PDA

View Full Version : G5 slower than my old PC..



Andres_cg
05-21-2005, 01:50 PM
Hey guys,



This is not a flame war or such, but...



I just got my duo dongle and ran some identacle renders on my Dual 1.8 G5 2gig ram, against my 2 year old 1.7 AMD 1.5gig ram machine.



I couldn't believe it, my PC was twice as fast as my G5. What's going on? I have the G5 set to 4 threads rendering. Is there some switch that need to flipped or somehting?



Thanks for any advice on how to get the G5 up to speed.

toby
05-21-2005, 02:43 PM
Can you tell us what you're rendering? Post a scene would be even better, then we can test it for speed. Check Blanos.com (http://blanos.com/benchmark) for more comparisons.

Lynx3d
05-21-2005, 02:46 PM
Hm...are you really sure you used the same settings (except thread count)?
And if you render with 1 thread, do rendertimes as expected get slower (roughly 80% longer, but doesn't apply to every scene but it should render slower obviously)

Did those scenes perhaps use any plugins that do not ship with LW?
Besides that, what Lightwave version are you using?

somnambulance
05-21-2005, 02:49 PM
Apples are crap, just ship it to me and go back to using your PC.

Darth Mole
05-22-2005, 04:10 AM
Yeah, somnambulance is right. I've got two and they ARE crap. Send it here and I'll make sure it's recycled properly (frankly, I think Mr ambulance is trying to trick you and he'll just use it himself!)

Maxkraft
05-22-2005, 09:10 AM
You should try the same thread count just to check. But if your using Voxels then the Mac needs to be set to one thread or else it will take foever to render. It's a lightwave bug. I was rendering a set of clou and it took 5 min with 4 theard and 1 min with one.

But in general an Amd 64 (fx or otherwize) should be clock eqivelant to a modern g5.

Captain Obvious
05-22-2005, 09:11 AM
Yeah, they're kind of crummy. I'd gladly take it off your hands. Unlike the other two "helpful" people (ie, sheapskates), I'll even pay shipping! ;)



Seriously though, there is obviously something wrong with the scene. Or it uses a lot of something that's slow on the Mac. If you look at www.blanos.com/benchmarks , you'll see that the Macs are fairly compeditive for the most part. At least compeditive enough for a dual 1.8GHz G5 to be faster than an AthlonXP 1700+.

Andres_cg
05-22-2005, 04:07 PM
You guys are just too generous. Just give me your street addy and I'll be sure to have the G5 to you by mon morning.

But, yes, the scene is the firewalkSample that ships with LW8. Same settings, I'm sure. But there hypervoxels in the scene, so that might be the prob.

I'll test some GI scenes after I install all of the content CD.

I'll keep it posted on what happens. Most interesting.

toby
05-22-2005, 04:27 PM
I'm seeing similar times on Blanos to an Athlon 1667, doesn't surprise me too much, the faster G5's have caught up quite a bit, but yours is almost half the ghz, and don't forget it came out at least a year ago -

Multi-threading HV has more overhead on both platforms.

Captain Obvious
05-22-2005, 04:57 PM
So rendering HyperVoxels actually takes a lot longer when you set the threads to more than one? Err... NewTek, care to fix that?


Edit: I tried it on a single frame, and it took almost 20% longer with eight threads instead of one. This is on a single-processor machine.

toby
05-22-2005, 05:49 PM
It can take many times longer, like 5 times longer depending on the scene. It's been like that since they were new. It seems that if your scene takes long enough to render though, multi-threading will still be helpful - I'm guessing that it's splitting the data that takes a while, so if your scene takes longer to render than it does to split the data it will help to have more threads.

I just tried 1 hv that filled the screen, and 2 threads was faster - 2 sec. compared to 3.3 with 1 thread

Captain Obvious
05-22-2005, 06:16 PM
I tried that sunset scene too... Bumped up the resolution a fair bit. 94.9 seconds with eight threads and 94.1 seconds with one thread (or something such). I guess it's the data splitting that made the firewalk sample take much longer with more threads.

Hervé
05-23-2005, 12:39 AM
ouch... I am so happy to have switched for a dual PC... Apple has problems afetr problems.. at least in 3D market...

Apple and 3D are definitly no friends... no rant, just facts... :(

toby
05-23-2005, 12:53 AM
whoa, hardly a fact, but an opinion - remember I work on both too...

nobody's even mentioned a mac problem in this thread!


