Experiencing Extremely Slow Viewport Performance: Quadro K4000 & K5000

Rhino Windows Hardware
1

Dear Mcneel Forum

I just put together 2 new Lenovo workstations for Rhino use in our studio. We opted for expensive Nvidia Quaddro drivers (A k5000 and a k4000), as it was highly recommended on the mcneel website. The results have been barely acceptable thus far. On both machines the viewport performances were almost exactly the same as our old computers running AMD Radeon HD7800 Cards - we were expecting a significant improvement.

Been searching the internet/ forums for a solution, and so far have tried the following:
-Clean install/ update to the latest driver (Multiple times)
-Set to Dynamic Streaming (also spent hours trying most different configurations on the nvidia control panel.)
-Power management mode to \Prefer maximum performance

Both machines are running windows 7, aero theme, and the specifications are as follows.

Machine 1
Motherboard: Lenovo Thinkstation D30
CPU: Dual Intel Xeon E5-2630v2 2.60GHz
GPU: Quadro K4000
RAM: 64GB Kingston D30 Specific RAM

Machine 2
Motherboard: Lenovo Thinkstation D30
CPU: Dual Intel Xeon E5-2620v2 2.60GHz
GPU: Quadro K5000
RAM: 64GB Kingston D30 Specific RAM

Currently we are extremely dissatisfied with the performance of our new machines, don’t really know what to do other than to just… Learn to be patient…

Is this normal? Are quadros actually just hyped up? Is there anything I’ve missed? Any advice would be hugely appreciated!

Best,
H

2

Could you post Holomark 2 results?

3

Cheers for the reply!
I should also mention that the slowness is on architectural models about 500mb in size.

4

A lot of CAD is single-threaded. Your processors, for all their multi-core might, have so-so single thread capability 1.41 VS 1.97 for $300 Quad Core i7.

Yes, I do think that Quadro Cards are over-hyped and priced, but Nvidia cripples gaming cards for CAD-like things, yet the dies are often the same. Quadro cards may or may not have ECC RAM though.

For display speed, you might want to check your anti-aliasing and aniosotropic filtering values, finding some compromise. For instance, unless you model is textured, you can shut off aniosotropic filtering for Rhino. If you have a very high-res display, you might be able to lower your anti-aliasing settings. If you are not careful, if might even be possible to stack Rhino’s AA settings, over NVidia, when using NVIDIA FXAA. If you are running high-res, vertical-sync values might affect performance. Shutting if off, or lowering it to the new partial mode might help. Make sure you have single display selected if you only have one. Make sure you have threaded optimization on. I would leave shader cache off if you have a SSD, as it just adds wear.

NVInspector may give you greater control over your video cards settings.

You might set you CPU power to maximum performance, as you did with your graphics card.

Try to hold our for Broadwell, that would give you another 10% over Haswell, if they ever make quads.

It might make things more enjoyable by adjusting the Adaptive Degradation, under /File/Properties/View/Dynamic Redraw…
Where FPS, is the minimum frames per second, you want to maintain.

5

BTW, my Lenovo W540 is bettering our your system for some reason

On adaptive:

Holomark 2 v2,2,03

Total Score: 43692
Total Runtime: 205.41 sec

GPU scores: 28270
…
CPU scores: 15422
…

LENOVO
20BGCTO1WW

NVIDIA Quadro K2100M - 2048.0 MB
DriverVersion: 9.18.13.4105
Intel® HD Graphics Family - 2112.0 MB
DriverVersion: 10.18.10.3412

Intel® Core™ i7-4800MQ CPU @ 2.70GHz
NumberOfCores: 4 NumberOfLogicalProcessors: 8
MaxClockSpeed: 2.7 GHz

TotalPhysicalMemory: 24.0 GB

Microsoft Windows 7 Professional

  • Service Pack 1 - 64-bit

On Maximum:

Total Score: 46464
Total Runtime: 163.57 sec

GPU scores: 30800
…
CPU scores: 15664
…

As an experiment, on adaptive GPU, I raised the antialiasing up to 8X, with a GPU scores: 25530, down from 2x giving 28270.

6

Hi Hans,

your results are really quite bad. My old GTX285 was better (for example if I look at the most interesting GPU4 value). Since Quadro cards are no limited by Nvida we should get the best available power within the limited power of Rhino. The best GPU4 values are around 65 fps. My old GTX285 show me results around 35 fps.

I don’t believe a lower AA will bring better results. This would help if your card would be on the calculation power limit. But your card is sure not on it’s limited.

General we have a basic Rhino display problem. If your display is slow caused by the complex model, than better hardware doesn’t help so much. The software is the problem.

But my impression is we have two other problems too here. First your CPU is slow, 2.1GHz is quite low. Best for Rhino you use less cores, but at max speed like for example a quad core at 4GHz. An other problem could be that you are using a Dual Xeon machine. I use a dual xeon too and my GPU results are much lower than on single CPU machines. It’s a new observation and I will post my result at the Holomark thread.

