CPU Render Benchmark

[robcat2075]

my little pink thing is soooo small

 

Smaller is better in that universe.

[Jaff]

Hi. I did the test and got this:

 

 

 

Version 15 F

14:54 min

AMD Athlon XP 4200

1.921 GHz

1 core

4 GB RAM

Windows XP 64 bits

 

I little slow i think , have to make some tests....

 

Saludos.

[jason1025]

Robert

 

I am really happy you did this. This data has helped me out in making a decision for my small render farm. I dont think you need to change a thing. It would be great to keep this topic open so we can continue to see results. In my opinion this is a definitive official benchmark test for a system and AM. It may not be perfect but now that I have seen this data, I can run this test on any computer and some up its performance in regards to Am and other systems. I can also compare new versions of AM against old ones.

 

I think in the years to come as technology improves and the code gets cleaner we will see this number get into 1 minute or less. To make a less intensive benchmark may not as clearly show the drastic range we have seen in results such as the difference between your results and the i7 3.3GHZ results.

 

Bottom line

Great job

[NancyGormezano]

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

 

No - it was Phil, wasn't it?

 

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

 

That's very promising, encouraging.

[steve392]

AMD Athlon ™64x2 Dual Core 5200+2.71GHz

3.68 GBof RAM

xp home

cores 1

 

v 15 g

0.12.30

[serg2]

..... but he didn't do a run on 14 for comparison.)

 

Report:

Version 14.0C

render time min:sec 07:12

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

A:M is using: single core from (4core (x2 HyperThreading is ON))

 

 

Report:

Version 15.E

render time min:sec 07:12

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

A:M is using: single core from (4core (x2 HyperThreading is ON))

 

[serg2]

MultiCore simultaneous Test Render

 

AM Version 15.H

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

(4core (x2 HyperThreading is ON))

 

8 copies АМ are started simultaneously

Each copy АМ one core is appointed (assigned from Affinity - pic mct)

 

 

Result of the simultaneous miscalculation on 8 kernels

1- 13:02

2- 13:04

3- 13:05

4- 13:04

5- 13:06

6- 13:07

7- 12:58

8- 13:25

 

pic mct2:

 

[Xtaz]

freak...... nice pic serg...

[KenH]

15h

8:52

Intel 8400

3 GHz

2 core

8 GB DDR2 RAM

Windows Vista Premium 64 bit

 

I also closed all AM windows. If this does save so much time, maybe it could be an automatic thing when anything is rendered. Not everyone will know that trick.

[jason1025]

15h

8:52

Intel 8400

3 GHz

2 core

8 GB DDR2 RAM

Windows Vista Premium 64 bit

 

I also closed all AM windows. If this does save so much time, maybe it could be an automatic thing when anything is rendered. Not everyone will know that trick.

 

 

can you make a feature request?

[John Bigboote]

15h

8:52

Intel 8400

3 GHz

2 core

8 GB DDR2 RAM

Windows Vista Premium 64 bit

 

I also closed all AM windows. If this does save so much time, maybe it could be an automatic thing when anything is rendered. Not everyone will know that trick.

 

 

can you make a feature request?

 

 

Not for me...I multitask while A:M is rendering. I don't care if it slows me down...I'm rendering right now as I surf the forum.

[jason1025]

15h

8:52

Intel 8400

3 GHz

2 core

8 GB DDR2 RAM

Windows Vista Premium 64 bit

 

I also closed all AM windows. If this does save so much time, maybe it could be an automatic thing when anything is rendered. Not everyone will know that trick.

 

 

can you make a feature request?

 

 

 

Not for me...I multitask while A:M is rendering. I don't care if it slows me down...I'm rendering right now as I surf the forum.

 

 

I think hes referring to closing all the pallet windows in AM before rendering.

[jason1025]

Closing all windows/pallets in AM before rendering as well as setting AM as a high priority in the task manager decreased my render time from 10:03-9:58.

 

Every little bit helps

 

Any other suggestions

 

[Rodney]

Closing all windows/

 

Just in case this is an important distinction...

