CPU Render Benchmark

[robcat2075]

It would be useful to find out how various CPUs do with A:M, for those who are considering new machines.

 

This PRJ is designed to test the major render time hits like reflection, refraction, multiple ray-traced lights, shadows, anti-aliasing and combiner materials without touching on disk and RAM hogs like bitmaps or hair.

 

Please download this PRJ

 

ThreeTeapotsBenchmarkv005.prj

 

select the camera view in the chor.

 

If you are testing v16 or later, load this render preset when you render to file.

 

benchmark.pre

 

If you are testing v13-15 set your Options>Render Settings to "use settings from Camera" before you render to file. The PRJ has every possible render parameter set in the chor camera.

 

In either case you will need to set a file path.

 

Then hit render and wait. You should get a small jpg (320x240) exactly like this in your "My Documents" folder (Mac Users may have to select a new location):

 

 

 

 

Report your:

 

version of A:M (see Help>About A:M)

render time min:sec

CPU Brand and model

Actual CPU speed in GHz

how many cores A:M is using

RAM

OS

 

 

mine was:

 

15h

19:59

AMD Athlon XP 3200

1.921 GHz

1 core

2 GB RAM

Windows 2000

 

 

Thank you!

[jason1025]

Report:

Version 15.H

render time min:sec 10:03

CPU Brand and model 2006 Intel Zeon Woodcrest 3GHZ Processor

Actual CPU speed in GHz

how many cores A:M is using: single core I think unless 15.H is capable of harnessing the power of multiple cores for render to file.

RAM 12GB 666MHZ

OS Windows Vista Ultimate 64bit

 

 

mine was:

V15.H

10:03

late 2006 Apple Mac Pro Tower intel

Intel Zeon Woodcrest 3GHZ Processor

1 core

RAM 12GB 666MHZ

Windows Vista Ultimate 64bit

 

 

My next test will be rendering the same project in multiple instances to see the results and at what point I get diminishing returns.

 

[serg2]

Report:

Version 15.H

render time min:sec 08:00

CPU Intel i7-860 2.8 GHZ RAM 4Gb/1066 WINXPx64

A:M is using: single core

 

[ludo_si]

19 minutes, it is too slow for this kind of scene. I hope that hash will make an effort on the rendering for the transparency.

there is refraction... but without refraction, it's very good

I think that

51 min 23 sec

 

AMD ATHLON 64 X2 DUAL

core processor 3800

2.01 GHZ, 2GO RAM

windows xp

[serg2]

What is the GHZ ? is that a 2.8?

 

Thank

yes - 2.8 GHZ 4core (x2 HyperThreading is ON)

I have corrected post

[jason1025]

Did some tests using 9 instances and then again with 8 instances all at the same time on my 8 core machine. All procs were pegged/maxed out according to the task manager.

 

Average time for 9 instances at once 11:16

Average time for 8 instances at once 10:12 (one instance for each proc)

time for 1 instances 10:03

 

 

Conclusion:

 

I think its safe to say you can run 1 instance of am per proc.

Diminishing returns when I added one additional instance above the number of procs.

Also I am not sure how healthy it is for your computer to run for hours on end with all procs maxed out. My system gets very hot. Recommendation keep the air conditioning on or step it down to 1 less instance then the amount of procs you have to give the system some head room.

 

Although not exactly, It appeared that approximately one proc was in use for each instance and as projects finished up rendering the task manager dropped off consecutively.

 

9instances

8Instances

[TheSpleen]

mine was:

 

9:48

Intel Core 2 Duo (2.4 GHz)

 

6.144 GB RAM

Windows vista

[jason1025]

I didnt know they made the i7 in an 8 core. I thought is was just a 4 core. Or is it? Is it the hyper threading that makes it look like 8?

 

 

If you have the time and are curiouse can you test 8 instances at once and average the render times?

