Happy New Year everyone,

Been MIA for a while now.

Some of you might remember, that I was messing around with the idea of a Bryce Render Farm, you can search in these forums for full details.

I had something like 6-8 PC's which was a mixture of Pentium 2's and Pentium 3's.

The results were not impressive a single Dual Core PC could outperform the entire farm.

forward to 2010 and I have aquired some hand me down P4's.

5 x Pentium 4's @ 3.0GHZ
2 x Dual core laptops  1.8ghz each.

they need some work to be functional, in the next couple of months I'll have them up and running on highly optimized windows xp machines configured for speed, with very small memory footprints.

I'm curious to see what kind of results i get, and I'll share those results here.

Stay Tuned... :)

OKay....   I'm keeping my eye out for your results...

I will look foreward to hear of your experinces with your setup of a renderfarm

I've tried this: I had 2 single core, 2 quadcore and one Duo-core running as a renderfarm - they reduced the rendertime with a factor 16 on a render that took 24 hours on a duo core. All machines except the 2 single-core, which were labtops, had a frequenzy in the range 2000-2600 mHz. I've discovered that using IBL can make the renderfarm useless, because suddendly some of the machines recieve the data, but stops rendreing, even with a restrat of lightning on the respective computers, i haven't found a workaround for that yet. Furthermore i've experinced a few times that the resulting bmp from the renderfarm doesn't look like the bmp from a standalone render. One time the colours just wasn't there on the renderfarm bmp.

Regards
Arne Fjord

Wow, those are interesting results.  Which version of Bryce and Bryce Lightning were you trying when you got these results?

I'm going to try with Bryce 6.1.

One other thing to note that has changed in the recent versions, the system that you run Bryce on in a renderfarm situation it does not do any rendering it simply acts as the render manager.

I think this changed with Bryce 5.5.

With Bryce 5.0 you could render with all the machines in the render farm.

So with the new version you would want to run Bryce on the slowest machine.

Well, i'm using breyce 6.1 as well, and you can start a local lightning on the system that runs as render manage as well that will participate in the rendering. Yesterday i furthermore discovered that when you want to render more images in succestion you have to do it in the same instance of the render manager, else bryce gets confused and you risk that it "mixes  " two render jobs - looks strange by the way :-)

Also - the biggest bottleneck in Bryce Lightning is the speed of the network itself. Gigabit networks would probably speed things up a bit, but most people still have 100megabit switches and older machines will have NICs onboard that run at 100 mbs.

Quote - Also - the biggest bottleneck in Bryce Lightning is the speed of the network itself. Gigabit networks would probably speed things up a bit, but most people still have 100megabit switches and older machines will have NICs onboard that run at 100 mbs.

Oh Yeah.. I forgot to mention.. all machines have Gigabit cards... I just need to purchase a Gigabit Switch.

Right now I'm looking on eBay to source one.

 Good deal - yeah you should be able to find a good priced one on ebay. Good luck.

I have 4 Desktops ready for testing.

1 Machine needs a new processor but for now I'm going to use it to manage the actual render jobs.

3 Machines have identical specs:
Intel Pentium 4 CPU 3.00GHZ / 2.99GHZ / 512MB (Yes I know ram is low)

I built these machines with the focus on processing speed and are running on very highly optimized versions of XP with a very small foot print.

All unnecessary programs / system files / processes have been stripped out.

I have attached a screenshot to illustrate how highly optimized these PC's are.  

You will notice that only 15 Processes are running at anytime, and only 82mb of RAM is in use, if you want a comparison, if you are a Windows User, do a CTRL-ALT-Delete and take a look at the task manager.  You will probably have 30 Processes running if not more, and probably using up 512mb or more.

So I need to buy a few things off ebay, Processor more RAM and 1Gigabit switch.

In the meantime I'd like to start testing.

What do you guys recommend? Should I just jump and install the latest Bryce 7 Beta and test with that?  6.3? 6.1? 5.5? 5.0?

I would go for 6.1 - i've had some troubles getting 7.0 beta to see the renderfarm - however bryce 6.1 can use the 7.0 beta lightning program ...strange

Quote - In October 2006, DAZ released Bryce 6.0 and has released an update (6.1), this includes a Mac Intel compatible update. New features include animation import, support for dual-processor systems as well as hyper-threading, random replicate tool, advanced terrain editing, HDRI support and other tweaks. The interface remained largely the same, but with a green tint to it, and different buttons in the create palette.

Ok so I've been looking at what else I can do to increase performance and noticed this blurb on Wiki and what stood out for me is that Bryce now supports Hyper-Threading.

The PC's that I have are Pentium 4 HT's, I wonder if Bryce Lightning does indeed support Hyper Threading.  If it does I can enable and hopefully get more performance per PC.

I'm not sure about hyperthreading, unless it's the same as ability to use multiple cores! - I have 3 machines with quadcore, which bryce is able to use. The way to do this is setting priority to high which have a large impact on the render time on multi-core cpu's. However ...you can not set priority to high on bryce lighning clients, and i don't think taht setting priority to high on the machine that is render manager has any impact on the clients

EDIT: - By the way - i don't think taht net-speed is an issue in renderfarms, if you use tile optimation it's only small 8realtive) chunks of data that is sent between the manager and the clients, both when delivering the task, and in sending the finished render-tile

EDIT EDIT Reading through my post...of cause the priority setting must refer to MT - but as far a i can see bryce  also take advantage of multi-core processors

Best Regards
Arne

Ok... I'm getting ready to try my first Render.