7

Your description of the CPU clock speed says 2.6 GHz, but Holomark shows it at 2.1. Is your description right? If so, does your motherboard have speed selection in the BIOS and do you have it set to the highest speed your chips will support?

Considering Micha’s observation about one chip running faster than two with Rhino, is it possible for you to turn off or pull out one of the CPU’s to see what happens?

And a really dumb question: do you have hardware acceleration turned on?

8

[Micha, nudge, I um, did prove that AA makes some difference in GPU 25530 VS 28270.]

9

Brenda, maybe AA shows a difference where speed is available yet, if Rhino isn’t the bottle neck. It should be the case if no NURBS/curves are at the display, in tests with meshes only. But if your display is slow, because the complex model, than the GPU is used with a few percent usage only most. In this case AA doesn’t matter. For example the bike model test, AA and hardware have no big influence on the test.

10

Sorry there was a small error on my original post. The higher spec machine, with the k5000, has the 2.6GHZ processor. (And the K4000 machine has a 2.1GHZ Xeon).

And yes I do have “Use accelerated hardware modes” checked!

11

Interesting and disturbing observation… I wish we had this info a few weeks earlier when we were ordering our machines…

I had a look at the Holomark thread, seems like we could build 2 or 3 machines with double the performance (on rhino), for the same price we spent on these.

So far the only redeeming point with these machines were fast rendering times with V-Ray.

12

some test information here:

http://www.simplyrhino.co.uk/rhinotraining/pny_quadro_test.html

we’ll update it, but it’s still relevant for anyone looking at higher end graphics cards and value for Rhino users

13

For sure concerning the Nvidia cards, but it’s only telling half the truth. I know, Nvidia cards were pushed at McNeel’s front, mostly for reason of supposedly better AA results and ancient times were the then ATI cards had some driver issues, that’s years ago, but my two AMD(ATI) HD 7900er GPU cards, the 7950 for 200$ with 3 GB DDR5, and the 7970 with 6 GB for 350$ are concerning all scores at pole position. And that means within all cards purchasable. That’s how it is with Rhino. It was always proud not to need expensive Pro cards. And I think, it is a shame, how these rich firms cripple their consumer cards by means of driver inaccessability. If the consumer cards would have a similar performance at Pro OpenGL, there would still be enough engineering companies, automotive, you name it, be left to give a fuck and still so buy the expensive Pro cards, because they could. If I would have a rich company, I know I would. Policy being, why shouldn’t newcomers with not enough starter money in this biz get a chance to full power, and I solidarize with them this way.
And yes, there were times, where same stands were friends.

14

After all that was said so far and I agree with, if you want to work with Rhino exclusivly on CAD, and you can still give back those cards, do it. And buy a solidly build kickass gaming machine, better yet, build one, or let build, with not too many cores, but a fast running cpu. There are elder i7 sandy bridge cpu’s, that run with over 5GHz without water cooling. Peng. That’s double the speed of those Xeons. Mine runs at 4.4 GHz, without having to turn any screw separately.
Put in another 16 GB RAM, and for the free money, you can put in several really fast M.2 SSD modules, so Rhino pulls in an 800 Mb file within seconds. Though this would need a modern board, for the latest PCIe 3.0 speed and DDR4 RAM.
Of course the more cores the better, I have six at the time, one never knows what Rhino 6 will bring, certainly some more parallelisations in number crunching, but several cpu’s is overkill. Not, if you also want to render with it. Other thing.

steff

15

Hi Steff- for what it is worth, my Fire-Pro 8000 (or something) has utterly useless anti-aliasing of curves/edges/isocurves. Otherwise it is great and plenty fast.

-Pascal

16

Hi RhinoUK,

I like the clear overview. Only the focused test could be choosen better. Since Rhino is a NURBS modeller the GPU2 test is not so interesting most, except the user extract all render meshes and meshes and hide NURBS/curves. In this case speed isn’t a problem most, because the GPU is heavy used. I think the shaded GPU4/8 test tell us more about the expected daily usage performance. From my experience the old bike model test is the simpliest test to see, will a card help to speed up a slow display. The bike model test is the test for the Rhino slow display problem of heavy models.

17

Micha,
I’ve looked for that bike a few times now but I must have parked it in the wrong place… Could you post that model here once more?

18

www.simulacrum.de/temp/Testmaxspeed.rar

1 Like
19

Hi Pascal, shit. Not good to hear that, though great to hear from you;-)
Of course I thought there must have been profound reason for McNeel’s before mentioned recommendations.
I am really not handling any models right now, so haven’t really worked with my cards so far, but from first
impressions, I could work with these even without AA turned on.
Also, I guess, you certainly tried out newer drivers from time to time…
Whereas, the latest ones aren’t the best performing neccessarily.