[Paul Forwood]

This test isn't very practical unless everyone is using the same version of A:M. Render times fluctuate dramatically between different versions as modifications to the code take place. Also the choreography should include things such as particles, fog and volumetrics if you want the test to reflect more realistic production demands.

[robcat2075]

This test isn't very practical unless everyone is using the same version of A:M. Render times fluctuate dramatically between different versions as modifications to the code take place.

 

Take the PRJ and show us. People would say that but they never posted actual comparisons.

 

 

Also the choreography should include things such as particles, fog and volumetrics if you want the test to reflect more realistic production demands.

 

Realistic production demands are whatever someone puts in their production. There is not single test for that

 

As I said at the top, this is a test designed to focus on the CPUs processing power for A:M.

 

While everything uses the CPU Particles introduce introduce other significant variables like RAM access speed and total available RAM. Those are not the variables I'm trying to look at IN THIS TEST.

 

I want to examine as few variables at a time as possible.

 

I just want to examine the relative processing power of CPUs. Those will be informative whether you use hair or fog or whatever in your particular production.

 

Nowhere am I claiming that three teapots is a typical scene. It does however invoke a set of loads that are mostly CPU intensive and much less of all other resources. That makes it a fair test OF THE CPU.

 

 

For the moment, I just want to examine the relative processing power of CPUs.

 

 

And if I haven't mentioned it up to this point, for the moment, I just want to examine the relative processing power of CPUs.

[jason1025]

Report:

Version 15.H

render time min:sec 42:34

CPU Brand and model 2006 Intel Zeon Woodcrest 3GHZ Processor

Actual CPU speed in GHz

how many cores A:M is using: single core I think unless 15.H is capable of harnessing the power of multiple cores for render to file.

RAM 12GB 666MHZ

OS Snow Leopard OSX 10.6

 

 

mine was:

V15.H

42:34

late 2006 Apple Mac Pro Tower intel

Intel Zeon Woodcrest 3GHZ Processor

1 core

RAM 12GB 666MHZ

OS Snow Leopard OSX 10.6

.

[Paul Forwood]

Take the PRJ and show us. People would say that but they never posted actual comparisons.

I am rendering at the moment but I have no intention of running it through every version of A:M that I have.

 

While everything uses the CPU Particles introduce introduce other significant variables like RAM access speed and total available RAM. Those are not the variables I'm trying to look at IN THIS TEST.

 

I'll will accept that, so forget the additional variables and stick to your setup, but the version of A:M that you are using must be part of the datum. If you are unaware of the differences in versions of A:M and the impact on render speeds then I suggest that you test that first.

[jason1025]

Take the PRJ and show us. People would say that but they never posted actual comparisons.

I am rendering at the moment but I have no intention of running it through every version of A:M that I have.

 

While everything uses the CPU Particles introduce introduce other significant variables like RAM access speed and total available RAM. Those are not the variables I'm trying to look at IN THIS TEST.

 

I'll will accept that, so forget the additional variables and stick to your setup, but the version of A:M that you are using must be part of the datum. If you are unaware of the differences in versions of A:M and the impact on render speeds then I suggest that you test that first.

 

 

Hey Paul

 

I would love to see your data on the different versions of AM. I am kicking myself because I started using AM on Version 11 and would love to do an apples to apples compare against the different versions.

[Paul Forwood]

 

You see, you must set a standard base for bench tests.

-----------------------------------------------------------------

 

Please don't anyone confuse this example as a benchmark for versions of A:M. There are too many other variables that would have to be added to compare one version of A:M to another. There are many improvements in later versions of A:M that are not reflected in this test. This is simply to show that when collecting data, to compare render times on different machines, you must ensure that everyone is using the same setup, (software version and render settings).

 

-----------------------------------------------------------------

I forgot to add that these were all rendered on a single core.

Background activity was minimal: I had a QT movie open, (but not playing), and a live browser window onto a text and image website.

I mention this beacause I am surprised by the results others have shown.

