Jump to content


Photo
- - - - -

performance test Comments


  • Please log in to reply
70 replies to this topic

#21 Justin Montoya

Justin Montoya

    Advanced Member

  • Members
  • PipPipPip
  • 334 posts

Posted 05 December 2014 - 12:57 PM

3dworks-

 

I'm not sure why, but your memory is only showing at 800mhz.  That is really slow for DDR3 memory.  If I were you, I would pull the ram and check to see if it is rated at faster.  It should really be 1333mhz for best performance.  If it's only rated at 800mhz, your slower memory is likely bottlenecking your system, and should be replaced with 1333mhz ecc DDR3.


  • Welchrisk, Palossole and Deroykeve like this

#22 3dworks

3dworks

    Advanced Member

  • Members
  • PipPipPip
  • 336 posts
  • LocationBerlin, Germany

Posted 05 December 2014 - 01:08 PM

yes there seems to be a multithreading issue...

whatever manual setting i use 6, 12 or 24 processsors, the program is only using a very small part of 'user' CPU potential.

 

just as an experiment, i did a fast - quick and dirty - conversion to lightwave in 2 minutes. export from FZ8 as layered LWO with textures. almost no material tweak or optimization, just replaced the default glass with a nodal glass material and added a bit of soft reflections on the stairs and the window mullions (just couldn't resist!). then i set the scene up with a simple sun + sky environment, matched camera and sunlight and rendered it with cached montecarlo 2 bounces full raytracing, medium/high AA.

 

at the same scene resolution as in imager, 5100x3300, it rendered in 8 minutes 21 seconds, included the GI calculation from scratch, which took 4 minutes and which at subsequent renderings would not needed to be recalculated. if someone wants to test this as well in a lightwave demo, if you want the LW scene, just contact me privately...

 

this is the unretouched result straight out of LW. probably the sun needs to be brighter, but thats only a rough test ;-)

Attached Thumbnails

  • Render Time Test_v8-Interior_LW.jpg

Edited by 3dworks, 05 December 2014 - 01:13 PM.

3dworks visual computing + photography


#23 Tech

Tech

    Moderator

  • Moderators
  • 4,110 posts

Posted 05 December 2014 - 01:19 PM

Hi Markus,

 

We did not optimize the rendering of this specific file for time -- just to take a "reasonable amount of time" for the render test to be valid -- so you can get similar quality in less time in formZ too -- but that was not the goal for the file we set up.  

 

With the Imager set to use 6 processors, it is expected that a little over 1 processor would be used for about the first minute, but after that, you should see all (6) processors being used.  Is that not the case on your computer?



#24 3dworks

3dworks

    Advanced Member

  • Members
  • PipPipPip
  • 336 posts
  • LocationBerlin, Germany

Posted 05 December 2014 - 01:21 PM

3dworks-

 

I'm not sure why, but your memory is only showing at 800mhz.  That is really slow for DDR3 memory.  If I were you, I would pull the ram and check to see if it is rated at faster.  It should really be 1333mhz for best performance.  If it's only rated at 800mhz, your slower memory is likely bottlenecking your system, and should be replaced with 1333mhz ecc DDR3.

 

that's what the specs for these apple mac pro v5,1 are... it makes no sense to replace this kind of EEC RAM with faster one (and afaik it is not possible), it's part of the machine design of that time- the bottleneck would be the controllers on board, as it was the fastest at that time. however, i have never had the feeling that this would be a 'bottleneck'. even when working with 5k multilayer 32 bit photoshop files (which can take up to 30-40 gigabytes of RAM), i never have to wait for the machine...


3dworks visual computing + photography


#25 3dworks

3dworks

    Advanced Member

  • Members
  • PipPipPip
  • 336 posts
  • LocationBerlin, Germany

Posted 05 December 2014 - 01:22 PM

Hi Markus,

 

We did not optimize the rendering for time -- just to take a "reasonable amount of time" for the render test to be valid -- so you can get similar quality in less time in formZ too -- but that was not the goal for the file we set up.  

 

With the Imager set to use 6 processors, it is expected that a little over 1 processor would be used for about the first minute, but after that, you should see all (6) processors being used.  Is that not the case on your computer?

 

the screenshot you can see is taken at the very end of the rendering process. it does not qualitatively change for all the settings i have tried.


3dworks visual computing + photography


#26 arkona

arkona

    Newbie

  • Members
  • Pip
  • 4 posts

Posted 05 December 2014 - 01:52 PM

I did new test for both versions.

