PDA

View Full Version : Quad core QX6700 and Maya Mental Ray


treborwal
12-14-2006, 05:59 AM
I would greatly appreciate some feedback on whether or not if it would be a wise decision to purchase a new pc with an intel quad core processing unit, qx6700, ati radeon x1950 video card, and etc. all for the purpose of low time rendering with mental ray in maya, version 7.0. Currently I'm running off of a p.4 and sometimes takes 8 hours for one render, through the comand line. I definteley cannot tollerate such pathetic rendering performance like that any longer.

And lastly, is maya 7.0 and up a multi-threaded software program to be fully utilized and accessible with intel quad core functionality?

thanks!
Btw, feel free to private message me if neccessary.

pixel mixer
12-14-2006, 07:48 AM
maya likes CPUS no matter how many. it will eat everything you got. get nvidia instead of ati cause in opengl it's far superior to ati.

treborwal
12-14-2006, 09:03 AM
maya likes CPUS no matter how many. it will eat everything you got. get nvidia instead of ati cause in opengl it's far superior to ati.

Nvidia seems like a good choice. But would I benefit from quad core capability?

o-v
12-14-2006, 09:30 AM
mind the licensing issue of mental ray... it are licenses per core!!! i gues maya unlimited has (for maya) 8 licenses, so it shouldnt make probs with licenses... but mental ray - I'm not sure.

tfritzsche
12-14-2006, 02:24 PM
treborwal,
I believe, and correct me if I'm wrong, that Maya will only use one CPU/core, it's only the rendering that is multi-threaded. so modeling,rigging,view ports,etc.. are all single core, at least Maya 7 is this way.

treborwal
12-14-2006, 10:37 PM
Sure, I don't care about rigging, modeling, etc. being one thread. I'm just concerned about rendering being multi threaded.

Roi-Danton
12-15-2006, 12:14 AM
The MR integrated in Maya Unlimited is limited to 8 CPU resp cores per computer licence. So you also can use two Quad-Core Xeons and still meet the licence.

From the technical point ... the Raytracing algorithm of MR loves multi-threading (for years now), so it will scale almost 100% with your numbers of CPU resp core.

treborwal
12-15-2006, 12:25 AM
That's great to know

*I called boxx to see how much the apex 8 system was; $35,000, which is insane. But with a 16 core processor nothing can't be done
Quad core will do for now

beaker
12-15-2006, 05:41 AM
That's great to know

*I called boxx to see how much the apex 8 system was; $35,000, which is insane. But with a 16 core processor nothing can't be done
Quad core will do for nowMultithreading quickly diminishes beyond 4-8 procs, so don't worry about that 16 core processor just yet. Also you never get 100% from each proc when multithreading, more like 60-80%. The way we do it in pro render farms is to not even multithread at all and just submit a frame per proc so you do get 100%. You need a lot more ram for that though.

treborwal
12-15-2006, 03:46 PM
aghh, so if 4 processors won't be fully uttilized, I suppose I'm better off getting an e6700 dual core instead. I didn't know from 4 and above that mulit thread wouldn't be sufficient

kevinseven
12-15-2006, 04:18 PM
//aghh, so if 4 processors won't be fully uttilized, I suppose I'm better off getting an e6700
//dual core instead. I didn't know from 4 and above that mulit thread wouldn't be sufficient

I don't think that's what they were saying. 1 CPU with 4 cores or 2 CPU with 2 or 4 cores each is going to be a great rendering solution. What Beaker was saying is that you're probably better off sending 1 frame for each CPU (or core) than letting Maya divide a single frame between the available CPU/cores. With enough RAM, you'll start to see 100% CPU usuage with that kind of workflow.

I'd also suspect that he's pretty much talking about batch rendering on a farm as opposed to the 1000s of preview renders some of us do in a day.

