Forum Moderators: wheatpenny, TheBryster
Vue F.A.Q (Last Updated: 2024 Nov 26 6:57 am)
Hmm.
Firstly, the Atom was designed for netbooks and appliance usage, so you probably are not dealing with a processor that has the math unit you need (CG needs processors that excel at floating point calculations). It definitely has a small L2 cache. The memory controller is in the northbridge, not on the processor, so that is going to be a hard smack on your throughput. The second smack will be the swapfile thrashing; with that small amount of RAM (which the OS will claim most of,) nearly all of your scene data will be loaded onto the hard drive, not kept in active ram. Which will add the access time of the drive to the penalty of having an old style frontside bus for memory access.
While it looks as if it does support hyperthreading, that only matters if the OS and the application -both- support it as well. The current Vue 8 rendercows support multiple threads, but Vue 6's only used 2. Some of 7's were limited to 2 as well, while the upper end cows (Infinite and Xtream) could use more than two threads. Anything 32bit only used 2 threads, I believe (but could be wrong about that). You =will= have thermal issues; CG is murder on hardware, and this class of hardware simply was not designed to handle this kind of load. You would have to have an oversized heatsink, or added active cooling to keep the chip from going into thermal shutdown to protect itself.
I would recommend either going AMD quadcores, or it you insist on Intel, the i7 (you get out cheaper with AMD though, and the tech in the Athlon X4 and Phenom II's is based on Alpha chip technology). I'd love it if my renderbox chips only pulled 8 watts each, too.....but consider that the ATOM is at best a single generation above the current field of desktop processors....all of which draw 60 watts minimum, and 95 to 125watts is common. That doesn't mean that your desired chip will only turn in 8% of the work that a mainstream processor will, but it does give you an idea of some of the things that -aren't- included in the ATOM. People have tried to put laptop chips in desktops, and it has rarely worked as expected, simply because the chip wasn't designed for anything high stress.....
I should have mentioned I was trying to keep both the:
physical space
power usage
and the price to a minimum.
The Atom 330 built into the MB in a small formfactor case takes minimal space, only 60W power and a little over 300$ AU each.
I can get 3 or 4 of these for the price of an I7 and still have plenty to spare of all three above points.
My main machine is an I7 with 9gig of ram, I'm trying figure out what extra rendering power I can add to this.
The only thing that worries me is the max of 2gig of ram, I'm wondering if I can minimise the OS install?
Also I was wondering how smart Rendercow is, will it just throw as many threads as the hardware will support at a render or only as many as the memory will support.
Honestly this is a terrible terrible plan. Render farms run as background services and never hit 100% utilization. So ONE i7 (2.8 to 3 ghz) running all 8 cores at 100% utilization (under a 64bit OS) as a standard render will still render twice as fast as 5 atom based cows. NO contest, not even close. Did i meantion terrible idea?
None of the desktop hardware out there has a scheduler that is intelligent enough to handle the kind of resource per thread scheduling you want; that is strictly still in the server and supercomputer camps. Currently, the hardware and software both handle first come first served; the cows themselves have hard coded priorities, so each additional thread has to take what resources are available.
If you build render faarm couple thing to know:
Problems with Atom, it's not design for number crunching, to slow. Minimum use i7 or Xenon.
Small factor big minus, when you load 90% for 24/7 overheating a big problem.
AMD use to be good, we use them at my last job for cluster supercomputer like this:
http://www.lbl.gov/CS/Archive/news081605.html
and i still have some opteron's in production, but at this moment, intel Xenon give best preformace. Core i7, cheap, easy to overclock, very stable. AMD big heating problem, lot power consumptions.
Also we build third fastest supercomputer in world from desktop hardware. (linux networx)
With out going brook, you can put high preformace nodes around $400
I have a small "renderfarm" myself and am using a Mac Pro for the main computer while the other two systems are my render nodes. Right now I have it setup with Windows 7 64bit on all computers and I can tell you it is VERY fast and helps with the LARGE scenes or animations.
But the answer is not going to be one you want to hear I am afraid. While one of my systems is an older AMD FX55 running Windows 7 64bit with 2GB, the other is an I7 with 6GB and the main system, the Mac Pro has 10GB. Therefore you are going to need at least one VERY good main system and one very decent 2nd system. The others won't really slow you down that much, or at least should not and are simply added to the chain. Sort of think of it like SETI At Home type of use where a lot of computers peck away at a task.
The bottom line is the other posters are right and I would advise you to just save your money for better hardware because the specs you posted will NOT begin to cut the mustard for anything very useful. Sorry, but that really is the best advice for right now given what you have posted.
Jeff
Development on: Mac Pro 2008, Duel-Boot OS - Snow Leopard 10.6.6 &
Windows 7 Ultimate 64bit, 2 x 2.8 GHz Quad-Core Intel Xeon , 10GB
800 MHz DDR2 RAM, NVIDIA GeForce 8800GT.
There was huge article in Tom's Hardware:
How To: Building Your Own Render Farm:
http://www.tomshardware.com/reviews/render-farm-node,2340.html
Hope it helps.
This site uses cookies to deliver the best experience. Our own cookies make user accounts and other features possible. Third-party cookies are used to display relevant ads and to analyze how Renderosity is used. By using our site, you acknowledge that you have read and understood our Terms of Service, including our Cookie Policy and our Privacy Policy.
Hi I'm currently planning on building a small renderfarm at home using small formfactor barebones systems.
I'm considering the ATOM 330 but all the motherboards only seem to support 1 slot giving a maximum of 2 gig of ram.
The atom is dual core with I think Hyperthreading which will mean four threads, split two gig by four threads leaves only half a gig per thread.
I'm pretty sure that wont be enough memory and the system will start thrashing the hardisk and slow down.
Does anyone here have any experiance regarding this, I want to keep the memory to about 1 gig per thread/core.