Imager was the only running program.

 

v7  0:04:31

 

v8  0:09:09



#27 Tech

Tech

    Moderator

  • Moderators
  • 4,110 posts

Posted 05 December 2014 - 02:15 PM

Hi Arkona,

 

If you carefully inspect the images, is the quality the same, or are there differences?

 

And you have the Imager Preferences: # Processors set to Use All Processors for both v7 and v8, right?



#28 ASONE

ASONE

    Advanced Member

  • Members
  • PipPipPip
  • 301 posts
  • LocationBoulder, Colorado, USA

Posted 05 December 2014 - 03:28 PM

Isn't it important to list how many cores are performing the test?  

 

i should also note that my test was v7 not v8.


Edited by ASONE, 05 December 2014 - 03:29 PM.

FormZ PRO v8.5.  2017 iMac 4.0 GHz quad core i7 24MB RAM and a 2015 Macbook Pro quad core i7 16MB RAM, Modern Architecture in Boulder and Denver Colorado www.ASONEarchitecture.com 


#29 Tech

Tech

    Moderator

  • Moderators
  • 4,110 posts

Posted 05 December 2014 - 03:35 PM

Hi Paul,

 

It is assumed that all cores are being used, and you can tell how many cores are available based on the generation of the Mac, or the specific processor being used.



#30 Justin Montoya

Justin Montoya

    Advanced Member

  • Members
  • PipPipPip
  • 334 posts

Posted 05 December 2014 - 03:40 PM

that's what the specs for these apple mac pro v5,1 are... it makes no sense to replace this kind of EEC RAM with faster one (and afaik it is not possible), it's part of the machine design of that time- the bottleneck would be the controllers on board, as it was the fastest at that time. however, i have never had the feeling that this would be a 'bottleneck'. even when working with 5k multilayer 32 bit photoshop files (which can take up to 30-40 gigabytes of RAM), i never have to wait for the machine...

 

No it's not.  It should really be using 1333mhz Memory...   http://en.wikipedia.org/wiki/Mac_Pro

 

Since it is showing 800mhz and not 1333mhz, you have a problem with your memory, whether you realize it or not.   You are leaving a good bit of performance off the table by the much slower RAM speed.  I suspect 1 of 2 things happened.  Either the system was upgraded using the wrong memory, or at least 1 of the memory sticks is failing or mismatched, causing them all to run at only 800mhz.

 

If the wrong memory, I trust the OWC group tremendously with their products:

 

http://eshop.macsale...ory#1333-memory



#31 3dworks

3dworks

    Advanced Member

  • Members
  • PipPipPip
  • 336 posts
  • LocationBerlin, Germany

Posted 05 December 2014 - 07:55 PM

justin, thanks for bringing this to my attention! i'm using 5 mac pro's of different types here, and after setting up the machines some years ago, i never double checked the RAM speed with the system profiler anymore and this got me confused. as a matter of fact, 800 mhz was the memory speed in older mac pros's!

 

indeed, i have 10600R 1333 mhz memory installed in all my slots, but apparently, the system is reporting it in a wrong way. i did some further tests and as i said before, all seems good here. then i found a thread on apple's forums about this issue, where it results unclear if this is because the mac pro memory manager is not reporting correctly or if it is a compatibility problem with certain brands of memory. that said, i also read that technically, the mac does never 'downclock' any RAM - it would simply report that i can't be used with that model. also, i never had any issues. all my other render applications (lightwave, maxwell, vray) work at the highest and expected speed. also, if i compare my maxwell benchwell values with other similar mac pros' in the database, they are perfectly in range.

 

in conclusion, the FZ RZ speed should not be affected in any way by this.

 

however, this story got my attention and i will further investigate to know more about that!

 

cheers and thanks again for the comments.

 

markus


Edited by 3dworks, 05 December 2014 - 08:07 PM.

3dworks visual computing + photography


#32 DennisA

DennisA

    Advanced Member

  • Members
  • PipPipPip
  • 408 posts

Posted 06 December 2014 - 11:28 AM

Markus,

It seems that if you limit a 12 core Mac Pro with this test file to 6 cores get much better results.

Give it a try.



#33 3dworks

3dworks

    Advanced Member

  • Members
  • PipPipPip
  • 336 posts
  • LocationBerlin, Germany

Posted 06 December 2014 - 12:38 PM

i did also try 6 cores but didn't get near anything reasonable... at this point i guess it's a compiling issue.