Another thing to consider is I'm a lot more productive on 2 medium level machines than I am on 1 high end machine. So you might think about dividing you budget up that way. Get 2 dual core machines and run one as a mray satellite slave. Put a cheap video card in the slave and use it to batch render when you're doing other stuff like modeling. If you configure it to be an almost exact duplicate of you primary machine then you have a spare when a power supply or MB dies on you the night before a deadline. Just a thought.

beaker
12-15-2006, 06:51 PM
I don't think that's what they were saying. 1 CPU with 4 cores or 2 CPU with 2 or 4 cores each is going to be a great rendering solution. What Beaker was saying is that you're probably better off sending 1 frame for each CPU (or core) than letting Maya divide a single frame between the available CPU/cores. With enough RAM, you'll start to see 100% CPU usuage with that kind of workflow.Yes, thats exactly what I meant. You will still get a speed up from multiple cores. Just that multithreading doesn't scale linearly like what Roi-Danton said. 4x3ghz procs does not equal a 12 ghz machine. You only get maybe 60-80% increase in speed with each proc you add. So rendering with 4 cores multithreading vs 1 core does not equal 4x speed up, more like 2.5x-3x.

As you add more cores the speed up diminishes because these apps are not written as massive parallel apps like for example something written for a cluster. So a 16 core box might only be 6-8x the speed of a single core. I'm totally pulling numbers out of my ass here and don't have any solid numbers to show with this. Back in the Sgi days we has 32 processor onyx and it was only 2x faster at rendering a single frame over a 8 proc.

treborwal
12-15-2006, 09:38 PM
Kevinseven, I will definetely think about what you stated as far as budgeting and having one cpu do all of the dirty work.

And most of the rendering is mainly through the command line instead of batch render

tbaypaul
12-16-2006, 01:39 AM
yes, maya's render is multithreaded as is mr...you get one maya license and either 6 or 8 mr satalite licenses depending on which package of maya.....

Roi-Danton
12-16-2006, 09:48 AM
I said almost 100%.

Look what a guy from Erlangen did with a special Quake 4 Raytraced version: There the four cores of the Intel Core 2 Extreme QX6700 scaled 3,84 (http://www.idfun.de/temp/q4rt/benchmarks.html) (with a realtime application) compared to a single core, what's more than 80% per additional core.

Here you can find a comparison between 2 Core and 4 Core processors (though the report is in german the charts are all what's interesting) with Maxon Cinebench & Lightwave:
http://www.computerbase.de/artikel/hardware/prozessoren/2006/vorschau_intel_core_2_extreme_qx6700/8/

Furthermore, beaker said "Multithreading quickly diminishes beyond 4-8 procs", so a 4 core processor is worth it's price for raytracing.

techmage
12-19-2006, 08:54 PM
I second the notion of nvidia over ati. Performance wise, the two companies are similiar. But nvidia drivers are much better.

kryoboy
12-23-2006, 01:59 PM
yeah, get an nvidia card for sure.. an nvidia card will be your best choice of performance/support/options for professional 3d packages.

bgawboy
12-28-2006, 12:59 AM
That 4-8 processor dropoff is considering rendering on multiple machines across a network. The multi-core machines don't drop off as much because they are all on the same machine. Then, make sure to get lots of RAM on that machine. mental ray scales very well, better than most other applications, on a mult-processor multi-core machine. I've heard of good performance from 8 to 32 total cores on those boxx machines.

Matt Grdinic
12-28-2006, 06:42 PM
On more cores:
Maya 6.5 and possibly < 6.5 mentalRay seems to limit you to just one core when rendering in the Application, e.g., the render view window. However, when you use Render > Batch Render: You can set how many rendering threads to use.

On my quad core Xeon I set it to 4, and sure enough, it uses all four cores. When in the render window though, it never uses more than two threads. Bug?

No matter, let me say this right now: It's worth it...every last cycle devoted to a render it always worth it. For reference, using the Cinibench rendering test on the quad core, I score over 1400 when all cores are enabled. My lowly Pentium D scores a mere 500. Bottom line: Get as many cores as you can.

On mental ray satellite:
Before my work place got me the quad core, I was using the Pentium D. It was too slow for our tight deadlines, so for one job we actually took the time to create a render farm with some dual Xeon's that normally host SQL or serve web pages. These servers were no slouch, but the sad fact is, the rendering times were still painfully slow, despite having 8 full time dedicated slaves on a gigabit network. Bottom line: Unless you're sending individual frames out to each slave, Satellite is no faster, and in some cases, is actually slower than using your host machine alone. Obviously this means if youíre not doing animation work, Satellite is basically worthless.


Reality Check:

http://features.cgsociety.org/story_custom.php?story_id=3583

"Rendering such a movie on film resolution is no small task: thankfully Bowie State University donated access to XSeed, their cluster of 240 Dual-core Xserves to render the movie. XSeed took over 125 days to render the movie, using up to 2.8GB of memory to render a single frame."

