Upgrading older PC for graphics use

[Stephen Thomas]

I currently have two PCs I use at work. One is great, an HPZ800 configuration with dual 6-core Xeon 5650s and a Quadro 2000 + Tesla C2075. I find myself using it 95% of the time. The other is its predecessor, a Boxx machine with dual 4-core Xeon E5420s and a Quadro FX1700. It really isn't up to scratch even as a render node anymore, I use it mainly to pitch in on distributed rendering to speed things up but that's about it.

 

All of which doesn't do much for my productivity when I have to have renders going during work time. So I propose to make the less powerful PC my main post-production workstation, with the HP one dedicated to majority of 3ds max use, Vray RT and rendering.

 

I need to do some upgrading though, the OS for starters as After Effects CC won't install on XP64. So apart from that what should I be looking to do as a minimum? Graphics card is a must I would think. Budget is whatever I can get my boss to agree to as necessary to improve productivity. Current specs below.

 

Intel Xeon E5420 (Harpertown) @ 2.5GHz x 2

Supermicro X7DWA mainboard

Seasonic 700HM

8GB (4x2Gb) PC2-6400 (400MHz) Kingston RAM

Nvidia Quadro FX 1700

Eizo Flexscan S2242W display

Edited December 19, 2013 by stef.thomas

[Guest dialog]

Well I know people may lynch me for this...but I just got the new K6000 and I must say my workflow efficiencies will have the cost of this card paid off in a month...if not less. (I can now use VRay RT) in mere seconds to do materials and lighting and set camera views for clients. On top of the increased viewport performance.

 

Now I am coming off a Quadro 2000 in my old machine... and the difference between the 2 is unreal (most noticeable is the VRay RT speed)

 

Just my opinion from using both cards and my workflow. Also is project dependant and software dependant too as to what level of hardware you need to run. Either way, I have found we will push our hardware to the absolute limit

[Dimitris Tolios]

Not really sure what you want to do with "PP", and what kind of money you want to spend doing it.

The machine you have is old and relatively slow, ok, but is that what you plan on doing that intensive?

 

Asking cause out of what you have little is salvageable, but in general PP work is not that crazy intensive if we are talking editing stills etc.

Even complicated composites in PS, assuming 5000 x 3000 outputs will probably be workable with your 8GB of RAM.

 

The GPU is also a joke, but again, what kind of GPU accelerated tasks will it be called to fulfill ?

 

So, if PS / AI etc is all you are going to use, you might get away with just the OS, and maybe a cheap 128GB SSD for the basics - placebo to give the thing some oompf - but my general approach would be "try to spend no money on it" (other than the OS).

 

If it works, it works. If it doesn't, few things are salvageable (DDR2 RAM, old & slow HDDs and whatnot), and perhaps from en economic perspective the best thing would be to sell it as-is, instead of parting it out.

 

The PSU is ok, so is the case, but the PSU is one of the components that get old with time (chemical capacitors) and I would probably replace it too with a fresh one if I was to build even a mid-range machine.

 

Again, what will the thing be realistically called to do ?

[Stephen Thomas]

I agree, the Boxx is definitely old hat. I had a look inside today, it's pretty sad in there! The more I think about it the more I'm actually thinking that I should change my approach. Maybe what I should do is keep the HP as what it is, a good all rounder and try to make the argument for getting a kick ass render node. It works well for modelling and RT, plus my post production software (After Effects, Photoshop, Magic Bullet). I just need to not be tied down to relying on it as my main renderer too. At least then I'd have two decent rendering machines when I'm in a pinch with deadlines approaching.

[Zdravko Barisic]

Well I know people may lynch me for this...but I just got the new K6000 and I must say my workflow efficiencies will have the cost of this card paid off in a month...if not less. (I can now use VRay RT) in mere seconds to do materials and lighting and set camera views for clients. On top of the increased viewport performance.

 

Mainly in what field?

VIEWPORT or VR RT?

Thanks!