3dworks visual computing + photography


#34 DennisA

DennisA

    Advanced Member

  • Members
  • PipPipPip
  • 408 posts

Posted 06 December 2014 - 03:02 PM

I am experimenting with different computers:

On a Mac Pro 2 x 2.93 GHz-6 Core Intel Xeon

6 processor=12:22:35

8 processor=12:24:37

12 processor=12:39:34
24 processor=1(hrs):46:28
 
On a Mac mini quad i7 processors=12:19:28
 
On a Mac mini (late 2014) dual 2.6GHTs i5 processors (4 cores)=2(hrs):15:48
 
Will try with less cores on Mac mini and also see how powerbook does .
FMB possibly post some different projects so we can compare speed with different standards (no radiosity/ final gather etc).


#35 Justin Montoya

Justin Montoya

    Advanced Member

  • Members
  • PipPipPip
  • 334 posts

Posted 06 December 2014 - 04:01 PM

I think there's some strange threading issues going on here...  at least on the OSX side.  

 

Mac Pro 2 x 3.46 Ghz 6-core Intel Xeons

 

v7 Test - # cpu setting in Imager preferences - test time

 

6 - cpu - 00:19:51

12-cpu - 00:12:31

16-cpu - 00:12:41

24-cpu - 00:13:19

 

So setting my Mac Pro at 12 cpus, thereby disabling hyper threading gives the best results.  I remember years ago, FormZ having issues with Hyper threading in OSX, but I thought that had been resolved.  My next step will be to test in Windows 7, which by looking at the other results, may have much higher performance.  If that's the case, we'll need to find out what's happening in OSX to make is so much slower.  

 

I feel like I should be rendering much fast than this, and that there is a threading problem in FormZ.  If I compare to my other renderer, Maxwell, results, I'm consistently at the highest on OSX, often faster than the same spec Windows machine as well.  Hopefully Autodessys can chime in here.



#36 DennisA

DennisA

    Advanced Member

  • Members
  • PipPipPip
  • 408 posts

Posted 07 December 2014 - 11:12 AM

Tech your fmz test file for 7 does not download in Safari or Firefox? They do not find file?

 

ftp://ftp.formz.com/pub/formz/formz/fz8/Forum_Files/Render%20Time%20Test_v7.zip



#37 DennisA

DennisA

    Advanced Member

  • Members
  • PipPipPip
  • 408 posts

Posted 07 December 2014 - 12:15 PM

These are results for v7:

12:24      Xeon     Mac Pro 2 x 2.93 GHz-6 Core (24 processors)              OSX 10.10.1         v7

19:25      Xeon     Mac Pro 2 x 2.93 GHz-6 Core (6 processors)                OSX 10.10.1         v7

vs Results v8

1(hrs):46:28     Xeon     Mac Pro 2 x 2.93 GHz-6 Core (24 processors)              OSX 10.10.1         v8 

12:22:35         Xeon     Mac Pro 2 x 2.93 GHz-6 Core (6 processors)                OSX 10.10.1         v8

         



#38 Tech

Tech

    Moderator

  • Moderators
  • 4,110 posts

Posted 08 December 2014 - 08:46 AM

Hi Justin, Dennis, and Markus,

 

Yes, in general, the RenderZone rendering is optimized for Actual cores, not Virtual cores.  Therefore, if you have a 12 core computer "with hyper threading" that makes it appear as 24 cores, you will get better results with the 12 core setting on OS X.  (The Virtual cores are handled better on Windows, where they can be more fully utilized.)



#39 arkona

arkona

    Newbie

  • Members
  • Pip
  • 4 posts

Posted 08 December 2014 - 08:59 AM

Both Imager versions are set to use all processors.

The quality of the images are almost the same (v7 5440x3520, v8 5100x3300).



#40 Tech

Tech

    Moderator

  • Moderators
  • 4,110 posts

Posted 08 December 2014 - 10:16 AM

Hi Dennis,

 

That v7 file was removed and included in the main download archive, that was updated to include that file.  You can download that again from here:

 

ftp://ftp.formz.com/pub/formz/formz/fz8/Forum_Files/Performance%20Test%20files.zip

 

Hi Arkona,

 

That's strange, the image resolutions are different?  Please download from the link above and try the tests once more.  Does that give you different results?






0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users