Help me solve the Quadro versus GTX video card debate !

[autoMick]

Hi Folks,

It's no secret that workstation cards such as the Nvidia Quadro series are recommended for Autodesk products. It's also no secret that there many people who use 'gamer' style consumer cards such as the GeForce GTX series card due to their apparently higher performance per dollar. 

Searching the internet will reveal many debates, but I've yet to find a good real world test of the advantages of Quadro over GeForce generally, and particularly for Civil3D. There are vague statements about Quadro = better accuracy, with the GeForce taking shortcuts to achieve better speeds. What does this mean in the real world?

So... I went and bought a Quadro K4000 to test against my GTX680. I installed it, and it works just fine, but no noticable difference to the GTX680 performance on first inspection. So now I want to do a proper set of real world test that I hoping you might be able to help with...

- firstly I want to optimise the Quadro performance. I just installed the default driver from the Quadro installation disk. I noticed there was a AutoCAD optimsed driver (or something to that effect) tucked away in one of the directories on the disk. But not sure if that is also what is recommended for Civil3D. What driver is recommended?

- second I want to put together a definitive set of real-world tests (or find some free software) that will allow me to gauge the best and worst of the two different cards. Any suggestions?

Any suggestions appreciated

Regards

- Mick

[Alfred.NESWADBA]

Hi,

>> I noticed there was a AutoCAD optimsed driver

No, not for current versions of AutoCAD (and so for Civil3D), this optimized driver was valid for AutoCAD versions were you could define (change) the driver that AutoCAD accesses for the display (up to AutoCAD 2009 I think). Since AutoCAD 2010 there is no chance to configure an external driver (that AutoCAD uses then), so there is no chance to get the (old) nVidia Quadro performance driver working.

And exactly that makes now the difference between GTX and Quadro's, Quadro's can have more performance, but as AutoCAD does not communicate via the optimized interface (but most close to Direct3D) you don't get all the performance that is in the card.

If you use older AutoCAD's or e.g. something like 3DS then you get more from the available performance in the Quadro.

>> So... I went and bought a Quadro K4000 to test against my GTX680

Wow, buy just for test ... great! 😉

Cadalyst has a benchmark for AutoCAD that also does some tests around the graphics performance, download this test from >>>here<<< ... and let us know the results!

My 2c, - alfred -

[autoMick]

Hi Alfred, thanks for the reply.

Hah... it sounds like I had a lazy $1K to chuck away, so why not on a Quadro ? It would be great to have cash to burn, but that's not quite the full story, I got if for a number of reasons, including a fair bit of photoshop work. But I also wanted to test the card in a definitive manner because I'm looking at the longer term upgrade path for some of our computers as well.

That's interesting information about the (old) driver... I wasn't aware of that, I had assumed (in fact a lot the internet chatter suggests) that each quadro driver is optimised for workstation style software such as autocad. In fact, a lot of the 'value' that people seem to justify the cost of a quadro is based on the better 'quality' of the drivers - which of course is one of the things I was hoping to test objectively.

I did actually come across the software you linked previously, but for some mistaken reason thought that was only for 2012. I'm using 2014 on the test computer, but will try it anyway.

Yes, I'll definitely post the results. If you guys can suggest the tests, I'll try my best to report back with some usable data.

Cheers

- Mick

[dgorsman]

There aren't any objective tests that can be performed, as the results are what is seen on screen and therrefore highly subjective e.g. "Taking too long to rotate" or "Regenerates quickly".  Accuracy differences are irrelevant and probably a misnomer, as the numbers are crunched based on model data not what is on screen, unlike visualization software like 3DS MAX where "looks good" is more important than accuracy.  For example, snapping to an endpoint of an object doesn't care how well its rendered to screen - the number is the same even with ye olde 8-bit gamer graphics.  Add in differences in hardware configuration like RAM speed/optimization, BIOS versions, processor style/speed, and so on means any benchmark values will only be applicable for exactly the same configuration.  For meaningful results the same dataset would have to be applied for each test as well.

[sboon]

There is some information in this thread which may be helpful. 

Steve
Please use the Accept as Solution or Kudo buttons when appropriate

[autoMick]

dgorsman, I totally agree regarding absolute benchmarks - meaningless unless you do exactly the same work on exactly the same computers. As I haven't got identical computers side by side, I can't really compare some of those subjective factors in any reliable way, so have to rely on things I can record. The intention for my tests is no so much to document a benchmark (say like passmark http://www.videocardbenchmark.net/high_end_gpus.html), but more the relative performance of the two cards on my system.

Interesting that you bring up the example of snapping to end points. My original thoughts were that 'accuracy' was that 'accuracy' is irrelevant in this context as well (which it is) - thinking exactly about this example. I now realise that most comments regarding 'accuracy' for GPU's is more related to rendering, in particular with reflective surfaces. Although... now I think about it, I remember having lots of accuracy issues in Sketchup with selecting vertices, where the location of a vertex was different to where it was displayed. I just revisited that problem with the Quadro card installed and the issue is exactly the same - so it's a sketchup issue, not GPU.