[Guest dialog]

The viewport is fast! but it still has the 3DS max issues to deal with and can only be so fast. I saw the payoff in VRay RT...i no longer need to wait for "test" renders. WOrk in almost real time. IT has been an invaluable tool for setting up quick and easy cameras for clients to see views and then going to the CPU for the production stuff (although I must say the 3.0 Beta really has dealt with the noise issue and could arguably be used for production)

[Zdravko Barisic]

Thanks Desmond!

I am still on Quadro 4000.

 

@Stephen Thomas

 

What are you planing w/your old PC?

In eastern Europe and Russia, that kind of machine, still has very respectable value, believe me.

...

You can make some kind of contest or whatever, for winnnig award of your PC, or sell it via PP to someone in that europe countries, for some small amount of money. Perhaps to make someone happy, in other words said...

[Stephen Thomas]

So I got the go-ahead to spec a new PC and did the build at the weekend. Got a decent budget to work with, so managed to build a pretty powerful workstation in the end. Spec below.

 

CPU: i7 4930k (OC'd to 4.6GHz with Asus 4-way optimization)

Mobo: Asus x79 Deluxe

Ram: 32Gb Corsair Vengeance LP 1600MHz

GPU: EVGA GTX770 SC ACX 4Gb

Case: Fractal Design Define R4

PSU: Corsair RM750

SSD: Samsung 840 Evo 500gb (job files stored on network server, so no HDD required)

Cooler: Be Quiet! Dark Rock Pro 3

Additional Case Fans: Be Quiet! 140mm Silent Wings 2 PWM

 

All components bought from Scan for around £1850. I also took their insurance which offers to replace any component damaged during the build, couldn't face telling my boss if I broke something! Build went well, no major hassles apart from CPU cooler was tricky to install on LGA2011 socket. Very quiet machine compared to the Boxx and after running a few benchmarks I'm very impressed with the performance. Might have to stay back some nights to game on it I think...

 

3D Mark scores

 

Cloud Gate -28709

Fire Strike - 7501

Ice Storm - 186454

 

Cinebench Scores

 

OpenCL - 118.78 fps

CPU - 1181 cb

CPU Single Core - 158 cb

 

Compared to the HP Z800 Dual Xeon X5650 with 12GB RAM and Quadro 2000 + Tesla C2025

 

OpenCL - 46.09 fps

CPU - 1309 cb

CPU Single Core - 90 cb

[RyderSK]

Nice one, hope it serves :- )

 

Installing the massive towers on LG2011 is always "fun".

[numerobis]

i7 4930k OC'd to 4.6GHz with Asus 4-way optimization

...

Be Quiet! Dark Rock Pro 3

 

Really?!? How are your temps and voltages?

[RyderSK]

Same interest here. I am running very conservative 4,2 on 4930k but then when I see this numbers, I get a tick to try to go higher :- )

[Stephen Thomas]

Will check tomorrow. It's adaptive, so values change according to load I think. Will run some stress tests and check before deciding whether to stick with it or back down a bit. Seems like a good tool though, creates optimised individual fan profiles as well.

[numerobis]

If set to offset, then VCORE, VCCSA and VTT will be different under idle and load, yes. Interesting would be the values under load. I have two 4930Ks here running air cooled. One does 4,4GHz and only 4,2GHz under 1,35v vcore if i remember correctly... They both hit a wall there and temps go up rapidly to 75-80°C if i go higher (cooled by Noctua D14 and Thermalright True Spirit). I have two 3930K where it was really easy to get 4,6 and 4,7GHz stable. To clock the 4930Ks was not so funny... But i hadn't the time to try if some of the other settings could be optimized to get it stable at lower values, maybe i'll do this some time.

It would be interesting to see what voltages the auto overclocking sets. Normally it tends to go higher to be stable and in the past there have been problems with too high vccsa and vtt values set by the overclocking functions that could degrade and kill the CPU (3930K in this case).

Edited May 14, 2014 by numerobis

[Stephen Thomas]

Ok, so I just used Furmark to test the CPU and GPU under load.

 

GPU temps rose to 80°C at full load, with VDDC rising to 1.125V from an idle value of 0.862V

 

CPU temps leveled out at 69°C with a vcore of 1.404V under load and frequency of 4.5MHz (100.0 x 45)

 

I'm quite happy with those temps as long as the voltage is safe, what do you think? I currently have 5 fans installed in the case. I replaced the rear fractal fan with a silent wings and also added a side and bottom intake fan. Things were so quiet though that I put the fractal fan I removed back in as a top exhaust, so I have 3 intake 2 exhaust.