For now it will be 1 Host/Manager  and 3 x P4's.

Hyper Threading is disabled.

I'll run some test and than enable Hyper Threading and see if we see any difference.

Hyper Threading is almost like having a Virtual Dual Core but not really :)

For the initial test I download and rendered the file from this site.

http://www.horo.ch/raytracing/tests/rendertest.html

I rendered the same file three times and I got 3 Minutes and 49 Seconds each time, I did not enable Hyper Threading and didnt touch the Bryce Lightning Priority settings.

Next few tests, I will do some seperate tests of  Playing around with Priority settings as well as enabling Hyper Threading.

Since Hyper Threading is enabled in the BIOS it will be a bit of a hastle since all 3 systems are being controlled using remote desktop, I would have to connect Keyboard, Mouse, Monitor to enable hyper threading on each one.

Ok So I turned on Hyper-Threading and although its nothing to write home about, it did shave off 16 seconds.  Yes 16 Seconds :)

Ok here are more results, all tests where done using network render with Bryce Lightning, Hyper-Threading and Priority set to HIGH.
Times are in MM:SS

Pentium 4 @ 3.0Ghz Hyper-Threading with 512MB of RAM.  Render Time 10:25 (Windows XP)
Pentium D @ 3.4Ghz Hyper-Threading with 3GB of RAM Render Time: 9:28  (Windows 7)

This is very interesting 57 Seconds difference between the two now is this a difference because of Processor Type? or the Extra RAM? Windows XP vs Windows 7? Whatever it is, difference is not by much considering one is more beefier system than the other.

1 x CPU = 10:25
2 x CPU = 05:17
3 x CPU = 03:34
4 x CPU = 02:38 <-- This is with the Pentium D Machine added to the mix.

These tests probably dont mean much when you start dealing with different files which some could be more complex than others.

It would be interesting to see if something that takes 1 day to render could be reduced down to 2-3 hours with this setup.

I plan to add Two more PC's dedicated to rendering, I also have 3 Laptops with Dual Core capabilities to add into the mix.

A small update.

I added a Core 2 Duo Laptop with 2GB of RAM and removed one of the Pentium 4 Machines and the Render time increased to 02:49, an extra 11 Seconds.

Will need to do more tests to figure out what is going on.

Ok out of curiosity I turned off "Tile Optimization"  and got 02:29.

It would be interesting to know what is going on between Bryce and Bryce Lightning 6.3 to get these strange results.

Ok so I've added a 2nd Laptop into the Mix.  For a total of 6 systems and the Render time dropped down to 1:38  a savings of 51 Seconds.

This is very interesting to me and without knowing the actuall internal logic, It almost looks like that Even number of PC's render fast than odd numbered.

So for a re-cap the results are as follows.

1 x CPU = 10:25 Pentium 4 HT 3Ghz
2 x CPU = 05:17 Pentium 4 HT 3Ghz
3 x CPU = 03:34 Pentium 4 HT 3Ghz
4 x CPU = 02:38 Pentium D HT 3Ghz.
5 x CPU = 02:49 Core 2 Duo T7500 @ 2.20 Ghz
6 x CPU = 01:38 Pentium Dual CPU T2370 @ 1.73 Ghz.

I'm comparing my results to the test file I got from  here: http://www.horo.ch/raytracing/tests/rendertest.html

and you can see that a Single Core i5 on a Mac can render the same scene in 41 Seconds, while a 2 x Xeon QuadCore for a total of 8 cores pumps it out in 46 Seconds.

So what do you guys think?  Now these tests are being done on a Single Frame.

I should do an animation and see what kind of results I get.. I wonder Bryce Lightning would behave different.
 

Ok so I've been doing more tests and here are some results for you to chew on.

In previous tests results you could see that on a Pentium 4 Hyper Threading machine running at 3Ghz with 512MB of ram with Bryce 6.3 Host/6.3 Lightning it would take 10:25 To render the test scene.

I did the same test but switched Bryce Lightning 6.3 with 7.0.0.151 and end result is amazing, the render time went from 10:25 down to 7:06.

Another thing that I tired was to upgrade my Core 2 Duo T7500 @ 2.20 Ghz Laptop from Windows XP to Windows 7 Ultimate 64-Bit Version.

With Bryce 6.3 Host/6.3 Lightning I got 07:11 when I switched out Lightning with 7.0.0.151 render time went down to 04:12

So I decided to re-do my earlier test using all 6 PC's but using Lightning 7.0.0.151 and my total render time was reduced from 01:38 down to 01:04.

So now I'm curious what render times I would see if I upgraded all systems to Windows 7 64-Bit Edition.

I imagine I must his some kind of speed limitation because of my 100Mbit network infrastructure.

PS: I use Prio Watch a neat tool that remembers my Affinity and Priority settings so I dont have set them everytime.

Prio - Priority Saver (FREE)
http://www.prnwatch.com/download.html