Steve, thanks for the link - a few little tidbits in there!

Alfred - I downloaded that test and tried to run it, but gettting errors someway through, so not getting to the results stage - might be something to do with 2014. So nothing to report yet, but will get something out of it.

Thanks for the suggestions so far.

Cheers

- Mick

[CADmgrMike]

Anxiously awaiting any further coment.

this is a big discussion in our company right now trying to come to some conclusion on if we should buy Quadro cards or gaming cards.

[Cadguru42]

My $0.02 worth is this: If you're only using C3D (or AutoCAD) then the GeForce gaming cards are fine. If you're using 3ds Max, Photoshop, Premiere, etc., then the Quadro cards are worth it. For those of us in the civil world, in AutoCAD we're only looking at a top, wireframe view the vast majority of the time. The difference between the gaming cards and the workstation cards doesn't show up when using that view. It's when you use the object viewer or viewport to rotate what you're looking at is when the real GPU power is being used and even then there's not much difference between them. Since 2010 I think AutoCAD is now using Direct3D for everything, which makes the workstation cards almost useless as it was the drivers for them that made the difference before. 

[CADmgrMike]

We are also starting to have reason to use point clouds, Recap, as well as Infraworks for conceptual design.

Do you think that would necessitate looking at the Quadro cards?

Doing any corridor design, I use object viewer quite frequently.

Without seeing any testing results specifically tailored to the above software, or doing the testing yourself, it's very difficult to explain and sell the idea of buying an $800-1500 card over a $400 gaming card.

[rl_jackson]

Personally I have dual gtx250 with 1gb each and have have absolutely no issues. I'm generally in plan and wire frame. And alothiugh there is some lag in orbit its something I've always been able to live with. I have not seen any major advantage over the higher cost short of the sticker shock.

[autoMick]

Well... as per my original post, I'm running two virtually identical computers, one with a K4000 and the other with a GTX680. One's at home, one's at work and I often work on the same DWG files between the two locations. I work with LiDAR, large surface analyses and large aerial photos (ECW), gradings and volumes in Civil3D. I don't do much with alignments, pipe networks, etc. so my experience is just with a small subset of the C3D capability.

In my experience, both graphics cards work well with Civil3D and there is no reason not to use either of them.

I never did do any definitive tests, but my feeling is that the K4000 is better to work with, but there is no particular difference (speed, clarity, smoothness, accuracy) that I can put my finger on. As the computers are in 2 different locations, I'm not comparing them side by side.

Is a quadro card worth the extra money? Probably not for Civil3D.

Regards

- Mick

[sboon]

For us the debate was less about speed than stability.  Have you noticed that one machine is more or less likely to crash during 3d operations, rendering etc?

Steve
Please use the Accept as Solution or Kudo buttons when appropriate

[autoMick]

Steve,

I'd like to say that stability was heaps better with the Quadro - that would make me feel better about having bought it in the first place, but there's no different that I could attribute to the graphics card for Civil3D. I run fairly 'clean' systems and really don't have many stability issues generally but I have to say that 2015 SP1 seems to be more stable that 2014 for my kind of work. So software version is probably a bigger issue than graphics card in terms of stability.

I also use revit, photoshop, blender - primarily on the home (Quadro) computer, which is I why that card is installed at home. I have to admit I haven't compared revit or blender renders between home and work - I suspect the quality of the quadro renders would be significantly superior. I'll have to do a test.

I tend to do more experimental type work on the quadro computer - (i.e. I wonder if this will work....?), and generally there's no problem. On that basis I regard that computer as a very reliable system (even though it doesn't have ECC memory, etc.), but that doesn't mean it's more stable that the GTX card computer. As I said, I'm happy with both.

Probably only one more thing I can think of... the GTX card drivers seem to get a lot more updates (e.g. specific optimisation for a new game release), whereas the Quadro drivers evolve less. Sorting out whether you want each particular GTX driver update or not could be a consideration for IT. I just accept each update as they come and have no problems.

Sorry to not be more definitive, but that's all I've got. I'm happy to run a test file or two if you have something in mind.

Cheers

- Mick

[dgorsman]

I'm not certain the Revit or Blender renders would show a difference.  AFAIK Revit uses the CPU (although its multi-threaded), not sure about whether Blender has switches to use GPU rendering.  I've *just* got my hands around using the program to edit together Navisworks outptut into movies and playing around a little with the noodles.

[CADmgrMike]

Well, you're no help for all those wanting to sell us expensive Quadro cards! LOL

Thanks for the update.

[Cadguru42]

---

@dgorsman wrote:

I'm not certain the Revit or Blender renders would show a difference.  AFAIK Revit uses the CPU (although its multi-threaded), not sure about whether Blender has switches to use GPU rendering.  I've *just* got my hands around using the program to edit together Navisworks outptut into movies and playing around a little with the noodles.

