Suggestions for a very good 3D rendering machine

[imported_Tick]

I'm not sure if tristor mentioned this in his posts, but definitly consider CHAOS live cd's for your nodes if you do anything with openMOSIX. Save you so much effort. And you need no hdd's in the nodes, and don't have to worry about netbooting. Also, from what I've seen, clusterKnoppix is using a very old kernel, so you might want to start with a new install of your favorite flavor, and install openmosix from there. There are several good guides out there.

Edit: Oh, also look at ROCKS and OSCAR.

 

[Tristor]

Originally posted by: Tick
I'm not sure if tristor mentioned this in his posts, but definitly consider CHAOS live cd's for your nodes if you do anything with openMOSIX. Save you so much effort. And you need no hdd's in the nodes, and don't have to worry about netbooting. Also, from what I've seen, clusterKnoppix is using a very old kernel, so you might want to start with a new install of your favorite flavor, and install openmosix from there. There are several good guides out there.

Edit: Oh, also look at ROCKS and OSCAR.

I knew about CHAOS, but had never used it. It is quite interesting though, and you are quite correct that with cheaper systems it might be a better option than netbooting or PXE with ClusterKnoppix, since it still supports an SSI system, but is in a bootable CD form that is small enough easily fit in the memory of any node, no matter how weak.

I may look into using it for my next cluster, actually. Seems like it might eliminate a lot of headaches dealing with PXE, although I must say that PXE is quite nice when it works right.

 

[Modelworks]

I wish I could get brazil to render on machines that I find discarded.
Its requirements are too high though.

I have had success with mental ray though.
Currently using an off site render farm.
It just worked out cheaper than having it local.

As for the original posting.
I do all my work on a dual core 3.4ghz pc with 4GB ram, winxp x64.
Vido card is a 7900gt.

There are the higher end quadro cards but I never saw the need.
The 7900Gt has worked for me without issue with zbrush, max.
I haven't used a "pro" card in years.

 

[Tristor]

Originally posted by: Modelworks
I wish I could get brazil to render on machines that I find discarded.
Its requirements are too high though.

I have had success with mental ray though.
Currently using an off site render farm.
It just worked out cheaper than having it local.

As for the original posting.
I do all my work on a dual core 3.4ghz pc with 4GB ram, winxp x64.
Vido card is a 7900gt.

There are the higher end quadro cards but I never saw the need.
The 7900Gt has worked for me without issue with zbrush, max.
I haven't used a "pro" card in years.

Just out of curiosity, what off-site render farm do you use, and do you know what sort of equipment they use for their nodes?

 

[Modelworks]

Currently its rendercore at www.rendercore.com
A routine render might be 1600 frames at a rate of 5 minutes a frame on a dual 3ghz intel machine.

Rendercore is able to do it in 1.3 hours versus 800+ hours that it would take the above workstation.
It ends up costing about 700.00 from rendercore. But thats cheaper for me than the outlay for building my own render farm.

this is the hardware info from there page:
Hardware
A. 3THZ COMPUTING POWER
B. 3.6GHz XEON Quad (4) FILE SERVER
C. 3.0GHz XEON DUAL(2) RENDER NODE
D. 500 machine
E. 10 Terabytes RAID level 5 storage
F. 2 TO 4GB MEMORY
? Operation system
A. linux(32,64bit) / B. windows(32,64bit) / C. macosx(32,64bit)
? Network
A. 2xT3 connection to internet with capacity up to OC-12
B. Gigabit switched internal network for servers

 

[Tristor]

I wonder what software they are using to cluster their render farm. I'm pretty sure they are probably using a queue manager system, mainly due to the fact that they probably have to do render work for more than one client at a time. At any rate, I think I'm going to build a farm for doing encoding, we will see how it works out. Maybe this time I'll document what I do a little better and post an article about it on the forums.

 

[wwswimming]

So I can calculate it as ((19.5 + 30) * .28) + 19.5 + 30 = system power draw, which is 63.36W, rounded up to 65W per system. I had 60 nodes, so that comes out as 3.9kW of power draw for the cluster.

that sounds super efficient. like a laptop. 19.5 watts for a P3.

i was initially surprised to read that some of Google's servers
used 1.4 GHz processors or something like that. but i guess
that's enough for fast data transfer.