The simple fact is when we start adding SSS, Global Illumination with Final Gather, and especially Motion Blur in mentalRay, render times will always be painfully long.

The best advice to follow is get the best hardware you can afford, Ďcheatí when ever you can, and use tools better suited for specific tasks. For example, Motion Blur in mentalRay is terribly slow for our tight schedules: The solution: Gelato. The catch: Only works with Nvidia's GPU's.

Check it out:
http://www.nvidia.com/page/gz_home.html

Their motion blur is as good if not better than Render Man, and it's free.

So yeah, best of luck!

bgawboy
12-29-2006, 11:28 PM
When using satellites, it is better to put multiple processors, multi-cores on a single satellite slave than to use multiple slaves each with a single core/processor. Again, its a matter of transfering data on a network vs. a system bus.

Also, satellites make more sense in more complex scenes. For very simple test scenes, the overhead may swamp out the resulting speedup. Always test with the appropriate scenes to make the decisions.

I have seen excellent satellite results used by a look dev team that shared a multi-proc satellite rendering accelaration box.

chilio
01-02-2007, 06:19 PM
I am using quadcore from intel (QX6700) and there is difference between the render engines you choose. Surprisingly if you use :
1. Maya standard render engine - all cores are going at full speed about 100% usage
2. Mental ray - first 2 cores are running at about 50%-51% usage, third at about 23% and fourth core at 75-80%

I believe that these proportions may slightly differ based on scene content.

Therefore it seems that mr algorithm isn't best optimized for multi-core processors.

*maya 8 used in batch render mode

If you need any suggestion I would go quad core for sure. Why?
In comparison to x6800 (dual core) which I also tested qx6700 has advantages which show almost only in rendering tasks. When application is unable to use 4 cores it is only few % slower (that comes only from slightly higher operating frequency) 2.93 Ghz to 2.66 Ghz. In all other situations you get a real perfomance boost. The increase as stated in earlier posts isn't 100% (to 2 core) but 70-80% sounds pretty good to me (and at comparable price).

And the last thing - the world goes multi core (and 4 cores are not the end) so we won't wait too long for all apps to use all cores :)

Hope it cleared the topic a bit.

By the way BEST 2007 for Yaa all.

DoomGen
01-02-2007, 08:28 PM
I use a dual - dual core opteron and get 100% on all processors with mental ray so I believe it is optimised fine for multi-core processors (if I understood your comment correctly). It may be different with 4 cores being on one processor but i wouldn't have thought so

Matt Grdinic
01-03-2007, 12:25 AM
@DoomGen

It uses four cores just fine, but only when I use batch render, not when rendering in the render window. It sounds like a bug, but keep in mind that when you use batch render it asks how many threads to use. It's almost as if the render window command doesn't use all available cores by default.

DoomGen
01-03-2007, 10:04 AM
ah sorry, I thought I may have misunderstood

bgawboy
01-03-2007, 05:21 PM
OK, there is a problem with a workaround for batch in Maya 8.

To clarify, we're talking about two different places to set threads, one for rendering in the render window, and the other, for batch render. Both are set by clicking in the little boxes to the right.

For Maya 8

First assume that from the Render menu, we set mental ray rendering
1 Render Using -> mental ray

Then to set the threads for rendering in the render window
2 Render Current Frame -> [] (the little box to the right)
-> mental ray Render Option pop up
-> choose how many render threads to use
(maximum: 8)

Then, one would typically do the same for rendering with batch, ie
3 Batch Render -> [] (the little box to the right)
-> mental ray Batch Render Option pop up
-> choose how many render threads to use
(maximum: 8)

However, this doesn't work in Maya 8. This is actually noted in the release notes if you are able to find it. Look in:
Maya Help > Using Maya > General > Release Notes > Rendering > mental ray for Maya rendering

From those notes:
"Changing the number of rendering threads to use in the mental ray Batch Render Option window has no effect

"The number of threads is saved in the Maya Preferences. Maya Preferences are not saved until Maya closes or preferences are saved.

"Workaround

"The workaround is to save preferences first by selecting
Window > Settings/Preferences > Preferences and clicking Save before calling Batch Render.


