Hints for optimizing render farm nodes?

I’ve got a render farmlet running with lokirender set up to handle my Blender renders, and it’s quite a boon, but some of the results loki spits out seems odd, specifically the render times for the various machines.

For basically the same frame from my Kata animation:

Main Master/grunt: render time 3m 40s
Athlon 64x2 5600+, dual core, 2.8GHz
Windoze Vista Home Premuium, Media CenterEdition
Other apps running at the same time, including Blender 2.49b

Grunt #1: render time 9m 54s
Pentium 4, dual core, 3.2 GHz
2.86Gb RAM
Win XP Home SP2
Only loki grunt running

Grunt #2: render time 5m 56s
Athlon 64, single core 2.19GHz
960Mb RAM
Win XP Media Center Edition
Only loki grunt running

I’m wondering why Grunt #2 with 1 core, a slower CPU and much less RAM is puttin’ a bad case of whupass on Grunt #1, like churning out frames in nearly 1/2 the time?

I generally don’t mess about with system optimizations if I can help it but I’d really like to get Grunt #1 to perform better. Any suggestions about where to start?

have you checked the task manager while it’s running, to see if anything else was doing something? (in the background, of course)

edit: or could it be an issue of bandwidth/bus speeds?

On Grunt #1 there’s one other app running with a small RAM footprint, I will uninstall that later (I’m running a series of frames now), but the Blender instance loki invokes is by far the heavy hitter of the processes (750Mb), of which there are of course many, this being Windoze. I think many of the processes are start-up 'ware I can remove, but nothing seems to be taxing the system unduly, no virtual RAM (disc-writes) is being used, and in between frames, the CPU usage often drops to less than 20% for both cores, while it’s at 100% when rendering.

loki also reports on memory usage, though it’s bit obscure (the numbers change a lot but these are typical):
Master/grunt – 1.42 / 2.0
Grunt #10.32 / 2.0
Grunt #20.85 / 0.94

As ratios, these numbers show that Grunt #1 isn’t utilizing near the percentage of RAM for rendering that the other two machines are, though Task Manager seems to show different.

well, if you go to the control panel, then under (i think) administrative tools, the services one will allow you to weed out a bunch of junk (like, for example, disabling google updaters and itunes launchers and adobe speedlaunchers, you get the idea. you can kill services, you can disable them completely, and so forth.
do exercise caution, you could accidentally disable something important. I would recommend you do some googling to find out more about each service before you disable it.

still, i’m not sure it all adds up.
is there a large difference in how long each copy of windows has been installed? (eg, is one 6 months old, and the other 4 years)

Not sure about that, since the machines were given to me, and I don’t know exactly when they were purchased, but all within the last three years, I’d say. The Vista machine is obviously newer, but between the two XP machines, I’d say there’s no more than two years difference. Both versions of XP are labeled Version 2002, Service Pack 2, the only diff between them is that Grunt 2 is a Media Center Edition (meaning more bells and whistles, I guess), but that’s the faster of the two, and probably the older as well.

EDIT-- Some observations
– Grunt #2 (1 core, less RAM, slower CPU) is running more processes, threads and handles than Grunt #1 but rendering faster.
– The ratios shown above are because of available physical memory – each file takes approximately the same amount of RAM to process, but on Grunt #2, that’s a higher percentage of the total RAM available.
– Is there that much difference in CPUs that a single-core Athlon with a slower clock can outperform the dual-core Pentium? I know hardware optimizations can be significant, but still…

hmm, if I remember correctly, right click on “My Computer” and open properties, I think it will tell you there when windows was installed.
in fact, if you wouldn’t mind posting a screenshot of that on each of the two grunts, that would have some more useful information.

are they both running the same version of blender?

I knew pentiums were lousy, but I didn’t think they were that lousy…

Nope, no dates reported there, or anywhere else I could find, just the general system description as I’ve posted above, plus the serial #s of the Windows installation (which I don’t want to publish).

After I get done rendering up the current batch of frames I’ll spend some time cleaning house, see if that helps some. It’d be good if it does because my next sequence has frame rendering times approaching an hour, on my fastest machine, and it’s 250 frame long.