 

We cannot close the Channels window with the Project Workspace (PWS) open.

The best we can do is Hide the Window - or close the PWS.

 

In order to hide it, move your mouse over the side pane until you see the mouse pointer change.

Then drag the Channels window until its out of sight.

 

At a guess I'd say the reason for the additional render time is that A:M has to take the time to update the User Interface.

Hiding that area seems to let A:M know updating the User's view is unnecessary.

 

Note: I've created keyframes only for the purpose of showing the Keyframes visible in the Channels window.

I'm calling this space the Channels window in order to distinguish it from the Timeline which can be closed.

If you have the PWS open you cannot close the Channels window. You can only hide it from view.

[John Bigboote]

Thanks for taking the time to point that out, Rodney. I did not know that trick.

[Fuchur]

This is my Benchmark:

- Version: 15h

- Renderingtime: 7:56

- CPU: AMD Phenom II 955

- Corespeed: 3,2 GHz (per core)

- Cores used: 1 core

- Cores Total: 4 cores

- RAM : 4 GB DDR3 RAM (1066 MHz), 2GB used

- OS: Windows XP (32 bit(

 

For the people saying, A:M is too slow for such a scene:

This project contains settings which wouldn't have to be used for the image it produces.

A reflection-level of 8 (for instance) is just much to high and will increase the renderingtimes significantly.

 

It is however good to test with such a setting, because it will increase the renderingtime and will show the differences between processors in a better way.

 

See you

*Fuchur*

[zeus1972]

I/you/they have not succeeded in understanding a thing.

or AM installed on windows 7 64 bit intel 4 cores 8 g ram

Does AM succeed in exploiting more than a core?

because when I send him/it in rendering the cpu it goes only to 25%.

I remembered me that in a version preceding wax the possibility to select the number of the cpu, but in the 15g I don't find this option anymore.

thanks

[robcat2075]

Does AM succeed in exploiting more than a core?

 

Short story... the renderer was rewritten in V14 to do MP but it only got faster results in certain situations and slower results in many others. Steffen is reintroducing MP in v15 for certain things that can benefit but it's not in the renderer yet.

 

You CAN use your MP computer with A:M and get great results by running a separate instance of A:M for each core and setting each to render a separate section of the frames you want.

 

This will get you almost 100 percent scaling with each new core, whereas multithreaded apps typicallyy do far worse.

[jason1025]

AM 15.H

 

Mac Book Pro intel core Duo

2.8GHZ

2 cores

4GB 1066 MHZ Ram

9.08 seconds

[robcat2075]

AM 15.H

 

Mac Book Pro intel core Duo

2.8GHZ

2 cores

4GB 1066 MHZ Ram

9.08 seconds

 

that beats me!

[jason1025]

I was surprised that it beat my 8core mac pro tower 3.0GHZ which rendered at 9:58

[robcat2075]

I was surprised that it beat my 8core mac pro tower 3.0GHZ which rendered at 9:58

 

Was the PWS closed on both? That made a big difference on mine.

[jason1025]

I was surprised that it beat my 8core mac pro tower 3.0GHZ which rendered at 9:58

 

Was the PWS closed on both? That made a big difference on mine.

 

 

Yes its was closed, along with all the other windows, including the model window that defaults to being open. Oddly on the mac pro tower closing all the windows only dropped the time by 6 seconds or so.

[jason1025]

Anyone try version 15.i to compare against H in speed?

[Fuchur]

Anyone try version 15.i to compare against H in speed?

 

Did the rendering. The same result with 15i -> 7:56 min

 

*Fuchur*

[pixelplucker]

9:34 on this system with 15.0i.

I didn't test it compared to previous versions but gained 2 seconds after a clean reboot.

 

xp pro 32 bit with sp3, quad xeon 2.66, 4gb interleaved memory. 34 processes running including AM.

