PC for heavy Computer Graphics and Animation

M

maxmax

Member
Hello everyone,

I'm trying to build a fast machine that I could use as my main workstation to do 3d graphics (3ds Max, Cinema4d) and compositing (Nuke, after effects). My main two questions would be:

1. which processor would be faster - 6 core i7 extreme or the Intel® Xeon® E5-2687W? A lot the times I'm doing lighting and final 3D rendering and I just need to get fastest preview possible, so would 8 core Xeon render the frame quicker then i7... I know you could potentially overclock that i7 to make it run even more faster, what about Xeons? Can you overclok them?

2. Would sticking another Quadro 4000 graphics card would make any difference for lets say Vray RT or any other program that has GPU accelarated stuff?

Here is my specs that I put together.. Feel free to share your thoughts and opinions..

Thanks

Max

Case AeroCool X-Predator Evil Black Full Tower Gaming Case
Processor (CPU) Intel® Core™i7 Six Core Processor Extreme i7-3970X (3.5GHz) 15MB Cache
Motherboard ASUS® P9X79 WS - SOCKET 2011, QUAD DDR3, USB 3.0, SATA 6 GB/s
Memory (RAM) 32GB KINGSTON HYPERX GENESIS QUAD-DDR3 1600MHz X.M.P(8 x 4GB KIT)
Graphics Card PNY QUADRO 4000 2GB GDDR5, 256 CUDA CORES, 2 x DP, 1 x DVI-I
Memory - 1st Hard Disk 240GB KINGSTON HYPERX 3K SSD, SATA 6 Gb/s (upto 555MB/sR | 510MB/sW)
2nd Hard Disk 480GB KINGSTON HYPERX 3K SSD, SATA 6 Gb/s (upto 540MB/sR | 450MB/sW)
3rd Hard Disk 3TB WD CAVIAR GREEN WD30EZRX, SATA 6 Gb/s, 64MB CACHE
Memory Card Reader INTERNAL 52 IN 1 CARD READER (XD, MS, CF, SD, etc) + 1 x USB 2.0 PORT
Power Supply CORSAIR 750W PRO SERIES™ HX750-80 PLUS® GOLD MODULAR (£119)
Processor Cooling COOLIT ECO II C240 DUAL RAD LIQUID CPU COOLER (Special Offer) (£55)
Extra Case Fans NONE
Fan Controller NONE
Sound Card ONBOARD 8 CHANNEL (7.1) HIGH DEF AUDIO (AS STANDARD)
Network Facilities 2 x ONBOARD 10/100/1000 GIGABIT LAN PORTS
USB Options 6 x USB 2.0 PORTS @ BACK PANEL (MIN 2 FRONT PORTS) AS STANDARD
Operating System Genuine Windows 7 Professional 64 Bit w/SP1 - inc DVD & Licence (£109)
 
Frank100

Frank100

Rising Star
Hi,

I'm waiting for my new Xeon 2687W order to be approved by the bean counters at work but I am expecting this PC to be a real monster. With hyper-threading enabled it will have 16 cores on a single chip and all will run comfortably at 3.1Ghz hour after hour and day after day. It's an industrial strength chip designed to operate pretty much flat out 24/7. It will auto overclock in certain situations, such as single thread tasks but it isn't something you'd normally overclock. In terms of speed it should also be noted the difference in instructions sets between i7s and Xeons. In certain tasks such as hashing, AES and Julia it will outperform chips running at higher clock speeds, be some margin.

I know the i7 chip you've listed is also a monster but it's cheaper partly because it isn't designed to run flat out all of the time. If this is ok then this chip is probably perfectly fine. It's about £600 cheaper so there's a price difference.

In terms of GPU the Quadros are based on the older Fermi chipset and are due for replacement very soon. I have a Quadro 6000 in a PC at work and for the GPU tasks I give it a GTX680 in the same PC is a little faster. I'm doing high end number crunching rather than rendering but it all uses CUDA and Open CL code for GPU acceleration.

Clearly the tasks I am doing are utilising double precision floating point calculations because the Quadro (which is good at that) is doing close to what the GTX680 is doing with over 1000 core less and with the cores running at a lower speed.

