I need renderslaves and have no clue where to start!

[yyk]

Hi all,
My rig broke down 2 days ago and since I need to order new components I thought this could be as good time as any to buy some renderslave machines. I've never done this before so I have a few questions and would appreciate it if someone could help me out.. I have no strict budget, I'd be looking at investing in 2-4 rigs with approximately the same efficiency as my current 2 year old Xeon5355, 8 gigs RAM.


1. I was looking at a thread on general section here: http://www.chaosgroup.com/forums/vbu...ad.php?t=44069 and was wondering would I still get the best value for my money buying slaves with i7 processors? if not, what would you recommend? I usually build my computers myself but TBH if it was an option without being to expensive I'd rather buy ready made builds this time around.

2. Many times when working on my current workstation, dual Xeon 5355 with 8 gigs of RAM and a Radeon 8800GTX, I managed to make 3dsMax completely unresponsive and eventually crash and cause excessive harddrive swapping when I tried to, for example, create closeup trees with very high polycount (around ten million) so I'd like to know is this something I could work around by buying more RAM or a better graphics card? or is this a software limitation? working on 64-bit Windows and Max.

3. Say RAM really was an issue here and I'd invest in a workstation with 16 or 24 gigs of RAM and worked on scenes that actually needed all the RAM available, to be of any use my renderslaves should have roughly the same amount of RAM, right?

4. For renderslaves I should look for processor/memory speed and get as cheap graphics card, monitor etc as I can find?

5. Setting up renderslaves like this is really nothing more than connecting computers through LAN?

6. Apart from the hardware, to get this to work am I overlooking something here? I need Windows for every comp but any other software I'd have to be prepared to invest in?

Thank you in advance, I'd really appreciate some help here!

[zoobadoo]

Hi,

1) didn't read the thread, but we have big fun with our latest 6 servers based on i7 (920 i believe) with 12Gb ddr3 mem. OC'd to about 50% renders fast, stable and very cheap. Dual Xeon boards are faster of course, but calculate the gigahertz v.s. $. I like building them myself as you know exactly what's inside and the hardware is more flexible then dell or hp prefab's.

2) there are some threads on optimizing max scenes and ram usage. more the better, but has a price tag... I still manage all our arch viz (pretty heavy stuff) on 8-12 Gb xp64 by using the right strategies. as far is i know the graphic card in a render slave does not matter at all. I only use it to install the pc.

3)right. actually i think workstations could be a little lighter than the renderslaves as you can use lighter settings/testing scenes on them. start on 12Gb, you might upgrade later to 24Gb.

4) right. but buy only one monitor and keyboard, both with long cables to connect to the slaves for installing. afterwards use remote desktop to acces them for trouble shoot or updates. and of course setup the backburner.

5) right. but use a fileservere when more then 10-12 pc's are connected as xp has connection limits. we use a linux fileserver so we don't need windows server. saves $. put this budget in really good net switches. and a seperate fileserver is good to setup from the start. No files on the workstation, centralize everything (especially libraries) right now to prevent a unc mess later on.

6) not really. though, when using a lot of pc's noise and/or heat will get an issue. we have 18 servers currently in a separate room, so thrustworthy airconditioning is needed. also the electricity demand can get too high for your electricity network. I calculate at about 250-300 watts per rendernode (@100%cpu usage).

[yyk]

Hi,

1) didn't read the thread, but we have big fun with our latest 6 servers based on i7 (920 i believe) with 12Gb ddr3 mem. OC'd to about 50% renders fast, stable and very cheap. Dual Xeon boards are faster of course, but calculate the gigahertz v.s. $. I like building them myself as you know exactly what's inside and the hardware is more flexible then dell or hp prefab's.

2) there are some threads on optimizing max scenes and ram usage. more the better, but has a price tag... I still manage all our arch viz (pretty heavy stuff) on 8-12 Gb xp64 by using the right strategies. as far is i know the graphic card in a render slave does not matter at all. I only use it to install the pc.