Thanks for the feedback!

been a little while since I’ve really done much with windows…

if you aren’t using an optimized version of blender, I would suggest that you do, just sse1 and sse2 can speed things up about 30% at least. (that’s probably all that those processors are capable of)
good luck digging up one of 2.49, though…
edit: this might work (Google FTW :))

Yeah, that’s one of the drawbacks of Blender’s development cycles, that fully functional versions get pretty much ignored, left behind and treated as “legacy” 'ware while the newer version is still in alpha. I can’t move up to 2.5 with Kata because the API isn’t yet finalized and even so there are too many changes to revise all my pydrivers script code.

EDIT: Many thanks, spacetug, I’ll give that version a shot! I hadn’t time yet to do a search so your effort’s much appreciated.

I gotta say, I think you’ve covered all the bases that I can think of. Not to sound like an Intel guy (I’ve got 5 intel cpus and 1 amd around my desk) but I can’t believe that is the deciding factor. I think Intel and AMD are pretty well matched. Seems to me the last thing I read about AMD’s latest processor was that it wasn’t as good as Intel’s, but the prices of the processors were at the point where you could build dual AMD cpu machine for the same money as a single cpu intel.

One thing that could be messing with the performance is the memory - specifically speeds and type. Dual channel vs Single channel and FSB speeds. Processor cache size could factor in as well. Even considering all those things still wouldn’t explain the difference in speeds.

I’m stumped, but I would be interested in trying the render on my 2.8 ghz/2g ram dual core intel system to see how it compares. Care to post the .blend somewhere and pm me with it?

I dunno if using benchmarking software would give you any clues…


Memory is something I’ll not be changing so if it’s a significant factor I’ll just live with it. Thanks for the test-render offer but I’m gonna pass on it, for now at least.

One definite improvement – the optimized build of 2.49b that spacetug linked to seem to be improving render times by 20-30% (as advertised :smiley: – thanks 'tug and metandrocto – love the theme, btw!), and the machine apparently most affected is Grunt #1, so now it’s much closer to Grunt #2’s performance, though still slower. This is after cleaning out the dusty corners of #1’s innards :wink:

Probably just chalk this one up to CPU VUDU, unless someone can offer more suggestions for optimizations on Grunt #1.

Hi chipmasque, i would like to advise to run the render benchmark on all systems,
http://www.eofw.org/bench/index.php?sortby=2 (I think you know it)
and than look to the list to compare your system with others.
Pentium 4 is a very old architecture in relation to amd64 or core2 or i7.
Try a linux live cd with the benchmark file, linux renders a little faster.:cool:
A lightweight distribution is xubuntu:
It is special for older/smaller systems and go easy on resources.

Cheers mib

Thanks for the suggestion, mib, I snagged the xubuntu image and the necessary Java and Blender distribs, but unfortunately Loki Render doesn’t play well with the other kiddies in the Linux sandbox :frowning: For some reason the Grunt running on the Linux machine never stops fetching the files and never gets around to doing any rendering. I’ve posted on the Loki Render forum, hopefully there’ll be a fix so I can continue testing the Linux option, but for now, I’m back in the 'doze (Win-doze, that iz :wink: )

Hm, i got it running here with opensuse (doesn’t matter what distri i think ).
Just for interest, did you compare the benchmark file with windows/linux ?
If no significant difference you could stay with your dose (german) happily.
Except the other 1000 reasons to change. :slight_smile:
But this is something completely different.

Cheers mib

Haven’t done the benchmark file yet, too busy getting renderings and editing done. I’ll be breaking from that farily soon so I’ll probably try the test file then.

I’m not at all familiar with the Linux world except by anecdote and reputation, so it’ll be interesting trying it out. Xubuntu was interesting, though somewhat unfamiliar, nor did I have much time to explore it. I need to get a bunch of frames rendered before I can delve back into it. Having the Live CD option was very useful, btw, easy to get a taste without having to eat the whole meal :wink: