PDA

View Full Version : How much does cores/threads affect modeling, dynamics and rendering?



OlaHaldor
10-21-2013, 02:22 AM
On my home computer I have an i5 2500K with 4 cores and 8GB RAM. I'm strongly considering upgrading to an i7 with 6 cores and hyper threading. I would also upgrade to new RAM sticks so I get a total of 16GB RAM, leaving four RAM slots open to upgrade to 32GB in the long run if it turns out I go nuts over not having enough.

Just wondering how much of an effect I can expect in terms of work with Modeler, calculating bullet dynamics, handling large amounts of polygons and textures and not least, rendering.


What I've looked at is the
Asus P9x79 Pro motherboard and i7 4930K CPU.


Thoughts?

Waves of light
10-21-2013, 03:21 AM
This is a good thread for information regarding the best PC setups for render times:
http://forums.newtek.com/showthread.php?133251-11-5-s-BenchmarkMarbles-lws-share-your-machine-s-render-time-here

Not much is going to make Modeler perform better, that's just the way it is. A faster CPU means better VPR (as VPR is CPU and not GPU dependent). More Cores and more RAM = better render times.

I've got a funny feeling that dynamic calculations are multi-threaded on a PC, but single threaded on a MAC, but I will wait for someone with more knowledge on that side of things before being so bold.

EDIT: I also seem to remember that multi threaded dynamics came in after LW 9.6

OlaHaldor
10-21-2013, 03:47 AM
Thanks! That really makes me wish I could double the budget and get a dual xeon setup!

spherical
10-21-2013, 03:47 AM
That's one of the boards and similar CPU I run. Recommend them highly. Makes a big difference in Layout but Modeler isn't thread-aware yet, so not much of a help. A good video card does help the OGL, though, in both.

OlaHaldor
10-21-2013, 03:53 AM
Thanks, spherical.

I have a GTX Titan. Originally bought it for CUDA intensive tasks for my Mac Pro, but turned out neither drivers or power supply liked it, so I swapped the Titan for the GTX 680 on the Mac, and got myself set for years to come on the PC. :)

Bottom line is; I don't think GPU will be a problem yet.

Waves of light
10-21-2013, 03:53 AM
Thanks! That really makes me wish I could double the budget and get a dual xeon setup!

I know, I'm with you on that one. The upfront extra cost would pay for itself in the long run, especially if you're doing a lot of renders, or frames to video.

@Spherical: Yep, a decent video card will help with OpenGL navigation, but it won't help with tumbling (where you rotate geometry inside Modeler) for large poly models.

OlaHaldor
10-21-2013, 03:57 AM
I have a Mac Pro (2xQuad 2009) for video, and I work a lot with color grading and finishing. I find it the choke point is hard drive speed, not so much the CPU. Having part of the processing utilizing CUDA helps a lot. Rendering 2K ProRes 4444 at more or less 50fps is quite ok in my book. :)


Might sound weird I have a Mac Pro but want to upgrade the windows PC. A lot of the reason is I want to use it with World Machine as well.

spherical
10-21-2013, 09:15 PM
Yep, a decent video card will help with OpenGL navigation, but it won't help with tumbling (where you rotate geometry inside Modeler) for large poly models.

And why do you say that? I'd venture a guess but don't want to make any assumptions. I know that a capable card sure does make a difference here, though.

JonW
10-22-2013, 12:08 AM
If you are mostly rendering frames a handful of 3930k bare bones boxes is the way to go.

OlaHaldor
10-22-2013, 02:13 AM
I've never done any farming with LightWave (even though I grew up on an agricultural type of farm. They're quite different. Who would've known :D ). Why isn't there a setup video made by LW officials like all the other cool stuff they show off on their YouTube channel?

Waves of light
10-22-2013, 02:24 AM
And why do you say that? I'd venture a guess but don't want to make any assumptions. I know that a capable card sure does make a difference here, though.

High end cards will make a difference when navigating around your model in Modeler. But you will see no major benefits from having a high end card when it comes to tumbling or manipulating high poly objects. Lewis seems to be the most knowledgeable when it comes to this sort of thing, here's a couple of threads on the subject:

http://forums.newtek.com/showthread.php?135013-NVidia-Quadro-cards-best-settings
(look at post 2)

And this one showed the speed of CORE (when it was in development) vs LW10 and Modo... LW3D Group have since brought in some of the CORE features into LW Modeler, but (as Lewis states) it still has slow down when you execute them, as it's presumably dumping data to the GPU:

http://forums.newtek.com/showthread.php?134694-LW-vs-modo-vs-Core-)

spherical
10-22-2013, 02:33 AM
High end cards will make a difference when navigating around your model in Modeler. But you will see no major benefits from having a high end card when it comes to tumbling or manipulating high poly objects. Lewis seems to be the most knowledgeable when it comes to this sort of thing, here's a couple of threads on the subject:

http://forums.newtek.com/showthread.php?135013-NVidia-Quadro-cards-best-settings
(look at post 2)

And this one showed the speed of CORE (when it was in development) vs LW10 and Modo... LW3D Group have since brought in some of the CORE features into LW Modeler, but (as Lewis states) it still has slow down when you execute them, as it's presumably dumping data to the GPU:

http://forums.newtek.com/showthread.php?134694-LW-vs-modo-vs-Core-)

That's what I figured. Not talking about Quadros at all. A Good Card with a decent amount of video RAM is what I'm talking about. Can't tell me that dropping back to a Geforce 6800GT is going to perform just as well as a GTX670.