And it is not clear about this, but because there is no actual setting in the preferences UI, I assume that one must set it from that mental ray Batch Render Options pop up, and then save preferences.

nerdfx
01-10-2007, 01:45 PM
I have a question regarding this topic: I have a small budget only. So what would be the best solution for increasing render performance using maya and mentalRay?

a.) getting a mutlicore workstation

or

b.) getting some strong singlecore renderslaves to distribute renderjobs across a network?

For example: would two 3Ghz singlecore CPUs with LOTS of RAM and huge harddisks be sufficient for smaller broadcast animation work? Or am I better of buying one mutlicore workstation?

Thanks

Matt Grdinic
01-10-2007, 11:52 PM
I have a question regarding this topic: I have a small budget only. So what would be the best solution for increasing render performance using maya and mentalRay?

a.) getting a mutlicore workstation

or

b.) getting some strong singlecore renderslaves to distribute renderjobs across a network?

For example: would two 3Ghz singlecore CPUs with LOTS of RAM and huge harddisks be sufficient for smaller broadcast animation work? Or am I better of buying one mutlicore workstation?

Thanks

A.if your going to be rending out still scenes.
B if youíre rendering animations.

mentalRay Satellite is slow with single frames, in some cases slower than rendering on a local box. However, if you render individual frames out to each slave, then itís faster.

However, if youíre trying to decide between just 2 single coreís or 1 multi-core, then itís a no brainer, go for the multi-core. With a descent amount of ram and hard drives your looking at least 2300k total for two boxes, whereas one quad core would run you about 2200k with the same ram/storage specs.

nerdfx
01-11-2007, 01:37 PM
Thanks for your reply! :thumbsup:

I'm considering spending about 4-5k USD maximum on hardware alone and I want to do short animated stuff mainly (broadcast animation). BTW, is there any advantage in getting a dual-processor instead of a dual-core system?

I'm open for explicit recommendations! And again: thanks!! :)

Roi-Danton
01-12-2007, 12:24 AM
Depending on your Maya and MR licences. If you have more than one:

3-4 times Core 2 Duo E6600 (http://geizhals.at/deutschland/Intel_Core_2_Duo_E6600_Sockel-775_boxed_BX80557E6600_a200142.html) or 2-3 times QX6700 (http://geizhals.at/deutschland/a233099.html) resp Xeon Quad Core (http://geizhals.at/deutschland/a225416.html), use each system to render own frames. Especially the performance/energy consumption ratio of the Quad Core is good (see this power consumption test (http://www.hardtecs4u.com/reviews/2006/intel_kentsfield_leistungsaufnahme/index4.php), made with a special mainboard to get realistic results) and you can overclock them easily (at least the C2D reaches 3GHz or more with aircooling).

Further examples:
Mainboard: Gigabyte GA-965P-DS4 (http://geizhals.at/deutschland/Gigabyte_GA-965P-DS4_dual_PC2-6400U_DDR2_a203496.html) - good priced mainboard, all what you need (if you need CrossFire or SLI get the DQ6 or the Asus P5B Deluxe) and good for overclocking, too.
RAM: Corsair XMS2 2048MB PC2-6400 CL4 (http://geizhals.at/deutschland/a203549.html) or similar from G.Skill. Qualified for overclocking. (don't think you need to buy ECC-Ram for rendering)
Graphics: ? ... I have a 965G-DS3 with integrated video chip for my C2D. For simple scenes this is more than enough. I've read nVidia cards would have better drivers for CG needs. If you use Hardware rendering, then have a look at the G80 (detailed, easy written report about the technology used by G80 (http://www.3dcenter.de/artikel/2006/12-28_a.php) (dt)).

nerdfx
01-12-2007, 10:29 AM
Thanks for your advice! I'll check it out later. :thumbsup:

treborwal
01-14-2007, 11:00 PM
I may know of a site where you can get hardware for cheper, nerdfx

lollygag
01-15-2007, 11:38 PM
The MR integrated in Maya Unlimited is limited to 8 CPU resp cores per computer licence. So you also can use two Quad-Core Xeons and still meet the licence.

From the technical point ... the Raytracing algorithm of MR loves multi-threading (for years now), so it will scale almost 100% with your numbers of CPU resp core.

16 as of today Maya 8.5.

calmasacow
02-28-2007, 09:28 PM
the 8 cpus for maya mental ray only apply to remote cpus. you can have as many cpus in your main machine as you can it doesn't matter.

CGTalk Moderation
02-28-2007, 09:28 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.