[steve392]

I don't understand how its differant in each version ,what changes to make this happen

[Paul Forwood]

The code is modified between versions to work around problems that people have found and these changes can have ramifications that affect other parts of the program. Just tweaking the routines that handle how alpha is handled,for example, could add, or reduce, the time spent in that routine by a fraction of nanosecond but if that routine is called thousands of times for every frame you render you will see a change in the final render times. Fixing things doesn't necessarily make them faster but hopefully more stable and more useful.

[steve392]

Ah I gotcha ,thank's Paul, I am now edgucated

[jason1025]

 

You see, you must set a standard base for bench tests.

-----------------------------------------------------------------

 

Please don't anyone confuse this example as a benchmark for versions of A:M. There are too many other variables that would have to be added to compare one version of A:M to another. There are many improvements in later versions of A:M that are not reflected in this test. This is simply to show that when collecting data, to compare render times on different machines, you must ensure that everyone is using the same setup, (software version and render settings).

 

-----------------------------------------------------------------

I forgot to add that these were all rendered on a single core.

Background activity was minimal: I had a QT movie open, (but not playing), and a live browser window onto a text and image website.

I mention this because I am surprised by the results others have shown.

 

 

Surprised in what way?

The only thing that surprised me was how poorly the render works on mac, I am convinced the application of AM is running through some sort of translation instead of a truly mac coded app.

Are you surprised that your computer is 2nd to last in the race, just kidding I know its not a race.

[Tralfaz]

Did this test on my HP G60 laptop, from a clean reboot of the computer.

 

Report:

 

CPU Brand and model - AMD Athlon X2 QL-64

Actual CPU speed in GHz - 2.1 GHz

RAM - 3GB

OS - Windows XP Professional SP3

A:M - Version 14.c

 

First Test

Threads - 1

render time min:sec - 13:17

how many cores A:M is using - Task manager shows both cores running at 50%

 

Second Test

Threads - Auto (2)

render time min:sec - 11:25

how many cores A:M is using - Task manager shows both cores running between 75-90%

 

Usually, I just leave the Threads set to 1, as I seem to get better render times than having Threads set to Auto.

 

Al

[Paul Forwood]

Jason, I was surprised by the difference in these three processors:

 

Robcat2075:

 

19:59

AMD Athlon XP 3200

1.921 GHz

1 core

2 GB RAM

Windows 2000

----------------------------------------

 

Ludo_si:

 

19:00

AMD ATHLON 64 X2 DUAL

core processor 3800

2.01 GHZ, 2GO RAM

windows xp

---------------------------------------

 

Paul Forwood:

 

A:M14c: 11:40

A:M15g: 15:43

A:M15h: 15:11

 

AMD ATHLON 64 X2 DUAL (4200+)

(utilising one core)

1.79 GHz

3 GB RAM

XP Pro

---------------------------------------

I didn't think there would be so much difference between Ludo_si's machine and mine, in particular. That's all.

 

Almost time to look for a new machine but I can hang on for a little longer. I still get alot of use out of my old P4.

 

By the way, I'm no genius.

[wedgeeguy]

I guess I'll throw my hat into the ring ... or shall I say my teapot?

 

Pentium 2.4 Windows XP

A:M Version 13t (yeah I know I'm old "skool")

1.5 GB Ram

Only got 1 core

 

Time: 35:39 ... ouch!

 

Screenshot:

Actual image:

[robcat2075]

That mac time is better than before Intel macs came along and better than intel times were some years ago.

 

And this is a scene designed to take absurdly long to render on any machine.

 

Compare this to the 11Second club entry I just did. That took 2 seconds per frame to render in "Final". So Mac would take 4 seconds per frame.

[robcat2075]

Paul Forwood:

 

A:M14c: 11:40

A:M15g: 15:43

A:M15h: 15:11

 

AMD ATHLON 64 X2 DUAL (4200+)

(utilising one core)

1.79 GHz

3 GB RAM

XP Pro

 

i got

16:44 in v13t

 

vs.

