Render farm from legacy PPC machines for Blender?

[disconap]

Hi all! Ok, so I'm learning to use Blender, and while I do eventually plan on buying a higher end machine once it gets added to our regular services/client workflow, for the moment it's a training/basic exercise in learning. And we have a number of legacy PPC machines in our office, all of which are working, stable workstations. Here's the full list (I plan to pick 4 for work, 1 for controlling; maybe 5):

Powermac G5 2X2.0gHz, 6gb RAM
Powermac G4 Sawtooth 2X1.3gHz, 2gb RAM
Powermac G4 Sawtooth 450mHz, 1gb RAM
Powermac G4 Mystic 2X450mHz, 2gb RAM
Powermac G3 b/w, 500mHz G4 upgrade, 1gb RAM
ibook G4, 1gHz, 1.13gb RAM
ibook G3, 700mHz, 312mb RAM
imac 400mHz, 1gb RAM

I figure the three G4 towers, the upgraded G4 tower, as the cluster with the G5 as the controller (and worker), or for when the G5 is needed using the G4 ibook as the controller (then maybe using the G5 as part of the cluster when it's not being used?). I've read up on server farms using older machine vs. buying a newer workstation, and I understand the differences in cost and results and the like, but this isn't a high-end environment and I'm basically trying to get some performance out of machines that are otherwise sitting collecting dust while I learn. I plan to use Blender and possibly just Xgrid for admin.

Thoughts?

 

[chris11]

Hi,

I'd use:

Powermac G5 2X2.0gHz, 6gb RAM
Powermac G4 Sawtooth 2X1.3gHz, 2gb RAM

and replace:

Powermac G4 Sawtooth 450mHz, 1gb RAM
Powermac G4 Mystic 2X450mHz, 2gb RAM
Powermac G3 b/w, 500mHz G4 upgrade, 1gb RAM
ibook G4, 1gHz, 1.13gb RAM
ibook G3, 700mHz, 312mb RAM
imac 400mHz, 1gb RAM

with a single new Mini. I'm petty sure the Mini is as fast as those 6 combined and the cost of the Mini pays back soon in terms of electricity
and system administration head aches...

Bye,
Chris.

 

[disconap]

Appreciated, but electricity is included in our rent (and is from renewable wind/solar sources, for any of the hippy minded out there), so electricity isn't my concern, not spending any money up front is, as we aren't sure this is something we'll pursue professionally. So I guess I'm more asking is it worth putting this together? I assume the benefit would be pretty much the same as rendering each frame individually on each machine, just all administered in one place, right?

 

[giffut]

Hi ...

... know this might sound peculiar, but one fast Core 2 Duo Quad would probably replace all G4s in your setup for rendering purposes, regarding pure speed. I don´t think - even with free electricity - that it´s worth the hazzle to use those machines for a rendering cluster, just too slow and the hardware too much outdated (IDE HDs, SATA requires new money to be spend on controllers, RAM limitations).

Don´t forget that you need the software speak equal on those systems, but with most of the G4s you are best served staying under OSX 10.4, as OSX 10.5 eats too much of precious resources which you lack for rendering, then.

If software wise it is doable, you could just use them to learn about the installation and mantainance structure. The best option would be to go Intel, either via some Mac Minis, Mac Pros or hackintosh nodes.

 

[disconap]

I don't think that sounds peculiar at all, but again it's right now about experimenting. I don't have the roughly $4k to pick up the workstation I want, but I do have boxes laying around that work. Here's my thought:

Controller/main workstation: G5
Nodes:
G4 2X1.3ghz
G4 2X450mhz
G4 500mhz
G3(4) 500mHz

So I would essentially have 8 processors all working simultaneously, though obviously at different rates. I'm just wondering if the lower end two are necessary; the dual 450 should be used simply because for the energy usage and heat generated, it's twice what the other two are, but I'm trying to decide whether the two 500s are worth bothering with. For the knowledge of how to set it up alone they may be worth it, and of course they'll each work slightly faster than each of the two processors on the dual, but adding two more configurations into the mix may confuse things. I'll probably start small and see how easy it is to add new machines to the cluster; ultimately the goal would be like 30 seconds of some kind of test animation, so while that will be demanding I think this could work, I'm just wondering how large an advantage it is over just using the G5 and taking a day off work.

On the OS, I was planning on paring down 10.5 or using 10.4. Or running the whole project on linux, but since most distributions (other than ground up Debian and a couple others who have maintained PPC for the PS3) have abandoned PPC support, OSX may be the way to go, especially if I want to become familiar with Xgrid.

So really it's an exercise, I was just curious if anyone had had any success using these legacy machines. I debated trying to piece a farm together from 1-2ghz P3s and 4s, but from what I've read on graphics render times they are roughly as fast as a 500mhz G4, so...