[Stephen Thomas]

This is the current settings from AISuite.

[Dimitris Tolios]

4.6GHz is actually an above average clock for 49xx CPU, and you did it on air. Grats (and silicone lottery).

[numerobis]

Ok, so I just used Furmark to test the CPU and GPU under load.

...

CPU temps leveled out at 69°C with a vcore of 1.404V under load and frequency of 4.5MHz (100.0 x 45)

 

As far as i know, AI Suite is reading the CPU package temp and not the core temp. So the core temp could be 10-15° higher. But i'm not sure (Coretemp or Realtemp is reading the core temp - but i'm not sure if they are working correctly if AI Suite is installed).

And i think Furmark is only for stressing GPU. For real CPU load you would need something like Intel Burn Test (IBT) with AVX (LinX AVX) or Prime95. IBT with AVX or Prime95 small FFT should produce the most heat. I like ITB because it is much faster in showing errors than Prime.

Or just render for a few hours and measure the temps for a real world test - depending on the renderer this can be more critical than ITB or Prime, even though the temps are normally lower than under ITB... Maxwell Render V3 is such a candidate - i had it stable for hours on ITB, but got a crash with Maxwell.

Cinebench is not very demanding i think.

 

A vcore of 1.4v is a bit on the high side i think... I would keep it below 1.35-1.375v, especially for air cooling - but if it is stable at 1.4v this is really one of the better chips .

VCSSA 1.2v is at the upper limit, but should be ok - if it is pinned at this voltage and not set dynamically per offset and raising under load...

VTTCPU is very good for 32GB i think - or maybe normal for 1600MHz, i don't know... i have only tried 1866MHz for now (not sure if this is adjusted dynamically too on the ASUS board).

[Stephen Thomas]

Not sure why the vcore goes higher under load than what the vcore setting in AISuite is, you can see it is set to 1.335v. Anyway, so if I try dropping the vcore value and still able to maintain stability running Prime95 this would be a good thing, correct? Realtemp does report higher temps, will try dropping vcore in stages and see how temps are affected and can I maintain this clock speed.

[Stephen Thomas]

After playing around with lowering the values and suffering a crash during rendering I decided to re-run the optimization tool. This time it settled on a slightly more conservative overclock of 4.5 on single core, 4.4 on 2-5 cores and 4.3 on all cores. I guess running it later in the day after stressing things earlier had an effect, but if that's a more realistic scenario then for peace of mind I'm going to leave it at that. VCore is set to 1.29v but rises to a max of 1.372v when fully loaded which seems like a safer level, and max core temps are around 75°C compared to 85°C with the earlier settings.

 

Still a good boost to stock speeds on air cooling, so overall I'm very happy.

[Stephen Thomas]

One thing I am pleased to report is that Geforce cards are just as good as their workstation counterparts in raw performance at least. The PC I inherited is running a Quadro 2000 and Tesla C2075 6Gb card in Nvidia Maximus configuration. I ran a VrayRT benchmark scene from Chaos group forum and the EVGA GTX770 4Gb card beat the workstation cards by 25%.

[numerobis]

After playing around with lowering the values and suffering a crash during rendering I decided to re-run the optimization tool. This time it settled on a slightly more conservative overclock of 4.5 on single core, 4.4 on 2-5 cores and 4.3 on all cores. I guess running it later in the day after stressing things earlier had an effect, but if that's a more realistic scenario then for peace of mind I'm going to leave it at that. VCore is set to 1.29v but rises to a max of 1.372v when fully loaded which seems like a safer level, and max core temps are around 75°C compared to 85°C with the earlier settings.

Yes, these temps sound more reasonable to me for a 4930K on air. You could maybe try to lower the VCCSA a bit, if it is still at 1.2V.

I think for quicker tests IBT is quite useful because it gives you a fast feedback. And then i would do a long run with your preferred renderer to see if it is stable.