The Quadros were designed more for 2D rendering initially and as the current ones are on their way out soon it might be worth checking what your various tasks will utilise. Although GTX cards are much slower on certain tasks their higher number of cores and faster clock speeds make up some of that difference and you can add multiple GTX cards together for the price.

If you went down this route it is likely you wouldn't SLI them. The programs will most likely make better use of them if it sees them as separate GPUs. Certainly that's the case with what I do. It's worth you looking in to this. Bear in mind more cards means a bigger case and a bigger PSU as well. My order is for the Cosmos 2 Ultra case which can hold four double width cards.

According to my sums the GTX660ti offers the best mathematics capability for the price in the GTX range. I'm not a fan of the 660ti for gaming but it offers better number crunching per £ than the 680. For example you can get 3 660tis for the same price as 2 680s and the 3 will be faster than the 2, just about.

Another option is the Tesla card. The Tesla cards now out are based on the new Kepler chipset with the K20 likely to be about £2500+VAT but it has over 2000 cores and has serious double precision capability. Again this comes down to your particular programs and what they will need. You can always post some links to the forums for people to look at and advise.

You might also be able to use ATI cards. If the programs aren't using Nvidia's CUDA then you could find ATI cards will work well. From a pure mathematics point of view the ATI cards completely blow the Nvidia cards out of the water. For example the ATI 7970 can do 947 GFLOPs of double precision (floating point operations per second) whereas the similarly priced GTX680 can only do about 350 GFLOPs double precision. These stats don't matter when it comes to gaming but with the way things are now the only thing holding ATI back from dominating the super computer world is lack of support for GPGPU computing. Simply put Nvidia got in there and ATI didn't do it quickly enough.

You won't be able to mix and match GTX cards with Tesla or Quadro but you can mix and match a Tesla and a Quadro together. The drivers are different and it simply won't utilise the cards fully if you mix GTX with non GTX, I have tried. The Tesla and Quadro use the same driver but with one being Kepler and one being Quadro that might not be the case anymore. I've mix and matched Fermi Tesla and Quadro and that's definitely ok.

Obviously a new Tesla K20 is a lot of money but it will give you many times the performance of the Quadro 4000 if your programs will support the Tesla's instruction set adequately.

I'd say as it's a serious purchase you make sure you are getting best value for money and that means checking the details of what programs will support which GPU type. I'm happy to answer more questions if you can point me in the direction of some small print for the programs you will be using.

One last point. You've picked a large and very expensive SSD. Will you really need 480GB? Dropping to a 240GB will save you a lof of money. If you were to rotate files onto the SSD that you are processing and then move them off you might be able to cope with less space.
 
M

maxmax

Member
Hey Frank,

cheers for your very in-depth reply!

For the CPU I would really value more the quickness of calculating one frame quicker rather then the ability to run non-stop for ages. I mean I do leave my current workstation to render overnights but the really render heavy stuff goes on the hired renderfarm anyway. So in this case my biggest goal is to get through those one frame test renders quicker so I can submit the rest of the sequences to the renderfarm once I'm happy with the look.

When it comes to video card I'm mainly using 3Ds Max for all the 3D at the moment so my main use of GPU will go on using Vray RT for previews. In all the stats that I've seen online Autodesk seems to love Quadro cards, but its not the first time I hear that you can get really good results by combining GTX cards, so do you think I could get better bang for my buck by combining GTX cards rather then going with Quadro..? btw, can you combine two gtx660ti's and lets say gtx670? as it seems I can't select gtx660ti as my third video card. Tesla seems like a really nice option but I think its bit too pricey for me at the moment.

as for the SSD, I know its pricy, but I notice that I keep getting these projects where I need to output and comp these huge multipass EXR files, and the project ussually exceeds the 240gb limit. so my idea behind that is that I would have a large SSD as my fast project drive and once I finish the project I can then move it to back-up on a separate drive.
 
Frank100

Frank100

Rising Star
Max,

I've not used the particular program you are using so I have to go with generic knowledge and doing sums on stats. I know myself that predicted results based on theory don't always add up as you'd expect. So I think if you aren't in a hurry I'd like to do a bit of sounding out over the weekend and I'll have more time to provide information then.

