Quadro vs. Geforce

[B. Ludin]

It's now time for the acquisition of new notebooks for our company. For a long time we have been wondering about whether it would be worth investing in Quadro graphics cards, or whether the GeForce cards would work just as well or possibly even better.

We got the hint that Quadros would be better at OpenGL while GeForces would shine with DirectX but apart from that, there seemed to be very little solid info on that subject.

Luckily, Schneider in Germany lent us two notebooks that were absolutely identical but for the graphics cards, so we were able to directly compare a Quadro K1000 to a Geforce GTX770.

We are always working with smooth shading and drawing parts edges and we usually have a lot of transparent parts in the scene. So that was the base for our tests.

 

The interesting outcome was that we couldn't see a significant difference between the two cards. The quality of the images rendered by the two cards were identical, pixel by pixel even, as far as we could say, whether we compared under OpenGL, OpenGL2 and DirectX. We also used tools to check the load on the GPU and on the CPU.

 

Our findigs:

Transparency rendering was best (but for an annoying bug) AND fastest under OpenGL.

Part edges are rendered by the CPU and not by the GPU. An this is what limits the speed. As long as we rendered part edges while rotating the scene, one CPU core would go to 100% while the GPU load rarely exceeded 25%. As soon as we switched part edges off, the frame rate would go up several times, CPU load would go down and GPU load go up to 60-90%. According to a (older) white paper by Nvidia, one of the differences between Quadro and Geforce cards should be, that the former are able to render edges. It appears that IronCADs rendering engine doesn't make use of that capability of the Quadro cards. Or at least, we haven't found a setting the would put this burden on the GPUs shoulders. IronFolks, can you comment on that?

Our conclusion: at the moment, it looks like the frame rate when rendering part edges is limited by the CPU not by the GPU, so it seems more beneficial to invest in a faster CPU than in a great graphics card.

 

Any comments and contributions are welcome, but please don't ask me for a protocol of all the settings we've tried. We were three people investing maybe 20 hours in total, but still, there so many possible combinations,that we could only test a small fraction.

 

Cheers, Beat

Edited January 18, 2014 by B. Ludin

[dleczynski]

I suggest you visit website: http://www.solidworks.com/.

bookmark (support -> System Requirements -> Graphics Card Drivers)

 

if you want to save money the best model for 3D is M6500 and Quadro FX2800 or fire pro, the big picture and high resolution 17inch Or m4700 quadro k2000m but low resolution and pictur 15,6 inch.

 

[B. Ludin]

I suggest you visit website: http://www.solidworks.com/.

bookmark (support -> System Requirements -> Graphics Card Drivers)

 

if you want to save money the best model for 3D is M6500 and Quadro FX2800 or fire pro, the big picture and high resolution 17inch Or m4700 quadro k2000m but low resolution and pictur 15,6 inch.

37316[/snapback]

 

Thanks for the link.

But as I said, we have the strong impression that the graphics card is not the speed limiting component, because IC's rendering engine doesn't seem to male use of the capabilities of the GPU.

Does SW use the same rendering engine as IC?

 

 

 

[dleczynski]

hello again,

 

recently purchased new equipment i7 dual core 2.67 GHz + Nvidia fx2800m 8gbram.

I am very surprised because there is no difference between nvidia fx1600.

The laptop allows for the construction of 5 thousand parts because later disappears, and the example on a desktop computer with a much weaker graphics can reach 40 thousand parts per Gefeorce GT 9800. I think the quadro cards are not good ... GTX will be better.

 

you could test a sample of of cubes?

[dleczynski]

CPU E8400 core duo 3.0GZH 3.0GHZ

RAM 4GB

Graphic Card Geforce 9800 GT

 

40 000 parts

[Bill C]

Hello Beat,

 

When you say Part Edges are ON or OFF I assume you mean you change the settings in the IronCAD Rendering page. The way we implement Part Edges is going to make a big difference. The best thing for you to do would be to export as HSF two models for comparison.

 

thanks.

-bill

 

 

 

 

It's now time for the acquisition of new notebooks for our company. For a long time we have been wondering about whether it would be worth investing in Quadro graphics cards, or whether the GeForce cards would work just as well or possibly even better.

We got the hint that Quadros would be better at OpenGL while GeForces would shine with DirectX but apart from that, there seemed to be very little solid info on that subject.

Luckily, Schneider in Germany lent us two notebooks that were absolutely identical but for the graphics cards, so we were able to directly compare a Quadro K1000 to a Geforce GTX770.

We are always working with smooth shading and drawing parts edges and we usually have a lot of transparent parts in the scene. So that was the base for our tests.

 

The interesting outcome was that we couldn't see a significant difference between the two cards. The quality of the images rendered by the two cards were identical, pixel by pixel even, as far as we could say, whether we compared under OpenGL, OpenGL2 and DirectX. We also used tools to check the load on the GPU and on the CPU.

 

Our findigs:

Transparency rendering was best (but for an annoying bug) AND fastest under OpenGL.

Part edges are rendered by the CPU and not by the GPU. An this is what limits the speed. As long as we rendered part edges while rotating the scene, one CPU core would go to 100% while the GPU load rarely exceeded 25%. As soon as we switched part edges off, the frame rate would go up several times, CPU load would go down and GPU load go up to 60-90%. According to a (older) white paper by Nvidia, one of the differences between Quadro and Geforce cards should be, that the former are able to render edges. It appears that IronCADs rendering engine doesn't make use of that capability of the Quadro cards. Or at least, we haven't found a setting the would put this burden on the GPUs shoulders. IronFolks, can you comment on that? 

Our conclusion: at the moment, it looks like the frame rate when rendering part edges is limited by the CPU not by the GPU, so it seems more beneficial to invest in a faster CPU than in a great graphics card.

 

Any comments and contributions are welcome, but please don't ask me for a protocol of all the settings we've tried. We were three people investing maybe 20 hours in total, but still, there so many possible combinations,that we could only test a small fraction.

 

Cheers, Beat

37315[/snapback]

 

[dleczynski]

Part Edges are ON works much better than Part Edges are OFF.

I tested it on multiple PC. I do not think that HSF here to help.

The GTX 770 is first class in performance http://www.notebookcheck.pl/NVIDIA-GeForce...I.104185.0.html

[Bill C]

Hi Beat,

I'd like to clarify my comments below.

Basically, we'd like to try to reproduce the CPU/GPU behavior you're reporting using only your graphical data. I'm thinking we can do that with just the HSF data. This eliminates the possibility that any of the behavior is related to other aspects of the IRONCAD application.

 

To more directly address your question, I do not believe there is any logic in our system that says "Part Edges are rendered by the CPU". Part edges are just another graphical object like facets that are send to the rendering engine.

That being said, we'd like to better understand the behavior that you are reporting.

 

thanks.

-bill

 

 

Hello Beat,

 

When you say Part Edges are ON or OFF I assume you mean you change the settings in the IronCAD Rendering page.  The way we implement Part Edges is going to make a big difference. The best thing for you to do would be to export as HSF two models for comparison.

 

thanks.

-bill

37463[/snapback]