[serg2]

Report:

Version 15.I

render time min:sec 07:01

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

[serg2]

Report:

Version 15.I

render time min:sec 07:01

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

[steve392]

Ok Ok no need to go on and on Serg

[serg2]

I do not know how to remove the duplicate post

[steve392]

Just joking Serg ,niether do I

[pixelplucker]

I did another quick test and set AM's priority to real time from the task manager and knocked 3 sec off the render time.

 

I saw that there are some utilities out there that do multi core optimization and clustering, wondering if one of those would speed up the render times since they change the update speed of prefetch data and dump unuseded computations opening the cpu usage for the task at hand. The utilities seem to be used for the real time visualization, workstation and servers but I can't seem to find any sales of where to purchase them.

 

http://multicore-optimizer.com/exludusflash.html

Same software used here http://www.jrti.com/products/software/exludus_index.html

[3DArtZ]

Pixelplucker.... I got a free one but have yet to test it. I will report back when I do, on a quad core boxx machine.

the free one is called: MultiCoreOptimizer

 

I found it with a google search of a similar name dont remember what I used exactly though.

 

Mike Fitz

[pixelplucker]

Cool do you have a link? Does it support the itanium chips?

Oh yeah, does it actually work?

 

Part of my slightly slower rendering is I have a 64bit system running in 32bit mode, even though AM is 32 bit the overall system isn't running at peak until I can muster up an extra grand or so to upgrade all my software and even then I will loose a couple of programs that have less than slow development and rely on 32 print drivers.

[3DArtZ]

I have not tested it yet...

I found it here:

http://www.ddlhere.com/pc-applications/814...zer-v1-0-a.html

[pixelplucker]

Thanks, when I googled for it, my AV warned me on most of the links, little bit scary. I'll check this out.

[pixelplucker]

I don't think those are legitimate copies of the Multicore Optimiser. They show up as torents or cracked versions. Most likely they are bootlegs from the one made by Excludus http://www.jrti.com/products/software/exludus_index.html

 

Thanks anyways.

[3DArtZ]

well, Im still going to give it a try on one of my quad core's. will report back... if you dont hear from in a week about this, I've been hacked and virused!!! lol

[pixelplucker]

Might be a good time to make a restore point on your system hehe.

 

I emailed one of the legitimate companies to see if it would help on render times, haven't heard back yet but probably will after the holidays.

[3DArtZ]

btw, I tested the optimizer on my quad core and it did absolutley nothing in terms of speed. I tried every combination of threads and affinity.

just fyi.

[pixelplucker]

Technically what the optimizer does is it dumps unused cpu calls and fiddles with the time it stores used calls in the cache. I'm guessing it probably works better for server type work rather than 3d work where moving files around rather than number crunching is more of a priority. There is some utilities on Intel's site that they probably built it from.

 

Guess we could pressure Yoda to make AM multithreaded and see if his head explodes with all the work he's been buried with hehe.

 

I still wouldn't mind a separate application that could just be the renderer and even if I had to purchase it as an extra it would be cool. The built in render is nice but would be handy to have something I could toss on another machine that is multi threaded and just grunts out the finish product.

[jason1025]

I was thinking of purchasing Net Render. I had a bunch of questions about the product and Jason S said that aside from open MP support it cant use multiple procs/cores. This got me thinking, Does open MP work. After testing the answer is no or at least it didn't work for me in regards to decreasing render times. You can see the results below.

 

Can some one concur?

 

 

[robcat2075]

Can some one concur?

 

Based on what Steffen has said in the release notes, OpenMP is only implemented for certain parts the modeler. It speeds up the "finding patches" maneuver, which is noticeably slow on very,very large models. This is the pause when you get when you edit CPs on a model of say, 10,000 patches or more.

 

So far, OpenMP hasn't lent itself to speeding up Final rendering routines.

 

Your best bet for making use of multiple cores is still running multiple instances. It makes almost 100% use of each additional core which is far superior to any single instance/multithread scheme in any software.

 

If you have multi-core machines NetRender probably isn't your best choice to make the most of them.

[jason1025]

thanks

[Fuchur]

Just to mention it: Preview-Renderings are faster.

-> green Buttons.

