As far as I know, hyper threading is an Intel thing, with AMD, you get the numbers of cores, that’s it!
For a render farm I would go with LGA 1150 and Haswell CPUs, pretty good bang for the buck, motherboards are real cheap, and low power consumption, which is important for render farm!
Now, if you want to build a GPU solution, a fast LGA 2011 might do better, faster bus and PCIE, although it is not crucial for GPU rendering, and of course 2-4 GTX 780 with 6GB of ram, that is if You can deal with the loss of some features, as not everything works on GPU yet!
Really depends on budget and what you want to achieve!
Lol Over 44 K here. Maybe not now but in a while.
Also, You might wannna consider Linux for rendering, as I get 10-20% more speed on Fedora then what I get on windows 7 64.
Well I thought I was getting 16 threads with AMD, then probably I will go with Intel quad or 6 cores.
Thanks for the mother board configuration. LGA 2011 will be able to support Geforce 780 in future?
What exactly are the features I will loss?
Linux on the main computer or the slaves?
Hi, on the slaves. The last changes to Blender development result in much better performance on Windows.
Linux is still faster on CPU, on GPU it doesn´t matter.
If a motherboard support PCIe 3.0 you are future proof.
The only feature not implemented on GPU is volumetrics (Smoke/Fire), other are only disabled like volume and SSS.
Cheers, mib
As it was said, any PCIE 3.0 will support GTX 780 cards and up, these cards are also backward compatible with PCIE 2.0, and frankly, either one is good for GPU rendering.
LGA 2011 supports more PCIE slots then 1150 motherboards, so you would get better performance if you plan on going GPU, especially with 4 cards!
But for renderfarm slaves, I would go with LGA 1150 micro-atx motherboard, maybe a nice looking Ikea Helmer, as it fits nicely under a desk, look at Kent Tremmel tutorial on how to build one of these!
If you want to build an Intel based Renderfarm you could get an Xeon CPU LGA1150 - they are a bit less expensive than the I7s, but you wont be able to overclock them. Just make sure to get one with Hyperthreading and an integrated graphics unit.
Assuming you want a CPU based solution to be able to use it with vue, how much do you want to spend?
No offense, but a 1150 Xeon is slower then any 1150 I7 of the same price, the motherboards supporting the Xeon’s are more expensive, usually you need ECC ram, which cost more while being slower, ands as you mentionned, Xeon’s cannot be over clocked, not that You want to overclock a render farm too much, but a bit does not hurt!
Xeon’s have two advantages over the I7, they run cooler in extreme environments (lower clock, thus lower speed), and can be run by pairs, while the I7 runs hotter, but faster, with higher clock, and can be overclocked!
For a home renderfarm, especially for something like a Helmer farm, Xeon’s are not the best, in my opinion!
No problem, maybe it is different where you are from but in germany an I7 starts at 250EUR(3.6Ghz non k so no oc, 270 for k), while the Xeon starts at 200EUR(3.4Ghz, without integrated graphics). The LGA1150 Xeons should work on all mainboards, but to be sure just look up on the site of the manufacturer.
Edit: Normal RAM works just fine for these, but you can´t use them in pairs, therefor you would need at least a LGA 2011 board with an Xeon E5-XXXX(the first X describes how many CPUs you can put on a single board with enough sockets).
My bad then, didn’t know about prices in Germany, or Europe for that matter, maybe its the same in India!
Also, a few more things are not supported on GPU, like OSL, but I suppose they will be eventually!
Depends on what I get for what amount. The other problem is I will have to buy the desk cabinet and get everything fixed myself, I am not sure whether stack server cabinets are available or not? any link for those?
so I can just make the dealer assemble it instead of trying to do everything myself.
On my system im using a Quadro 2000 for blender renders, but i googled for some opinions and i found that using a GTX video card from nvidia will decrease considerably your render times. So i see that u have a quadro 4000 in your system.With a GTX you will render faster. These cards are the best choice for rendering in blender. Especially the GTX 500 series because the newer ones like 600, 700 series are more optimized for games. Just do a google search about this fact and youll be on 
Cheers!
Another thing that you should always try to do is: break the problem down into smaller pieces that you can composite together individually. For example, a shot like this one could be broken down into:
- Background – a stock shot of a mountain panorama mounted on a billboard.
- Foreground dirt-only texture.
- Rocks and scattered limbs (and the shadows they cast); other computationally-cheap window dressings.
- Clumps of grass (and shadows); other potentially computationally-pricey window dressings.
- Less-important background characters.
- Sidekicks.
- The Star™,and his shadow.
Now, let’s say that we are doing a “truck shot” in which the camera, mounted on a truck, rolls left-to-right to follow the Star as he, also, walks left-to-right. You render one frame of the first four items, at a frame-width wide enough to encompass the entire shot, and you achieve the sense of “trucking left-to-right” by moving the background right-to-left, perhaps at very slightly different rates to give depth perception. Yeah. The corny old 2D-animation trick.
You then render the Star and (separately) the Sidekicks, walking in front of a fixed camera. This being the only component that you actually render more than one time.
“Comp” all of this together, and whaddaya got? You get a shot like this.
Mind you, I don’t know how they actually did it. But they could have done it this way.
When one part of you (or some other person) says, “wouldn’t it be cool to do <X>?” Counter with: “perhaps, but is it necessary?”
Throughout all this process, you look for “cheats” anywhere you can find them. For example, even though motion blur is accurate, the present implementation causes the same data to be rendered multiple times. So, cheat. Superimpose a slightly blurred version of the Star, taken from “two frames ago,” on top of the tack-sharp Star from the current frame. Presto: motion blur. Or, maybe, use vector blur.
Sure, your peers here at BA might cluck and say, “you cheated!” But you got the shot done, and you didn’t spend a day and a half to get one measly second of animation. And, in addition, if you now step back and look at some little aspect of it and wish that it were just a wee-bit different, you have the pragmatic capability to do something about it without “starting over.”
P.S. The “OpenGL Preview” renders are extremely useful for quickly setting-up a shot to see (in great detail, actually) how it’s going to look. And-d-d-d-d… sometimes those “preview” outputs are good enough to use in the final comp. Or maybe they just need some “sweeteners” smeared over the top of what OpenGL did. Look for every chance to do that. Build the “comp” first, and replace the stand-ins with the final versions as each one comes out of the printing-press, reviewing the shot each time.
Yes man I know that but it actually helps me to create scenes in Vue with over billion polygons when ecosystems are used. It has more view port stability.
For the render slaves I will use the geforce ones.
Thanks for these great tips.
I am actually going to cheat a lot, because if I try doing every thing by the book, I will never actually complete rendering my movies.
Hi, with the new Cuda Toolkit 6.5 the GTX 500 series performance get down and memory consumption get up.
GTX 6/700 series perform better.
Next month the GTX 800 are coming, better performance, less power and 8 GB VRAM.
If you can get a used GTX 580 3 GB very cheap, ok, it is still a good card.
Cheers, mib
^^ So that Toolkit 6.5 is nothing more than marketing card played again? Why would they cripple older HW performance which was working on par with a new otherwise?