 

19:59 in v15h

 

 

That's comparable to your difference between v14 and v15

 

 

So 15 does seem slower.

[NancyGormezano]

version of A:M (see Help>About A:M)

render time min:sec

CPU Brand and model

Actual CPU speed in GHz

how many cores A:M is using

RAM

OS

 

ver 15e CD

8:31

Dell 3GHz Intel Core2Duo

1 core? 50%?

3GB Ram

XP pro sp3

 

EDIT: with ver 15g CD

9:33

 

EDIT2: with ver 14c

8:18

Edited December 3, 2009 by NancyGormezano

[John Bigboote]

Thanks for putting this together, Rob. Wow, Nancy's Dell blew my Boxx out of the water! This is my work machine, I did not reboot before. I will next test my new machine at home.

 

ver 15g subscription

15:16

Boxx Dual Core AMD Opteron Processor 275 2.20 Ghz

1 core? 25%?

5GB Ram DDR2 I think

XP 64 sp2

[Rodney]

This can be a useful measurement.

 

If joining into the discussion please stay on topic and report your benchmarks as requested.

(No need to ask why your post was edited... you already know why)

 

Results:

A:M 15g subscription

17:40

Sony Vaio Core 2 Duo CPU T5800 @ 2.00GHz

1 core? %?

4GB Ram DDR2 I think

Vista 64 SP2

 

Nothing optimized here and I suspect better render times would result if I cleaned things up and optimized a little. I made no effort to shut down extra progroms or process and was connected to the internet. Will render again...

[NancyGormezano]

So 15 does seem slower.

On my system 15g is slower than 15e

[Darkwing]

im gonna give this a whirl soon because my mac could render particles and stuff like that roughly 6 times faster then the windows could

[Rodney]

Second Time Rendering...

 

Results: 15:15

50% on 1 core

 

I changed two things as far as I know:

1) I made sure the Channels window was hidden (supposedly it takes longer to render with it cracked open)

2) In the task Manager I Set Priority to High for Master.exe

 

I guess cutting over two minutes off render time is worth tweaking settings a little.

[robcat2075]

It occurs to me that since the render time variation between different versions of A:M appears consistent among the various CPUs that have reported results for more than one version, I can make a multiplier that factors that out and leave us with something approaching a CPU comparison only

 

 

for example if Nancy's 15e rendertime is 9 and 15g is 10

 

and my 15g is 17 and my 15h is 19

 

I can estimate what 15e would do on my CPU (15.3) and predict that 15h woudl do about 11.2 on Nancy's.

 

By this, I could include results from someone who only tests with 15e on their CPU

 

With what I have now i can convert times from

 

13t

14c

15e,f,g,h

 

This conversion will be more reliable with the PC data since we have more of that so far

 

 

But keep posting your results if have time.

 

Thanks!

[Shelton]

Correct me if I am wrong, but 15 was compiled differently? Could this be the reason it takes longer?

 

Steve

[TNT]

Jason, I was surprised by the difference in these three processors:

 

Robcat2075:

 

19:59

AMD Athlon XP 3200

1.921 GHz

1 core

2 GB RAM

Windows 2000

----------------------------------------

 

Ludo_si:

 

19:00

AMD ATHLON 64 X2 DUAL

core processor 3800

2.01 GHZ, 2GO RAM

windows xp

---------------------------------------

 

Paul Forwood:

 

A:M14c: 11:40

A:M15g: 15:43

A:M15h: 15:11

 

AMD ATHLON 64 X2 DUAL (4200+)

(utilising one core)

1.79 GHz

3 GB RAM

XP Pro

---------------------------------------

I didn't think there would be so much difference between Ludo_si's machine and mine, in particular. That's all.

 

Almost time to look for a new machine but I can hang on for a little longer. I still get alot of use out of my old P4.

 

By the way, I'm no genius.

 

Could the differences be due to bus speed between the CPU,memory, and page files?????

