Hi, do not forget that usually you are at geometry level, when you move the camera, go up to scene level and compare that to mayas performance. Would be nice to have the same performance at geo level.

Yeah, I was at scene level.

This is interesting:

chrism
We're getting around 150-170 fps when tumbling around with that torus/grid scene on our k4000. Not sure what the difference is between our machines, but at this point you should log a bug and provide support with your specs from Help > About Houdini, along with your scene file. Thanks!

How did they get 40 times better performance on k4000 comparing to mine?

Apparently I've got the same issue as the author of this topic, was hoping that he found the solution by now.

also i just tested eg. when I create a 1000x1000 (rows, columns) grid in houdini my fps drops down to 25 (GL 3.3), in maya OGL viewport I can create a 6000x6000 grid and it still holds fps over 150. Pisses me off cause I prefer houdini over maya

andrej22
I'm rly starting to wonder what's the point of quadro cards. Nvidia is conviencing their users these cards are made for applications like houd, maya, max and yet GTXs are owning them badly in all apps.

I think you're onto something

A whole lot of non-workstation cards are supported in the next Houdini release.
http://www.sidefx.com/index.php?option=com_content&task=view&id=2680&Itemid=390 [sidefx.com]

andrej22
also i just tested eg. when I create a 1000x1000 (rows, columns) grid in houdini my fps drops down to 25 (GL 3.3), in maya OGL viewport I can create a 6000x6000 grid and it still holds fps over 150. Pisses me off cause I prefer houdini over maya

Assuming that you're viewing the entire grid, Maya must either be measuring draw time differently or dropping triangles. A GEforce 780 can process 4 tris/clock, and has an average boost clock of 900MHz, giving it a peak triangle throughput of 3.6 billion triangles per second (4*900 million). A 6000x6000 grid has 36 million quads, or 72 million triangles. This means the absolute maximum framerate you could get, ignoring everything else (rasterization, vertex transform, etc), is 3.6B/72M or 50fps. A Quadro K4000 can do 890M tri/s, for an absolute maximum of 12fps. Since ignoring everything else would mean a boring, empty viewport, you're probably going to see a lower rate than that by at least 20%.

Houdini measures draw time by using GPU timer queries, which grabs timestamps from the GPU itself, and is a very accurate way of measuring rendering time. If you instead grab timestamps from the CPU, you'll get an inflated fps. OpenGL commands are processed asynchronously, meaning that the GL command can return before the GPU finishes processing it. This means that some time won't be measured at all. Perhaps this is how Maya is measuring draw time.

Smooth wire shaded is also the most expensive shading mode in Houdini. Smooth shading is often quite a bit faster for large models. I get 100fps for a 1000x1000 smooth shaded grid on a GEforce 690 (one GPU only, so more like a 680), and roughly double that for smooth shaded. However, since my monitor caps out at 60fps, I'm okay with that :-) A 6000x6000 grid is 14fps in smooth shaded, and 1.3fps in smooth wire, which is definitely bumping into the hardware's limits. Are you measuring Maya's performance in a similar shading mode?

Maya - houd comparison

Grid 1000x1000. WireF on Smooth shading

Houd : http://oi57.tinypic.com/b7eusl.jpg [oi57.tinypic.com]
Maya : http://oi61.tinypic.com/eajep2.jpg [oi61.tinypic.com]

Maya - houd comparison

Cube 600x600x600. WireF on Smooth shading

Houd : http://oi58.tinypic.com/iptv6p.jpg [oi58.tinypic.com]
Maya : http://oi62.tinypic.com/24uxhew.jpg [oi62.tinypic.com]

fps info is in bottom right corners.Tbh I don't care that much about the difference between maya and houd viewport performance. I'm interested in why quadro k4000 beats gtx 780 in maya GL viewport - (about 5x faster) while in houdini gl gtx 780 is twice faster. Just seems that Maya is maybe utilizing a quadro card better.

Can you tumble at the scene level please - those fps rates for houdini are at the geometry level.

PS. The forum does not work well with large images. It would be great to resize them down. Thanks!

Viewport - obj level.

I'm interested in why quadro k4000 beats gtx 780 in maya GL viewport - (about 5x faster) while in houdini gl gtx 780 is twice faster. Just seems that Maya is maybe utilizing a quadro card better.

My guess is that Maya is using a polygon offset surface and smooth lines (2 pass technique). Quadros have historically had better line-drawing acceleration that GEForce cards. If that's the case, that is an artificial market segmentation feature imposed on the drivers by Nvidia.

Houdini draws wire-over-shaded surfaces using a single-pass, polygon outlining technique which involves no line drawing - it's all done by the geometry & fragment shader. This technique favours raw shading power - which the 780 has a lot of (3x the shader cores of the K4000).

Houdini also relies on multisampling to smooth lines and polygons, rather than GL smooth lines, as that has some performance issues and artifacts associated with it. This could also play into the 780's strengths, higher memory bandwidth and more output units. Houdini's default AA for GL4 cards is 4x multisampling.

I recently switched the Hidden Line mode from a single-pass technique to a 2-pass line/surface one in H13 (build 325). Do you notice any difference with that?

I noticed that it looks like it should again.

MartybNz
I reckon you should re-run your tests now that Houdini 13 is out

50+fps 3.2mil primitives waves headlight only, smooth shaded
25+ fps 3.2mil primitives high-quality lighting & shadows(antialiased-area), smooth shaded

H13.0.222, AMD 7950, OsX 10.9, GL2.1 viewport

Yes, viewport performance on AMD cards boosted tremendously

I get stable
75+ on HD7950 headlight only, smooth shaded
45+ high-quality lighting & shadows(antialiased-area), smooth shaded , HDRI lighting

this is on GL 3.3 and 13.0.343 and nor more glitches and text corruptions.

Previously pixel representation wasn't working for me for points, and now it is working,

Thanks SideFX

https://www.sidefx.com/index.php?option=com_content&task=view&id=415 [sidefx.com]

Then this I suppose is not rly a good recommendation:“Workstation-class OpenGL graphics cards, such as NVidia Quadro and AMD Fire Pro, are required for professional use in production.
Non-workstation cards, such as GeForce, Radeon, and Intel integrated graphics can be used at your own risk. They may be used for learning and personal use but are not supported: you may experience display problems, slow performance, and the software may exit unexpectedly.”

That recommendation is fine,.

Did you see the one for the next version of Houdini; it would be fair to say it's currently a transition time from that recommendation to the next one.

Graphics Card requirements for the upcoming Houdini release
http://www.sidefx.com/index.php?option=com_forum&Itemid=172&page=viewtopic&t=30927 [sidefx.com]

twod
I recently switched the Hidden Line mode from a single-pass technique to a 2-pass line/surface one in H13 (build 325). Do you notice any difference with that?

Wow I didn't know you made the hidden line mode match the regular wireframe mode. Thanks so much for doing that I also didn't realize you did the previous one for performance reasons since it's using 2 pass now like you said, but it's worth it.

pusat
I also didn't realize you did the previous one for performance reasons since it's using 2 pass now like you said, but it's worth it.

Agreed.