Specs for Modeling PC, NOT Used for Rendering

[erickaiser]

I am an IT Tech that's been given the fun task of reducing costs for my company's 3D modeling team. They do large and complex static 3D models and presentations of buildings, landscapes, but nothing animated. They use 3DS Max, Vray and Photoshop. Every year these guys ask for more and more powerful and extremely EXPENSIVE computers:

-Superfast processors with as many cores as possible

-RAM RAM RAM RAM

-Graphics cards that cost as much as the rest of the PC

 

Their justification has always been the 3D Grunt "We need it. If not, we can't work." and so our company has been buying these guys what they want without question, basically because the decision makers wouldn't know the difference between dual core and hardcore (although my guess is that they're fully into the latter from the traces I can see in the firewall logs.)

 

I've been looking into this and it seems to me that the hardware requirements for Modeling and Rendering are completely different, so a machine designed to do both is obviously expensive.

 

Hence the idea behind Render Farms.

 

So, I plan on building these guys a Render Farm with about a dozen or so of their old PCs, which will give them about 80 Cores to render on. It then stands to reason that I should be able to spec a machine Just for Modeling that is significantly cheaper than one that does both Rendering and Modeling.

 

Sorry for the long story, but here's my question. Everywhere I look, I can find hardware specs, advice and discussions for building cost-effective systems designed for either Rendering or Rendering and Modeling combined, but never Just Modeling alone, without the Rendering part. Why? Am I missing something?

 

Also, could someone please give me their best advice on the following:

i7 or Xeon? Which path should I take for a Modeling PC?

RAM - ECC or nonECC, does it matter for Modeling?

Clock Speed is more important for Modeling than number of cores. True?

How important is disk speed for Modeling?

The graphics card I get ... Just buy the biggest, baddest Quadro or Firepro your budget will allow.

 

Thanks for your time.

[Jeff Mottle]

Before you start making any major changes to their machines, I'd definitely look closer at their workflow. As someone who had butted heads with IT people more times than I care to remember over machine and network specs you do need to be sure you fully understand HOW they are working. I first commend you for coming to a place where you'll be able to get the correct information and not just make assumptions or take the rendering teams word for it.

 

1) How much RAM do their current models consume when loaded? How big are their Photoshop files?

 

Due to the type of models you are describing, they potentially could require a lot of RAM just to load the models into 3ds Max. Also, depending how many layers they are setting up in Photoshop could also change their RAM requirements and even drive requirements for that matter.

 

2) How are they interacting with them in the viewport (wireframe, shaded, active shade etc), previewing animations?

 

When they are navigating these models in a viewport do they have to hide parts of the model to allow them to easily navigate? What is the rendering method they are using to shade the viewport (Nitrous, OpenGL, Direct3D)? Are they ever using active shade with something like V-Ray RT? If they are, then this requires powerful GPUs and often many of them per machine that can not be put into a farm.

 

3) Do they make test renders or region renders during the process

 

If they don't have a farm now, I'm sure these guys will love you for it, but depending on their workflow, they are more than likely still going to need to do at least some of their rendering locally. Are they doing any sort of distributed rendering or network rendering now?

 

If it were ONLY a modelling machine depending on their version of 3ds Max and how large these models are, will greatly affect how you set them up. CPU could certainly be cut down dramatically as you would not need it for rendering, but also depends how they render. 3ds Max has traditionally been a VERY CPU bound application that usually did not leverage the video card much except in wireframe mode. A lot of progress has been made in recent versions, but 3ds Max version, model size etc will affect the type of video card you can get. Generally you get very little benefit with the professional cards (Quadro vs GeForce), but again depends on the specific case. If your guys are not doing GPU rendering, then getting the biggest and baddest video cards is a complete waste of money. For many years a decent higher end gamer cards is about as much power as you'd need in 3ds Max. Buying a more powerful card would not net you any more gain. Again, things have been changing a bit, but the majority of advancements in cards you are seeing now are on the GPU compute side which is for , among other things, GPU rendering (V-Ray RT, iray etc.)

 

Get some more details about their workflows if you don't already know and do a quick interview with them to find out how they are working and how they would like to work. Post back with more details and I'm sure many here will be able to also chime in.

[Scott Schroeder]

Stop buying them new computers every year, that's step 1 to save money. With a well invested build, you probably won't have a need for upgrades or new systems for at least 2-3 years and maybe longer.

 

You can also stagger the machines if you have people doing set tasks. If you have the more senior workers dealing with the heavy scenes, they get the higher end boxes. The new guy modeling rocks, he gets the older and slower ones because he doesn't need that powerful of a system.