OlaHaldor
10-22-2013, 02:39 AM
Would the Titan be considered "good enough" then?
And wow.. I wish Core had continued development when I see that. Insane!

Waves of light
10-22-2013, 03:17 AM
That's what I figured. Not talking about Quadros at all. A Good Card with a decent amount of video RAM is what I'm talking about. Can't tell me that dropping back to a Geforce 6800GT is going to perform just as well as a GTX670.

For navigating around an object in OpenGL, no it will not outperform a newer faster card. But for tumbling a whole object and/or transforming/moving polys and points in a large poly model you will see no major benefit between say a 9500GT (my previous card) and a Asus GTX 580 (my current graphics card). This is because these functions use old code, so GPU doesn't come into it. The new features (Axis Translate, Transform, Axis Scale, etc.) are quicker, but I don't believe that has anything to do with GPU, it's just that the code for those features uses a new geometry core and runs quicker... but you still get a small lag when you activate and deactivate those functions, as it's copying the mesh information.

EDIT: @OlaHaldor - A TITAN will be great for OpenGL performance and GPU renders (e.g. Octane) but it will not boost manipulation performance of high end poly models in Modeler.

SECOND EDIT: And there's a reason why I went for a second hand 580 over a new 680/690 and that's because there's an issue with 600 range because Nvidia crippled the CUDA performance a bit in the Kepler series.. I don't think this has much bearing on LW, but some of the older 500 series out-do the newer 600 models in 3DCoat, especially:

http://3d-coat.com/forum/index.php?showtopic=14766&hl=

erikals
11-04-2013, 02:47 AM
for Modeler speed i think i'd do this >
use the fastest known CPU technology with high MHz speed but few threads
this one is reasonable > Intel Core i7-4770K

also, as far as L/M speed goes, check >
http://www.youtube.com/user/erikalst/search?query=speed

Carlo_Jongen
02-09-2014, 06:09 AM
I was about to post the question concerning Dynamics and Multi Core use. I'm curently working on a project with a lot of Bullet Dynamics. ( Using a Dual Quad Core machine ) It takes some time for the symulation to complete. Because of this I looked at the Processor use in the Task Manager. And it apears that the PC is not using all of it's Cores to calculate the symulation. So same here, I can tell that Bullet Dynamics in LW is not using all of it's cores on a PC.

I was wondering why LW is not using all of it's cores with Bullet Dynamics?

hrgiger
02-09-2014, 08:55 AM
Not much is going to make Modeler perform better, that's just the way it is. A faster CPU means better VPR (as VPR is CPU and not GPU dependent). More Cores and more RAM = better render times.

Well more cores and faster cores will increase rendering speed but more RAM wont help at all. Unless youre runnning out of RAM and its being forced to segment the rendering.

And yeah, modeler is a lost cause as far as improving mesh manipulation with better hardware.

Carlo_Jongen
02-09-2014, 10:15 AM
Well more cores and faster cores will increase rendering speed but more RAM wont help at all. Unless youre runnning out of RAM and its being forced to segment the rendering.
And what about the amount of cores with LW Dynamics? I have the feeling that LW is not using all of it's available cores.

hrgiger
02-09-2014, 02:00 PM
And what about the amount of cores with LW Dynamics? I have the feeling that LW is not using all of it's available cores.

No. I have 12 threads and I only see one showing any activity during a dynamics calculation.

OlaHaldor
05-05-2014, 02:15 AM
Due to budget, I went with the i7 4770K and I've overclocked it to 4.3GHz. Didn't get it stable at 4.5GHz.
I'm using a Corsair H100i for cooling, and it's mostly dead silent and performs very well. During load at CPU stock speed (3.5GHz) the temps would rise only 10 Celcius degrees to a maximum of 39 degrees! At 4.3GHz it's about 65-70 degrees.


I made a little benchmark scene for me to render. 625 frames long, a camera swooshing over a ground plane with an 8K texture and a few low poly structures with one or multiple 2K textures UV mapped to it. It may not make sense for most of you, but for me it's a valid benchmark as it's these kinds of things I currently do the most.

With the i5 2500K I used previously, the scene rendered in about 40 minutes. With the i7 4770K at 4.3GHz, it rendered in 15 minutes! Hugely impressed with that as well. And very satisfied! :) In comparison, my Mac Pro 2009 (2 x 4-core XEON 2.26GHz) would use just under 30 minutes.