3)right. actually i think workstations could be a little lighter than the renderslaves as you can use lighter settings/testing scenes on them. start on 12Gb, you might upgrade later to 24Gb.

4) right. but buy only one monitor and keyboard, both with long cables to connect to the slaves for installing. afterwards use remote desktop to acces them for trouble shoot or updates. and of course setup the backburner.

5) right. but use a fileservere when more then 10-12 pc's are connected as xp has connection limits. we use a linux fileserver so we don't need windows server. saves $. put this budget in really good net switches. and a seperate fileserver is good to setup from the start. No files on the workstation, centralize everything (especially libraries) right now to prevent a unc mess later on.

6) not really. though, when using a lot of pc's noise and/or heat will get an issue. we have 18 servers currently in a separate room, so thrustworthy airconditioning is needed. also the electricity demand can get too high for your electricity network. I calculate at about 250-300 watts per rendernode (@100%cpu usage).

Thank you for your answer. Since I have a project with deadline tomorrow I had to buy one i7 920 since as I said my old motherboard died, and I'm very happy, the speed gain is really noticeable and I haven't even overclocked it yet! and this is the first time I build a rig of my own (took less than an hour to assemble) and got it right the first time! Again, really happy with the i7 920

Also to clarify, on point #2 I was actually referring to the main workstation (confusing since the thread is about rendernodes), I'm still curious about the graphics card.. for closeup shots I like to make foliage as detailed as possible but am having really hard time managing Max while creating trees with GrowFX, maybe I'm just pushing it too far.

[jonahhawk]

For reliability, speed, power savings and general ease of use, I recommend the Boxx Render Nodes. for around $2800 you get 4 XEONs in one Unit. It is actually two complete, dual proc nodes per Unit. We have five of these at work and they have been very reliable and fast. We don't have the i7s but our new workstations are i7s and are about 1.8 times faster (than our older Render Boxx nodes) with the Hyperthreading turned on (16 cores).

[werticus]

get: no monitor, no keyboard, no mouse, no dvd drive
cheapest possible hdd, cheapest possible video card.

i7 920 a $50 cooler better than stock - overclock to 3.5 - 3.8ghz
12gb ram (any ram does not need to be fast, this does not really affect much)

Just have a USB DVD drive for all of them to share the one time you actually need it.
Windows7 64bit.

Bios options allow the machine to boot with no keyboard, and you can just install VNC or other remote control software and control them from your workstation. (maybe have one set of keyboard and mouse and small portable lcd to set them up initially)

Set the machine to auto login, and auto start backburner and DR slave software, and i like to auto start a explorer directory looking at the z:\maps directory so its polled the network right away.

Easy!

[dellis]

I've been thinking of building another slave and with all that's going on with the VRay RT stuff, it makes me re-evaluate whether or not i should be loading up the slave with good video cards as well. If the future direction is in fact using the graphics card processing abilities that surely effects how one should approach building. I hope Vlado or someone chimes in on this...cause i wonder if its a good idea to get motherboards that can handle multiple graphics cards. My workstation can take three running in SLI mode...which I'm not sure if the graphics card version of RT uses. Anyway, hope someone has some input on this.

[beestee]

Video Card tech advances so fast that it would probably be best to wait until VRayRT GPU usage is set in stone and available for use before stocking up...if nothing else, the GPUs that you are considering right now may very well be considerably cheaper once that usability has opened up.

EDIT: It seems to me that SLI basically converts multiple GPU's into one, it takes care of the overhead of unifying the processes of the multiple GPUs...so logically an application coded with CUDA should be able to treat an array of cards in SLI configuration as it would a single card, but with many more pipes for handling the data it is being fed. It seems pretty logical that this would work as you would expect. Of course I am not a programmer, and I have not looked at the cost differences for a motherboard that can support parallel speeds over multiple slots, but I think it is a pretty safe bet to go ahead and get the mobo with that capability in anticipation that even if it doesn't work out of the gate, it will in the not too distant future.