Will I see a performance increase with this build?

[Eric]

I currently have 3 render nodes, with one of those being my workstation. All 3 computers are identical. They are:

 

Dell Precision T3500 with Xeon E5640 @ 2.67GHz CPUs and 12GB of memory with ATI FirePro V5800 graphics cards.

 

I'm using Vray 2.0 and tested Vray RT today. It's been a while since I tried it and wasn't overly impressed with the speed versus the standard Vray Adv renderer. I'm assuming it's because it only works with CPU on my workstations and seems actually slower with OpenCL with my current graphics cards.

 

Will I see any increase with the following build, perhaps to the point I could render to Vray RT Production renderer via GPU (provided I'm not using Vray displacement, and whatever other plugins aren't currently recognized)?

 

(Home built) Intel i7-3930K with 16GB of memory with (3) Nvidia GTX 570 HD 2560MB cards

If I added a pair of these video cards to each of my 3 current Dells, would they notably contribute to my home-built workstation for an even faster RT distributed rendering? If so, any idea how much?

 

I should also note I will be doing a little work in Premiere, which I believe still only utilizes CUDA, so I guess that ties me to Nvidia cards. However, I may only use it on projects once or twice a year, so if there is something better for Vray RT, I'm cool with that.

Edited June 8, 2012 by Eric

[Dimitris Tolios]

VRay RT CPU is more fledged than the GPU version. The latter is much faster in most cases, unless you have access to multiple systems, as RT CPU can be easier distributed to more than one machine. The GPU version does not support many of the VRay 1.5 and 2.0 options, but for simpler scenarios it can produce production quality renders amazingly fast, or help you setup and test your lighting before you commit to a long CPU rendering session.

 

RT GPU is VRam limited, as you have to fit your whole scene + materials + maps (which are afaik automatically downsampled to 512x512p to save on memory) and the 2.5GB 570s are passable, yet you should test out scenes that match your actual work scenarios to see if that buffer will be enough. Remember that all the cards used by the renderer have to fit the scene, or are excluded by the engine automatically. You don't need to have SLI / Crossfire enabled and your cards bridged for unbiased GPU renderers (VRay / Iray / Octane etc) to recognize and use your cards. Also you might keep your ATI as your display card, and add the GTXs only as computation cards - that will save some extra hundreds of MBs of VRam that driving your displays would take out of the equation.

 

Generally the card of choice for such apps is the GtX 580 3GB. Eventually this will be out-classed by the GTX680 4GB which is not as fast as the fermi 580, but the governing power for most demanding applications (read complex scenes) is usually the buffer memory and not the absolute speed which is already pretty good, while the kepler based 680s are consuming up to 50% less power. That is quite important as in order to drive 3x 570s or 580s alike, you will need to invest in a pretty serious PSU ontop of the already expensive cards.

Unfortunatelly most games do not stress ultra-fast GPUs like the 670/680 even when gaming in 2-3 1080-1200p monitors and "just" 2GBs of VRam, so it will be some time before we will see cheaper gaming oriented cards getting 3 or 4 GBs.

 

In your case, I would try to sort out the card choice 1st, buying a single card to test on your existing workstations, see what the limitations are (and if those actually affect what you plan on doing with them). If 2 - 2.5 or w/e buffer size you choose is enough, its nice to know that the performance in GPU accelerated rendering is usually linearly scalable: 2x cards will give you 2x the performance. This is absolutely true in teraflops etc, but to get half the time after a point you usually need 4x the computation power etc - that's the nature of the beast, as GPU accel. unbiased engines approach all rendering scenarios in brute force mode.

 

Get a card, leave your ATI as the display accelerator and use the nVidia just as a VRay or Iray accelerator.

 

Remember that for the card to start playing part in your workflow, the program needs to load a lot of data into it - perhaps nearly fill the buffer with nearly 2GB or more of geometry, light data and bitmap information, and that does take some time. I don't know the difference between loading those through a SSD drive or a fast HDD makes a big or small difference, but people with more experience could advice you one that.

 

edit: also make sure you are using the lastest VRay 2.0 update, and look up in Chaosgroup's forums for the recommended Catalyst Driver version that works best with the current Vray 2.0 RT engine.

Edited June 11, 2012 by dtolios

[Eric]

Thank you dtolios - I have to admit I've been relatively unaware of the size of some of my models. According to the attached, I guess my model is around 5.5GB? The model was generated in Sketchup, then imported into Max for lighting and final materials. I don't know if the fact that it's imported from Sketchup is what's bloating the file, or if it's all my textures (I'm sure very few, if any, fall under 512x512 in size). The actual file size is around 165MB.

 

Also, my lights are almost all photometric lights emitting from the actual light fixture location, with a few vray plane lights in various locations for additional lighting in dark areas.