[robcat2075]

Could the differences be due to bus speed between the CPU,memory, and page files?????

 

I'm sure those are elements; the CPU has to talk to the RAM just to run the program at all, but to try to minimize that bottleneck this test uses no bitmap textures and renders to a small image.

 

But I suspect CPU power is the #1 factor in how fast A:M renders.

[robcat2075]

Correct me if I am wrong, but 15 was compiled differently? Could this be the reason it takes longer?

 

I recall Steffen is using some different libraries since v15f to enable some new features so that might be part of it.

[NancyGormezano]

for example if Nancy's 15e rendertime is 9 and 15g is 10

 

I just added test with 14c

 

to summarize:

 

14c - 8:18

 

15e - 8:31

 

15g - 9:33

 

Not that much difference between 14c and 15e on my system (render time), BUT 15e is a bazillionbilllllion times better, faster with hair, dynamics.

 

I think OS has a lot to do with it as well as cache, bus speeds, cpu, memory speed, amount. Vista, and maybe ? 64 bit OS seem to be dawgs with A:M.

 

I bought this system over 2 years ago.

[Paul Forwood]

BUT 15e is a bazillionbilllllion times better, faster with hair, dynamics.

Yes. Overall performance in later versions is much faster and stable than earlier versions, when you start adding in things like hair/particles, volumetrics, dynamics, etc., etc. One of the things that Steffen is intergrating into A:M code is OpenMP, so that might be affecting performance in recent updates, but it seems to me that introducing OpenMP points towards efforts to get A:M firing on all cylinders, or rather all cores.

[robcat2075]

Here's a first attempt at a chart with the raw data, I need to find something better to visualize this. Note the lack of labels.

 

 

if you go from front to back you have the same CPU running different versions of A:M. Relatively small differences

 

If you go from left to right you have different (unidentified!) CPUs running the same version of A:M. Fairly large variations.

 

I'll try to find a chartmaker that handles this better.

[3DArtZ]

11min, 4 seconds

am v14.0c

BOXX machine running winxp pro serv pack3

amd opteron 250 - 2.40Ghz(single core machine), 2gigs ram

[photoman]

27:42

AM 13l

Intel Core 2 Duo 2.44ghz

2gb Ram

Mac OSX 10.6

 

Ill test it on my windows partition tonight for comparison.

 

Photoman

[jason1025]

27:42

AM 13l

Intel Core 2 Duo 2.44ghz

2gb Ram

Mac OSX 10.6

 

Ill test it on my windows partition tonight for comparison.

 

Photoman

 

 

I am guessing the windows side will be around 9-11 minutes

[Phil]

version of A:M 15H

render time 6:38

CPU Brand and model Intel I7 975

Actual CPU speed in 3.33 GHz

how many cores A:M is using 1

RAM 3GB

OS. Windows 7

[jason1025]

version of A:M 15H

render time 6:38

CPU Brand and model Intel I7 975

Actual CPU speed in 3.33 GHz

how many cores A:M is using 1

RAM 3GB

OS. Windows 7

 

 

WOW and the winner is the i7 975. That is so fast. Can you run 4 instances at the same time and six instances at the same time after that. I am curious how well it keeps that record time when rendering multiple instances.

 

Am I correct that that is a 4 core system? How many cores come up in the task manager? does hyper threading create suit o extra cores?

[TNT]

Here are my stats to add to the mix

 

 

version of A:M - 14.0C

render time min:sec - 10:20

CPU Brand and model - AMD Phenom 9600 Quad-Core

Actual CPU Speed - 2.31 GHz

how many cores A:M is using - 1

RAM - 3GB

OS - XP Pro SP3

[robcat2075]

Second Time Rendering...

 

Results: 15:15

50% on 1 core

 

I changed two things as far as I know:

1) I made sure the Channels window was hidden (supposedly it takes longer to render with it cracked open)

2) In the task Manager I Set Priority to High for Master.exe

 

I guess cutting over two minutes off render time is worth tweaking settings a little.