 

But Final Renderings aren't. For the non-programmer that sounds odd, but

this is because of many final-rendering-operations which are not programmed to be multithreadable.

 

NetRenderer is for several computers, not for several cores!

That means, you can get much out of it with 100 single-core-pcs but not with a 4-core-system.

For that you should open instances, etc.

 

*Fuchur*

[jason1025]

Just to mention it: Preview-Renderings are faster.

-> green Buttons.

 

But Final Renderings aren't. For the non-programmer that sounds odd, but

this is because of many final-rendering-operations which are not programmed to be multithreadable.

 

NetRenderer is for several computers, not for several cores!

That means, you can get much out of it with 100 single-core-pcs but not with a 4-core-system.

For that you should open instances, etc.

 

*Fuchur*

 

 

Unfortunately I dont fit either of those scenarios. I have 6 computers that have a combined 30cores. Sounds like Net render needs to be updated. I know they wont because its not economical but at my last staff job we had a Net render that looked exactly like Hashes, it worked the same way. It easily took advantage of all cores as if each core was its own computer. We gave it access to two 8 core machines. It was proprietary software that that could plug into almost any render er. In our case our net render was utilizing Red's trans coding software. I think it was called Red alert. Its only job was to transcode Red Raw 4k Video footage and turn it into a Targa sequences.

 

 

Its spooky when I see the gui of net render because it looks just like ours. There must be some sort of universal component there.

[robcat2075]

I'm sure it can be done. Submit it as a feature request, if you haven't already.

 

In the meantime I can imagine someone making a simple text editing app that, given a Chor set to render all the frames, would spit out X copies with each set to a different frame range. Then you'd just have to load each copy on to a different core and set it rendering without having to dig into each one to reset the frame ranges.

 

Lessee...

 

-this app would search thru the .CHO file for the defined frame range,

-extract the start_frame and end_frame numbers,

-extract the number_of_cores to be used from a string embedded in the PRJ title

 

then, in a loop, write number_of_cores copies with

 

Copy_0.cho set to render start_frame to end_frame, step=number_of_cores

Copy_1.cho set to render start_frame+1 to end_frame, step=number_of_cores

Copy_2.cho set to render start_frame+2 to end_frame, step=number_of_cores

 

and so on until number_of_cores .cho files had been written out.

 

(alternatively each copy could be set to render contiguous ranges if that was preferred.)

 

that would be a simple program to write, there's probably a text editor out there with sufficient macro ability to do it.

[jason1025]

I'm sure it can be done. Submit it as a feature request, if you haven't already.

 

In the meantime I can imagine someone making a simple text editing app that, given a Chor set to render all the frames, would spit out X copies with each set to a different frame range. Then you'd just have to load each copy on to a different core and set it rendering without having to dig into each one to reset the frame ranges.

 

Lessee...

 

-this app would search thru the .CHO file for the defined frame range,

-extract the start_frame and end_frame numbers,

-extract the number_of_cores to be used from a string embedded in the PRJ title

 

then, in a loop, write number_of_cores copies with

 

Copy_0.cho set to render start_frame to end_frame, step=number_of_cores

Copy_1.cho set to render start_frame+1 to end_frame, step=number_of_cores

Copy_2.cho set to render start_frame+2 to end_frame, step=number_of_cores

 

and so on until number_of_cores .cho files had been written out.

 

(alternatively each copy could be set to render contiguous ranges if that was preferred.)

 

that would be a simple program to write, there's probably a text editor out there with sufficient macro ability to do it.

 

 

 

Is this for net render or for cd version with multiple instances?

[robcat2075]

Is this for net render or for cd version with multiple instances?

 

multiple instances.

 

I realize this is not as convenient as if Netrender transparently handled multiple cores.

 

I just see it as a shortcut to setting up multiple instance renders.

[jason1025]

Is this for net render or for cd version with multiple instances?

 

multiple instances.

 

I realize this is not as convenient as if Netrender transparently handled multiple cores.

 

I just see it as a shortcut to setting up multiple instance renders.

 

 

Can you make a feature request? It may have more impact coming from you.