 

 

I spent a lot of time researching this weekend, and think I've landed on a final build list, but would like additional feedback if possible. Anyone see any problems with this build? All prices are from Tiger Direct and exclude any rebates that might currently be offered. They are literally across the street from my office.

 

My goal is to build myself a new workstation, then add additional GPUs to my other render nodes for additional GPU DR processing power.

 

Case -$169.99 -Corsair Special Edition White Graphite Series™ 600T Mid-Tower Case

Power Supply -$204.00 - OCZ OCZ-ZX1250W ZX Series Modular Power Supply - 1250W, 80 Plus Gold

Video card - Display - $279.99 - MSI N560GTX-Ti Twn Frozr II 2GD5/OC GeForce GTX 560-Ti Video Card - 2GB

Video card - GPU - $499.99 - EVGA 03G-P3-1594-KR GeForce GTX 580 (Fermi) Classified 3GB 384-bit GDDR5 PCI Express 2.0 x16

Video card - GPU - $499.99 - EVGA 03G-P3-1594-KR GeForce GTX 580 (Fermi) Classified 3GB 384-bit GDDR5 PCI Express 2.0 x16

Hard Drive - OS -$229.99 - OCZ VTX3-25SAT3-240G Vertex 3 Solid State Drive - 240GB, SATA III, 2.5"

Hard Drive - Cache - $84.99 - OCZ Vertex 4 VTX4-25SAT3-64G 2.5" 64GB SATA III

Motherboard - $307.39 - ASUS P9X79 PRO - LGA2011

CPU - $589.99 - Intel Core i7-3930K 3.2 1 LGA 2011 Processor

Memory - $45.99 - Corsair CMZ8GX3M2A1600C9B Vengeance - 8GB (2x 4GB), PC3-12800, DDR3-1600MHz

Memory - $45.99 - Corsair CMZ8GX3M2A1600C9B Vengeance - 8GB (2x 4GB), PC3-12800, DDR3-1600MHz

Operating System - $139.99 - Windows 7 Professional 64bit

DVD Drive - $17.99 - Lite-On IHAS124-04 Internal DVD Writer

CPU Cooler - $24.99 - Cooler Master RR-B10-212P-GP Hyper 212 Plus CPU Cooler

Thermal Compound - $7.99 - Cooler Master Thermal Compound

Case Fan - $9.99 - Thermaltake AF0060 DuraMax 12 Case Fan - 120mm, Dual Ball Bearing, 4-Pin

Case Fan - $9.99 - Thermaltake AF0060 DuraMax 12 Case Fan - 120mm, Dual Ball Bearing, 4-Pin

Total: $3,169.24

Total: $3,430.70 After Tax

Edited June 11, 2012 by Eric

[Dimitris Tolios]

Your build is solid for the most part.

I don't know about the 4x4GB Ram configuration - I would go with 8GB sticks. I know 32GBs are a lot, but since you are dropping such a chunk of money, I would go full (half, you have 4x dimms left!) way. Doubt you will actually "feel" the difference, but I would go for 4 sticks ofc to get quad channel speed - weeeeee and yeah for the placebos that rock us whenever we plan a new build!

 

The 560Ti is a nice addition - It will assist the 580s in lower complexity scenes, and relieve (one) of them for the high, so that you have 100% of the 3GBs available for the computation tasks.

 

Again, if I wanted a new workstation oriented almost 100% to GPU accelerated rendering and did not know my ways with VRay GT (which I don't as I don't have a GPU working with it good enough for me to try it over CPU), I would buy one and "work my way" up, testing if GPU DR would work etc. Daydreaming and spending money in vain is...bad.

 

The possibility for multiple 580s working in DR for VRay gpu is great, but 3-4 Xeon machines slaves on a 3930k will do VRay RT CPU quite fast, DR is more likely to work and you won't have the VRam limitations, while enabling more features. Don't mean to shoot the dream down, just take it easy if there is the slightest probability that your build will have 3x massive cards collecting dust and burning though 350Wh every hour (and $1500 to boot) idling for nothing.

 

Also, OCZ "matched" system would be cool, but I would stick with Crucial m4s and Samsung 830s for SSDs. What is the deal with "cache" disk? To accelerate your magnetic HDD drives etc? You can partition 20GB out of the 256GB OS drive and do the same, i doubt you need a separate drive, and 64GB is not enough anyways. Plus it is slower than the 128/256 drives anyways (less NAND chips in parallel = less through-output).

 

 

Oh, and my b@tt would itch to overclock that beast even a little bit, so i would probably get a better cooler (otherwise the 212 in any version is amazing for the price @ stock speeds).

Edited June 11, 2012 by dtolios