In terms of joining Nvidia cards together different cards have different limits to numbers of cards in SLI. The 660 for example can run 2, the 670 can run 3 and (most) 680s can run 4. The 660ti is based on the Kepler 104 chipset so I'd have expected it to run 3 as well. The 660 uses the Kepler 107 chipset.

However, not all GPGPU tasks require or even benefit from cards being in SLI. Which means you could in theory have 4 GT650s (which can't SLI at all) and get the combined power of the 4 cards. It's another thing I can look into and find out.

If you are running cards in SLI they all need to be the same type. As for mixing and matching if you aren't going to SLI I don't see why it wouldn't work. I haven't tried it myself but I will be testing it at work when I've got a handful of new cards in to play with.

Certainly GTX cards can do what you want and certainly the Quadro was a great card but I'm not sure its worth the money anymore.

If you can give me a couple of days to do a bit of snooping I might be able to say something more solid.
 
Frank100

Frank100

Rising Star
Hello I'm back,

I've been doing some scouring of forums, reading tech specs and other things and I think I've got useful answers for you.

3DS Max and Cinema4D are going to be more CPU intensive and will respond well to more cores and faster cores. There is a lot of use of SSE 2 instructions and all processors these days support that. I couldn't find any stats comparing i7s to Xeons on the sorts of things these programs will be doing but I would imagine the i7 will be good enough considering its price advantage. From the graphics point of view it uses quite old Open GL standards and you shouldn't have any issues with current cards.

Cinema4D recommends Quadro cards but I don't think they've updated the page very recently on that point and I sometimes think its a bit of a case of scratching each others backs!

Vray RT is able to use GPU support and indeed the beneficial results can be seen by this test. http://www.spot3d.com/vray/images/stuff/rt_gpu/ This was using a GTX480.

There is some other useful info here. http://www.spot3d.com/vray/help/rt100/render_gpu.htm

Vray RT uses Open CL but doesn't appear to support ATI cards yet. What a shame because ATI cards will seriously rock in these sorts of tasks.

So it looks like Nvidia is the only way forward for you.

I've seen threads suggesting larger on-board memory is advantageous but I can't quite understand why when I read what Vray RT is doing. It's never as simple as all that anyway because you have to factor in memory bandwidth speeds as well.

Another unknown factor is how much support has been written for the newer Kepler chipset cards so far. Older Fermi cards like the GTX500 series and the current Quadro are getting hard to find now, I doubt they are still making Quadros and they definitely aren't making GTX500 series any more. The developers might be a little behind so I would expect improved performance from the same Kepler cards in the months and years to come.

A Quadro 4000 is about £600 new at the moment. You get 256 cores @475Mhz, 2GB of memory @ 750Mhz and a 256 bit memory bus. Yes it hasn't been nerfed but that isn't really that much to start with.

Here's the alternative. 2 x GTX660ti. You get 2 x 1344 cores @ 915Mhz, 2 x 2GB of memory at 1502Mhz (6008MHz effective) and 2 x 192 bit bus. So it will have been somewhat nerfed for computational work but it's starting position is so much further ahead. Price for 2 of these cards is about £460.

I've gone for GTX660ti cards over the GTX670 because I don't think memory bandwidth is going to be a significant factor. The cards are otherwise pretty much identical. The price difference between the two cards is about £60 per card.

Of course if you fancied doing some gaming on a break from work the GTX670 is far superior to the 660ti on that front. If you did go with 2 x GTX670 it wouldn't be the worst £120 you'd have spent. You'd still be spending less than 1 x Quadro 4000.

It would appear you shouldn't SLI the cards (if you go for two or more) for VRay RT. If you did a quick bit of gaming you would need to for both/all cards to be utilised.

Without having used these programs myself and without doing any testing I feel reasonably confident that a PC based on the i7 extreme 3970X with two high end GTX cards installed will represent excellent performance without having to pay the premium cost associated with 'premium' hardware.

As you won't be using SLI you could add up to 4 cards if the case and motherboard will support that many. You'd also need about a 1200W PSU for 4 cards.

I think for 2 cards an 850W would be about right. You could even go to a 3rd card with that PSU.
 
M

maxmax

Member
hey Frank,

Big thanks for your advice buddy! I think I'll roll with the two GFX cards and the i7 then.

Regards
 
You must log in or register to reply here.
Top