Wich Features Are Multithreaded In Maya

[kafl]

Hi there

I just bought a new machine, with dual Quad xeons, and where pretty dissapiointet about the speed of animated skin, fluid calclation and fur rendering (in mental ray) so just to be sure, does anyone know which features should be multithreaded? and also if not in 8.5 is it fixed in 2008? its a petty that I can not use the 8 cores for anything else than "just" rendering, if that is really the case. It could also just be my setup or settings in maya.

Hope someoine knows and can help since i dont see any specific topics in the helpfiles or on the forums I've searched so far.

Thank you

[deke-kincaid]

Rendering, Paintfx, Geometry cache, nCloth cache(not the ncloth solver), Python scripting, Hair collisions, skinning, draw tesselation, wrap deformer, display of polygons/subdivs.

Maya 8 was the first version to start adding threading to anything beyond Rendering and Paintfx. Remember that 99% of all applications out there don't even take advantage of more then one processor. Also most 3d applications don't use more then 1 processor for anything other then rendering.

[kafl]

Hey Beaker

Thank you for the answer.

The once you mention here is that in Maya 8.5? or 2008. I just opened a animated skinned model and pressed play, and it did not use more that 13 % but the framerate was about 3 fps.? Where did you find that info? is it in any help files/homepages?

Thank you a lot
Karsten

[joojaa]

Playback is is related to the interplay of graphics card, memory and main bus. Possibly deck (disk or rray of them). If you choke up any one of theese your performance drops.

Mainly on screen playback is gfx card. but the idea is that all of these components need to match up in speed if one lags all rest must wait. You now have 4 times the power of computation but also 1/4 iof the bandwidth to allocate for them.

Not everything can be threaded in a useful manner. It does not matter how good harware you have if you work by dumping everything in you suffer a penalty. Working smarter is the game of play here. Its really easy to kill the performance OF ANY computer, no matter how good it is. Its been like that forever. I mean the computational power we have currently is still one 1/100000-100000000 (its a estimate based on the computational capacity of our frontal lobe, problem with theestimate is that it goes up every time we do something more powerfull) of what weed need so having a 10 faster computer does not really mean all that much here unless you work smart.

Then theres the fact that you don't have 8 processors you have 4 so if the computer says to you its using 13% it means 1/4 might be in full use so one proc going full (the virtual processor only kicks in if you have need for it). Besides even in ultimate situations you'd only ever a bit more then one proc anyway.

So in the end having a super multi core machine only means you need to pace your work differently so you have some high processor low memory bandwidth calculations going in the background. Such as some simple rendering. Also threading does not necessarily mean all processor use, it could mean just use 2 and thats it, because synchronizing processors also means more work, but not necessarily better performance.

And this is also why computer power hasn't gone some much further as wed liked. Our theoretical maximum speed is astonishingly much more, but the true computational capacity hasn't. Welcome to the world of scientific computation, thats why we have cluster machines and super computers side by side, only supercomputers can solve certain problems.

QUOTE
Where did you find that info? is it in any help files/homepages?

relese notes is one place where you will find it.

[kafl]

Hey Joojaa

Thanks for your answer, i think I follow you a long way, but still have some questions.
I'm just trying to understand this in a simplyfied way, so i'm not sure about all this, but the way I see it is that for me to be able to see advanced big fluids and dynamics scenes, and extreeme high poly models, the graphics card needs to work hard, and that might be a bottleneck. If I just want to render one of the above i guess that the graphisc card does not play the biggest role?

I know that as you say there might be bottlenecks in different areas, but when "just" rendering "normal" stuff, it renders 10x the speed of my "older" machine. (wich was pretty old.) I dont know how or why, but I hope that its possible to make these other things use more of the processor power too. And I heard that more features in Maya 2008 was being optimized for multicores.

Btw the scene is pretty simple, two objects skinned to a skeleton, polycount is about 40.000, the system is two Quad core Xeons on an intel s5000xvn board with 2 gigs of ram. The graphics card is Quadro FX1500. (dont know if it would have been better with a new Geforce card?)

Thank you
KArsten

[yashu]

8.5 threading
8.0 threading
in 2008 fluids are further threaded, as well as skining

[dwgregg3d]

Opening the Windows Task Manager (control-alt-del) is the way I like to get an intuitive sense of multi-tasking in software. If all processors are pegged at nearly 100%, you know multi-threading is cooking good. If one processor is pegged and the other just bumping along the bottom, you can bet that what ever you are running isn't multi-threaded.

If the procs are all cooking but not all the way, it could be a mixture of any number of variables.

In any event, a list of what is and isn't is the ultimate guide.

[deke-kincaid]

QUOTE(kafl @ 08/29/07, 12:34 PM) [snapback]272478[/snapback]
I'm just trying to understand this in a simplyfied way, so i'm not sure about all this, but the way I see it is that for me to be able to see advanced big fluids and dynamics scenes, and extreeme high poly models, the graphics card needs to work hard, and that might be a bottleneck. If I just want to render one of the above i guess that the graphisc card does not play the biggest role?
Real time viewport or hardware rendering has to do with the video card. Rendering to MR or the software renderer has nothing to do with the video card.

QUOTE(kafl @ 08/29/07, 12:34 PM) [snapback]272478[/snapback]
I dont know how or why, but I hope that its possible to make these other things use more of the processor power too. And I heard that more features in Maya 2008 was being optimized for multicores.
You can't make things multithread if they aren't programmed that way.

Look at using the 64 bit version of maya and get more ram if you want to see more speed on large scenes. Another way of speeding up renders(pending you have enough ram) is to not multithread it but instead submit a render per processor. That utilizes more ram and more hard drive resources, but does render much faster. The trouble is that somewhere in the 4-8 range you get diminishing returns on multithreading. Multithreading is not a linear, so 8 processors does not equal 8x the speed(more like 4-6x) all depending on the operation. The other trouble is that some operations just don't thread very well, like anything simulation based because it needs to calculate a result for one frame before it can go to the next.

[joojaa]

QUOTE(beaker @ 09/01/07, 03:57 AM) [snapback]272616[/snapback]
The other trouble is that some operations just don't thread very well, like anything simulation based because it needs to calculate a result for one frame before it can go to the next.

remember the saying more cooks dont make better soup? We all have had this experience, some tasks are just easier to do alone than do in groups like calculation. If it can reliably be split up then no rob, if not then its pretty damn hard to do anything about it. So if the guys 2 -8 must wait for guy 1 to compute something its a problem.

That said 3d can mostly be treaded, but simulation cant.