---

The GPU doesn't change the output or quality of the render itself. There are some renderers that can use the GPU to handle it, but the majority of renderers use the CPU, such as the scanline renderer, VRAY, and MentalRay in 3ds Max. Renderers such as Octane, IRay, etc. are GPU based, which speeds up the renders, but the quality is based on the renderer itself, not the GPU. In fact, if you're using a GPU based renderer it's recommended to get a gaming card as they are much, much faster than the Quadro cards and are cheaper to boot.

EDIT: Here's a YouTube video showing the difference between a CPU render versus GPU render using VRAY. The quality is the same, but the time to get it is much, much faster.

http://youtu.be/eHh9x_3EZ98

[autoMick]

dgorsman, I have to admit I didn't know about revit, but Dr Google confirms your theory. With Blender you can use 'cycles' which can be set to GPU compute, which is what I use, but I haven't done any definitive comparisons.

engertech, you may be right for some software, but I equally seem to remember youtube videos of side by side renders clearly showing differences, particular with reflective surfaces. When there are claims of greater 'accuracy' for workstation cards, I think the reference is more render/visual accuracy rather than xyz linework etc.. I'm genuinely interested in the recommendation regarding getting a gaming card for rendering as that's completely counter to my understanding. It probably depends on what the target audience is for the render too...

Back to Civil3D -  I still have no idea if any of the claimed benefits of a quadro card are actually verifiable.

Cheers

- Mick

[Cadguru42]

---

@autoMick wrote:

engertech, you may be right for some software, but I equally seem to remember youtube videos of side by side renders clearly showing differences, particular with reflective surfaces. When there are claims of greater 'accuracy' for workstation cards, I think the reference is more render/visual accuracy rather than xyz linework etc.. I'm genuinely interested in the recommendation regarding getting a gaming card for rendering as that's completely counter to my understanding. It probably depends on what the target audience is for the render too...

Back to Civil3D -  I still have no idea if any of the claimed benefits of a quadro card are actually verifiable.

Cheers

- Mick

---

There won't be a difference between the "accuracy" in a render. A render is a render and is based only on the renderer itself. If a single renderer can use both a GPU and a CPU, then the output should look identical between them, but there are very, very few renderers where you can pick between GPU or CPU. VRay has a separate renderer for GPU rendering, therefore it'll look slightly different than the regular CPU VRay renderings, but it's very close. The faster video cards only help with viewport visualizations and the "accuracy" of what you're seeing, but renderings are always going to look the way the renderer itself is designed regardless of GPU or CPU being used.

Some software, such as Photoshop and Premiere, will take advantage of a Quadro card to speed up effects, but the output will be the same whether you have a gaming card or a workstation card. Really, with the advent of Direct3D and the 3d software using it, it's rendered (pun intended!) the workstation class cards almost useless. They are slower than gaming cards, don't get updates as often, but are considered more reliable under heavy usage, which for a company like Pixar or render farms would be better, but for us average Joes they just aren't worth the money. 

[Homsey-CCH]

AutoMick, my issue with Revit 2015 isn't with final rendering, but with modelling in shaded and realistic modes.  Our current project site is about 1800 feet by 1800 feet of topography with three 3-story buildings (Revit files linked in to the site model) totalling about 150,000 SF, plus the access roads and parking lots to go with them, which are subregions on the topo.  We have about 300 trees and 300 bushes of various species, plus cars and people.  I have a GTX 560 card and the model is very cumbersome to pan, zoom, and rotate in "shaded" mode.  In "realistic" mode, it takes 3 or 4 minutes to update a view.  A 2400x1500 full render with just sunlight takes about 30 minutes, but that is using the CPU, not the video card.

The same GTX video card runs the model smoothly in Lumion, which gets the most bang for the buck from gaming-type cards.  Before I buy a Quadro card, I wish I could tell if it would make a major difference in Revit during the modelling process.  I don't mean a difference you'd need to benchmark to see, but rather a difference of a factor of 10 or more.

The reason I think it might, is that we have an old station with a Quadro FX 580.  That only has 32 cuda cores and 512 MB, and yet it rotates the Revit shaded view almost as well as the GTX 560 1GB cards with 336 cuda cores, on the other stations.  My hope is that a higher end Quadro card with a lot of memory might be able to handle a big model and allow a smooth workflow.  Your K4000 has 768 cuda cores and I think 3GB of memory.  Your GTX 680 has 1536 cuda cores and probably 2GB of memory.

Suppose you make a large flat site, populate it with some textured buildings and a bunch of trees.  See how Revit handles it with both cards, and if the difference is night and day, or hardly noticible when zooming, rotating, and panning the site in 3D shaded and realistic mode, with and without shadows.  If the drivers were equal, the GTX should be twice as fast, but I suspect it will be much slower than the Quadro, and I'd love to know for sure.

(if you have time to try it: thanks!)