 

Rodney, you are RIGHT!

 

I closed all A:M windows except the render window and my time improved from 19:21 to 17:10, an improvement of 12%!

 

Curiously, minimizing A:M during the render only dropped the time to 17:46

 

Setting priority to High dropped the time just a little bit to 16:56

 

Closing the other windows is easy enough to do and will probably eliminates some Graphics card overhead.

 

 

 

SO, anyone reading this... I'm going to CLOSE this thread and come up with a new, shorter, more consistent CPU test in another thread. Be on the lookout for it.

 

Thanks to everyone who participated so far!

 

Edit: Temporarily reopened to catch straggler's posts.

[jason1025]

What was your conclusion based on the data below? It looks to me that AM14c is a screamer when it comes to rendering.

 

 

 

 

Jason, I was surprised by the difference in these three processors:

 

Robcat2075:

 

19:59

AMD Athlon XP 3200

1.921 GHz

1 core

2 GB RAM

Windows 2000

----------------------------------------

 

Ludo_si:

 

19:00

AMD ATHLON 64 X2 DUAL

core processor 3800

2.01 GHZ, 2GO RAM

windows xp

---------------------------------------

 

Paul Forwood:

 

A:M14c: 11:40

A:M15g: 15:43

A:M15h: 15:11

 

AMD ATHLON 64 X2 DUAL (4200+)

(utilising one core)

1.79 GHz

3 GB RAM

XP Pro

---------------------------------------

I didn't think there would be so much difference between Ludo_si's machine and mine, in particular. That's all.

 

Almost time to look for a new machine but I can hang on for a little longer. I still get alot of use out of my old P4. smile.gif

 

By the way, I'm no genius.

[TheSpleen]

my stats not mentioned?

[robcat2075]

my stats not mentioned?

 

 

[robcat2075]

What was your conclusion based on the data below? It looks to me that AM14c is a screamer when it comes to rendering.

 

 

 

It may be faster , but the problem is we only have two results by users that also gave results for v15. One g and one h.

 

Paul's was 24% faster but Nancy's was only 14% faster.

 

However, Paul's h was faster than his g but my g was faster than my h. Explain that.

 

It's possible that different versions do differently on different CPUs. It's possible 14 might not do any better on some and it looks like most people have moved to 15 for the features and stability.

 

(the fastest render of all was Serg with 15h, but he didn't do a run on 14 for comparison.)

 

But I want to come up with a benchmark that takes less time to run so more people will be likely to contribute data. My goal is to compare CPUs.

 

BTW, to qualify as "screamer" for me it has to go > 50% faster.

[TheSpleen]

my stats not mentioned?

 

 

my little pink thing is soooo small

[robcat2075]

It looks to me that AM14c is a screamer when it comes to rendering.

 

 

One likely reason a later version of the render wouldn't be as fast is that it has to accommodate rendering features that have been added since the earlier version. It has more things to test for and handle properly and has to do it on every pixel.

 

There are switches in the renderer you can set to tell it to not bother with certain time-consuming stuff. For example you can disable Shadows or Hair or AO. You can avoid refraction calculations by only setting IOR to 1.

 

But you couldn't have a switch for every possible rendering element. There are already more render parameters than some users can manage.

 

It's also possible that the earlier version failed to properly handle some combination of existing features (some shadows and semi transparency rings a bell here) that were technically legal but so rare that it wasn't noticed by most users or perhaps not rare but noticed only by a few discerning users. So the earlier version may have finished its work faster but not always correctly.

 

Once the problem is noticed the code needs to be expanded to always test for that particular situation and that adds to the render time.

 

My test case PRJ probably doesn't invoke any of those problem situations since everbody's renders all look identical. But there's no way for A:M to know those situations aren't there and that it didn't need to test for them until it's finished the last pixel. And since this is animation, that situation that wasn't in the last frame might be in the next frame. Back to square one on testing.