 

I've never known anyone to use max and never hit the render button at some point, so you may want to have machines that can still get them test renders in a fairly decent amount of time. The other side of the argument to getting them top of the line computers is that, while you may build them a decent daytime render farm, when everyone goes home at night you now get a very powerful nighttime render farm when you add in all of the user boxes. Most of the rendering tends to get done at night when people send stuff off before they head home, so that's generally when you need full power.

 

The entire other can of worms is that maybe the team needs to look at their process. All too often people just toss more power at really horrid rendering settings and modeling pipelines. You have people sitting there rendering tests at 140,000 pixels and using a final render preset. You have the modeler that makes a 4.6 billion polygon chair that never is more then 1% of any scene.

Edited April 22, 2014 by VelvetElvis

[Dimitris Tolios]

Re-purposing old machines that you already have is not a bad idea - the contrary.

Just don't drive expectations up: old architectures are sometimes much slower than we think: a 5-6 yo dual Xeon machine with 8 cores, might be significantly slower in modeling and notably slower in rendering vs. a modern i7 with 4-cores and HT (8-threads).

 

That dual Xeon will also draw much more power than a relatively current s1155/s1150 system, while doing "less" (i.e. being slower).

So having a few dozens of old systems like those on "standby" just in case is one thing. Relying on them running 24/7, might turn around and bite you in increased power bills - you also have to take into account that multiple machines require an extra OS license each (which you might have) and an extra rendering node licence each. That might add up to a few $100s just in soft-costs. Along with power efficiency/savings, getting "more per box" tends to favor newer systems.

 

Q an A:

• i7 vs. Xeon: both share the same architecture give or take. There is no "magic" in Xeons. You need Xeons for ECC or systems with more than 1 CPU or more than 6 cores.
   
  
• RAM: ECC is not that important anymore - in my opinion. Not only in modeling, but for nodes and most stuff that are not very delicate calculations.
   
  
• Clock speed: yes, it is pretty important / more important than # of cores. Most of the operations we are dealing with while modelling in 3DS/Maya/C4D etc are single threaded. And that is not because devs are lazy - not everything can be multithreaded, and not everything can be scaled to multiple threads without diminishing results after 3-4 cores are thrown into the mix. Image Rendering/Raytracing is one of the exceptions, and for the rest utilizing the parallelism of modern GPUs is more promising than increasing core count in CPUs.
   
  No matter what happens, nomatter how much CPU cores or GPUs you are using, you need 1-2 "master" CPU cores that synchronize the rest, delegate tasks and keep all the resources (questions and answers) in check. So single thread/core performance will be important for a long time, unless the way we think of computer code changes radically.
   
  
• Disk speed is important. After you experience the speed of a decent SSD in launching apps, reloading assets, saving (anyone pissed waiting for auto-saving which happens always the worst moment?) etc, you won't turn back easily.
   
   
  
• The benefits of a faster card are relative to the programs (and program version) used and the scene complexity. There are hugely diminishing returns after a price point.
  

[Tommy L]

Really? Staff wanting RAM? Compare the cost of salary against RAM. Give them the RAM.

That aside, yes, your staff need a render farm. If your company has a 'team' that does 3d and a IT guy (yourself) who has the time to be tasked with reducing the running cost of the 3d team then Im mighty surprised they dont have a farm already. Do these guys just sit around watching their workstation render? If so then you can just build a farm and leave their workstation specs just as they are right now. I model on a 3 yr old machine, its a i7 920 with 24G of cheap RAM and a GTX 580(3gig). Its totally slick, I need for nothing. I also have 12 x i7 nodes sitting next to me which carry the rendering load.

Ive forgotten how may times Ive said it... get the rendering tasks off your workstation. If you invest in a decent farm it will be like doubling your team. Or get them all rendering off the Amazon cloud....

[Chris MacDonald]

Nothing much to add really other than there is no way you need a new PC every year IF you buy a decent spec one to begin with.

 

Provided you are absolutely sure they don't need to be doing any rendering (keep in mind test renders, very important) then a dual/quad core with the highest clock speed possible is probably the better option as almost everything modelling wise is single-core. RAM is cheap so buy plenty because the chances are they aren't lying when they tell you their models are massive, but anything faster than 1800MHz and you're into the land of diminishing returns for a modelling PC. And finally as you stated, the biggest quadro/firepro GPU you can afford though again speak to them about their workflow, because they might be using the GPU to do their test renders.