test scene

[jstrob]

Now I have 9.3.3.3000

I will install the last one: 13.352.1006

[jstrob]

Same crash with latest driver. And btw 2 textures are missing:
Furnishing.Fabrics.Lace.jpg
Furnishing.Fabrics.Lace.Mask.jpg

[Nicinus]

Is there an updated benchmark chart somewhere?

I'm looking to upgrade my graphics on my work station and trying to decide between 2 x 780 6GB or a Titan Black. It is for the moment about the same money, but I will get more cores with the dual 780s (4,600 vs 2,880).

Any opinions on pros and cons would be appreciated. I drive four monitors, 2 30" and 2 1920x1080.

[Valtiel]

By far I'd say go with the dual 780's. Titan's only a real advantage if you need dual precision performance, which RT does not leverage.

[Nicinus]

Thanks! Besides the cores, would I notice any difference in screen performance by having two cards compared to the one Titan?

Btw, what software uses dual precision? Max? Revit?

[Valtiel]

Honestly, as far as I know and can find on the first page of google results, nothing in CG uses double precision. It's mainly for scientific simulation and research like GROMACS and folding@home. Titan had a core advantage over the earlier 700 series but that's gone with the 780Ti. If I understand correctly, it doesn't even have the professional application drivers that optimize performance in viewports like the Quadro. Besides being a cheaper alternative to fully-fledged Tesla, I have trouble understanding its niche in Nvidia's lineup.

Oh yeah, if you can, go for the 780 Ti rather than the 780: 2880 cores/unit vs 2304 for ~50-100$ premium and still much cheaper than Titan Black (which I did just notice has a small clockspeed advantage but I wouldn't consider that worth 300$ more).

[Nicinus]

The reason I'm considering the 780 is really for the 6GB, which isn't and apparently won't be, available for the Ti version. There are rumors about a 790, which I assume means dual 780Ti, but that probably won't be 6GB (per card) for the foreseeable future.

The EVGA 780 6GB is also over clocked, somewhat making it more comparable with the 780Ti, at least the reference version. The Ti has higher memory speed (7 vs 6 GHz) but those kind of differences are in my experience rarely worth it in practice. The 6GB makes a difference though so the only real comparison is the Titan Black. Having said that I can't say I'm using RT that much, at least not GPU, so I may be chasing ghosts to begin with.

Any opinion on which solution would have the best screen performance? I have no experience of SLI at all, but since I have two 30" and two for HD I'm hoping it can make a difference.

[Valtiel]

You're probably right on the 790. I wouldn't expect it to have more than 3/processor but then again the 295X slapped 4GB apiece so it is possible.

I haven't seen any apples-apples 780 3GB vs 6GB benchmarks in the forums but the closest I’ve noticed is a Titan (on RT 2.4) and 780 Ti (on RT 3.0) getting essentially the same time. Vlado has said that 2.4’s code is simpler so you could expect the 780Ti to perform a bit better when tested on the same kernel. In the case of the test scene at least, it seems cores+clocks trumps memory size though you would want to match their clockspeeds to isolate the advantage.

I know workstation apps are more likely to need larger RAM stores than consumer applications but you should still verify that a) Your scenes will actually need more than 3GB and b) RT will actually leverage it. I’ve posted a small lament in the other benchmark thread about how RAM size is one of the most oversold features on GPU’s; it makes a difference until you have ‘enough’ for your application and any more beyond that does nothing. It’s probably best to ask the devs how necessary additional VRAM is for RT seeing as how the test scene doesn’t actually seem to use that much. You could also take one of your own scenes that you know consumes a large amount of system RAM at render time with VRay Adv and try to run RT on a fixed card to see what diff it makes. I’d test for you but I’m on (still non-functional) AMD cards.

[Nicinus]

It's tricky. I usually keep my hardware for a good three years so every time I upgrade I try to make a decent decision. The 780 Ti seems to be a great card and with two I would have almost 6000 cores, but only 3GB compared to the Titan Black.

Anyone dare to guess which would have faster screen performance (given I'm driving 4 monitors), two 780 Ti SLI or a Titan Black? Will SLI make difference for Max and something like Revit? Or would it basically be an unnoticeable difference?

[Valtiel]

It's hard to say but I'm guessing only one or two of those monitors will be driving viewport (3D) content so any that can actually do 4 monitors, and you may have to dp daisychain to make this so, should be fine. As far as I know, SLI and XFire are structured for gaming and creative pros are best served keeping those plastic straps off of our cards; for what we want they cooperate just fine without bridges. Probably best to get multiple input on this one though (Dmitris? )

[fraggle]

I do have 2 780 OC and one 670 OC too.
From my own experience, my old Quadro FX5800 was performing a lot better, really a lot, in viewport, compare with theses ones.
Moving around the scene is very slow and I sometime I need to give up on wireframe mode.

But I can used the RT GPU, so you can't get the butter and the money of the butter.

[Nicinus]

Fraggle, that is frustrating to hear. I've never used a Quadro, but I have experience with SGI (yes, I'm getting old) and the smoothness and antialiasing of those systems is something I miss dearly.

You may not have tested them separately, but if you did, was there any improvement in terms of screen performance using the two 780's compared to one? Are you using SLI?

[fraggle]

I'm not using SLI and I don't think that SLI will make any difference in a 3D application like 3Dsmax.
It was not so easy to get all of them worked fine, so I will avoid to unplug them.

Fraggle, that is frustrating to hear. I've never used a Quadro, but I have experience with SGI (yes, I'm getting old) and the smoothness and antialiasing of those systems is something I miss dearly.

You may not have tested them separately, but if you did, was there any improvement in terms of screen performance using the two 780's compared to one? Are you using SLI?

[tricky]

Just ran the test scene on my new workstation. I wasn't expecting much as the card is a Quadro4000. However, at some point I got some sort of error crash and the vray cmd box reads like this:-

Error Invalid command queue (-36) at line 1249 in file ./src/ocl_tracedevice.cpp !!!
Error Invalid command queue (-36) at line 1028 in file ./src/ocl_tracedevice.cpp !!!

Writing crah dump to ".....VRay.dmp"
Unhandled exception occured, will exit render loop. Unhandled exception c0000005

Error Memory object allocation failure (-4) at line 1224, in file ./src/ocl_tracedevice.cpp !!!
Unhandled exception occured, will exit render loop. Unhandled exception c0000005

What's this all about then?

EDIT:-

I just tried again and similar thing happened - a warning came up to say the display device had stopped responding. I then got a crash with the following info in the event viewer:-

Faulting application name: vray.exe, version: 1.0.0.1, time stamp: 0x53024831
Faulting module name: KERNELBASE.dll, version: 6.1.7601.18409, time stamp: 0x5315a05a
Exception code: 0x80000003
Fault offset: 0x0000000000033ca2
Faulting process id: 0x247c
Faulting application start time: 0x01cfc83f1e2c3e69
Faulting application path: C:\Program Files\Chaos Group\V-Ray\RT for 3ds Max 2014 for x64\bin\vray.exe
Faulting module path: C:\Windows\system32\KERNELBASE.dll
Report Id: 9df7218e-3432-11e4-b0a8-f8b156c2a61d

[svetlozar.draganov]

What V-Ray version is that?
Is the issue scene dependent or it happens on every scene?
What version of the gpu-drives are installed on the machine?
Have you tried to render the scene with another V-Ray version?
Which V-Ray RT engine you are using Cuda or OpenCL? Does the issue persist with both engines?