm curious - how many render nodes (or broken down into buckets) does the average visualization firm have at their disposal?

[paulison]

well we have around 20 recent dual/quad cores and the 8 or so workstations on top of that. its never enough even though we mostly do still images. I'm sure that this is less than the average visualisation firm but i guess we just add them as needed or else if really stuck will use one of the online rendering services but we haven't had to do that yet. we get by on ~20 though for purely arch vis

I'd expect medium size offices may have more like 50-100 machines?

I wonder how many the companies churning out all of those recent 3d movies are using?. 1000+?

[jujubee]

Well, I recently used a rendering farm for my first time and I'm terrified now. I took a x64 scene and tossed it over to respower - had to convert all gifs over to jpgs, kill glossies, and turn down all the settings on my frosted glass.

Kicker was that they aren't on x64 so barely anything would render. Luckily I used a day pass, complained (they were rather nice people over the phone), and they helped me out with an extra day.

I ended up rendering a large portion of frames on my own network. After that experience, I'm wondering if it makes more sense to buy my own computers instead of coughing up big bucks for one of the newer firms.

But yes - I am curious to see what others use and if I have any remote possibility of catching up.

[QuakeMarine1]

jujubee - try my farm and you will do what my last client wrote me

"I’m downloading the frames now, and they look awesome. On time, on budget. Music to my ears. I’m doing a little frame dance around my room as I watch them come in."

[jujubee]

You know - I could never figure out how to calculate a gHz hour...

I'm afraid I send you a file and encounter problems, then a day later my bill is $3,000...

[jujubee]

Yep just looked at your site - at 10 mins a frame on my dual quad xeon 5355s (had a scene with full glossies) it would have been over 3k. I know you're inexpensive but OUCH that would have killed me. Please don't take any offense, but I'd almost rather by 5 QX6600 systems for that kind of cash.

I'm just trying to figure out how a professional rendering farm is more effective in the long-run. Hence my original question, how many render nodes does a typical archvis firm usually have?

[Neilg]

We have 20 dual quad cores and 16 workstations, all on dual quads too.

I'd say as far as output goes we hit it hard, it's almost never free for more than a few hours. We stagger animation deadlines and it never tends to get too tight, although it gets close to it a lot.

[jujubee]

So if QuakeMarine1 at RebusFarm hit 1200ghz running dual quadcore xeon 5355s at roughly 20 GHZ per machine, that would be about 60 units/comps total.

Price per unit would be about $4,000.
An estimate for all the comps would be $240,000.

A Single QX6600 is a total of ~$600 and is about 12 GHZ per machine.
To match that renderfarm you would need about 100 units/comps.

An estimate for all the comps would be about $60,000.

Assuming only half of those computers on his farm (or others) are dedicated to your scene at any one time, going with a QX6600 solution would come to around $30,000 or 50 units/computers.

Seeing that you're (or me for that matter) isn't going to be a full-time render farm, half that number would be a pretty decent internal farm.

In total, 25 computers or $15,000. (which takes us back to Paulison's number of 20 computers at his firm.)

Of course this doesn't include backup power supply systems ($150 per backup/3-4 computers = $1,500-$2,000) and a nasty electricity bill.

[andrewjohn81]

For 50K we put together a little farm with Dell 670's (this was about a year and a half ago).

We've got 15 machines, dual Xeon 3.4Ghz with 2gigs of ram. Now they have 4gigs for an extra 3K. Just an 80gig sata drive and crappy NVS nvidia card. I couldn't get them lower than that. I think each machine was about 1600. about 10K went for the air conditioner we needed for that and the electric. 300 for a nice wire metroshelf type shelving unit to hold them all on nice wheels, and 3K for a KVM switch along with a few hundred for a 24port 1000T switch. The rest was all for the machines. Oh, and some ethernet cables, but that was less than 100 bucks.

I'm not really a Viz firm, I work for a corporation in the Viz group here. It took forever for us to talk them into spending the money for it, but it wasn't hard when you take a look at the prices of sending this work out.

Not sure if that was what you were looking for, but that's what we did. Hope that helps somebody.

[QuakeMarine1]

if I get a scene I normaly render each 100th frame to get a feeling of the rendertime
with that info I quote for the render and send some preview renders to the client to get his approval
until that point the client have no costs
I dont know how other farm handle that issue but you never send Pandorras Box to me

when asking the amoung of rendernodes you should take to account the costs for purcase, amortization and maintenance
it makes no sense for a small 3d guy to have 10 octa core PC inhouse
he do 1 animation and maybe 1 still per month
he better use an external farm if the deadline is too tide and job too big

a better answere for your question is
-runtime vs idle time of the inhouse rendernodes
-amoung of jobs per month
-"size" of the 3d firm

how helpfull is andrewjohn81 answere to you ?

[andrewjohn81]

quake makes a good point with idletime vs. rendertime.

We usually pump out 1-3 animations every two weeks and about 2-3 high res (5k square+) per day. That's average. We need the farm to accommodate the animations, but many times it is idle with nothing to do for a few days at a time. Most of our animations are 30seconds or less and my frames don't last longer than 1 hour. Most are 5-15 minutes. large renders take 1-5 hours.
This is just for the work we do here. We have definitely thought about using external farms. They just don't make sense for us in most cases.

I just wanted to give the information so people would know what some of us smaller groups do and a little bit no how to set up the same thing there.

Also, being in a corporation it doesn't come down to knowing if it's worth it. If someone offers you extra budget for something you could use you take it. You can't take it and use it for something else so we took it and built what we could.

[paulison]

we are kind of similar with the budget. The farm started off with a budget for 4 machines, then 8 , then 12. So you either take a big gamble and buy lots of machines, or steadily build it up adding and upgrading on a regular bases so that you never have a huge debt. If you are in a position where you have to prove to the people with the money that the investment is required then it may have to be a staggered investment like this

Actually we use a rentsmart system now so that we rent the machines and replace them as the contracts are paid off which is a good way of staying up to date. all you need is enough cashflow for the rent which works out well when doing tax because its a deduction for us i think.

I think if you buy machines outright then in a few years you will end up with a patchwork farm and a lot of obsolete computers that arn't worth plugging in.

Power is a big issue...I think ours went up a LOT since we started to get saturated with work but i guess that pays for itself. Older machines just arnt worth plugging in though when you consider the power they use relative to their speed.

I think you have to consider power, cost of additional equipment, A/C when you work out how many you can afford as pointed out already.

Personally I'd start with the bare minimum required to get the job done so that the machines are not often idle. Dont base it off just one animation job and then have 20 machines sitting there for a week idle afterwards

[QuakeMarine1]

can i ask the monthly costs + power + X for that rent farm ? (12 node ?)

[jujubee]

Quake - do you have any plans for day passes?

[QuakeMarine1]

no
you run better and cheaper with the current system
and me too
sure - other farms have a day/week flatrate but you run then in low priror
that farms do it to sell the idle time and hide the low priror often
in worst case some other regular clients render and your stuff finish 3 days later or run on 20 nodes only

your job never have 24 hour 100% CPU usage
you never have 100% of the farm if another guys jumps in
thats why you can not be sure all renderings are done after the flatrate expires

the last point should explain my point of view in that manner