I guess it's the data splitting that made the firewalk sample take much longer with more threads.
I just did a test with a couple hundred hv particles filling the screen - 8 threads was twice as fast as 1 thread, so it really does depend - surface hv interact with each other if they're close enough, that has to be calculated too...

Lynx3d
05-23-2005, 01:52 AM
Er...are we talking about the "firewalk.lws" on the Content Disk 1? A frame (i tried frame 45) just takes 3.0 seconds to render on my dual Athlon 1800+ with 1 thread, such scenes have never really benefited from more threads (from within Layout with preview enabled at least), it merely goes down to 2.7s with 4 threads (3.2s with 2 and 3.0s again with 8 threads).
If you want to render the whole anim you better launch two screamernet nodes...

Ozzie
05-24-2005, 12:53 AM
On a Dual 2ghz I got 7.5 seconds with 8 threads and 1.1seconds with 1 thread. This for frame 0 hitting F9 after opening the scene.

For frame 45 I got 8.1 seconds with 8 threads and 3.2 seconds with 1 thread.

Hope this helps.

CHEERS.

Mark

toby
05-24-2005, 01:35 AM
AAArrrgh!! I keep forgetting!

Andre_cg, check your System Preferences/Energy Saver/Options and make sure Processor Performance is set to 'Highest'. It can make quite a difference.

Captain Obvious
05-24-2005, 03:52 AM
I just tried some random scene. With highest, it took 30 seconds. Same with automatic. With reduced, it took 37 seconds. Odd. Oh, and I should note that this is in Tiger. They seem to have improved the energy saving quite a bit. In Panther, there was a huge difference in quite a few things between automatic and highest, especially in games and such. In Tiger, the difference seems much smaller.

Lynx3d
05-24-2005, 11:38 AM
Wow, Lightwave on OSX seems to have quite a bit more threading overhead...for whatever reason.

Captain Obvious
05-24-2005, 11:50 AM
That seems odd... In most of the benchmarks (for other applications too) I've seen, a dual G5 does better against a dual Xeon, for example, than a single G5 does against a single Xeon. I suppose it only really makes a difference on scenes that take a few seconds to render.

archiea
05-25-2005, 02:15 AM
are you sure you aren't running LW on the mac using virtual PC!?!?!?!? :D

toby
05-25-2005, 01:42 PM
Wow, Lightwave on OSX seems to have quite a bit more threading overhead...for whatever reason.

I must be missing something, what leads you to believe that? I've tested the HV thing on pc and as far as I can tell it's the same deal...

Andres_cg
05-30-2005, 12:03 AM
Hey guys,

Well I ran a few more scenes against each other. And yes, the dual G5 was 2x as fast as my 2100+ AMD. I ran the Ninja scene, the one with global illumination, and the G5 clocked in just over 1 min @ 8 threads. While the AMD clocked in at just over 2 min. Not bad.

Then I re-tested the firegirl HV scene. I set the G5 to 1 thread and that did the trick, much faster than the 4 thread setting.

Come on Newtek, PLEASE get the multi threading to work on HV's on the G5. That would be so **** cool.

toby
05-30-2005, 02:40 PM
That's good to hear! Faster than a dual athlon or single?

Multi-threading actually does work, put one HV in a scene and make it big enough to fill the entire camera. This renders faster with 2 threads than with 1.

But I'm sure it could work *better*...

Ge4-ce
05-31-2005, 01:21 AM
Here's (http://vbulletin.newtek.com/showthread.php?t=35198&highlight=Ge4-ce) a thread I started some months ago. It gives a pretty good explanation why some scenes benifit more from multithreaded hypervoxels renders, and others don't.

It all has got to do with keeping the processors busy.

If you switch on view render in progress, you'll notice what I mean.

When rendering volumetrics, LW does not automatically start a new part when an earlier part has finished.

3dworks
05-31-2005, 08:34 AM
Hey guys,

Well I ran a few more scenes against each other. And yes, the dual G5 was 2x as fast as my 2100+ AMD. I ran the Ninja scene, the one with global illumination, and the G5 clocked in just over 1 min @ 8 threads. While the AMD clocked in at just over 2 min. Not bad.

Then I re-tested the firegirl HV scene. I set the G5 to 1 thread and that did the trick, much faster than the 4 thread setting.

Come on Newtek, PLEASE get the multi threading to work on HV's on the G5. That would be so **** cool.

andres, there must be something definitly wrong with your setup:

the ninja GI scene took 22,4 s with 8 threads (29,6 with 1, 22,5 with 4 threads) on a G5 2x2.7. but now comes the 'funny' thing: it took 28,9 s (8 threads) on my 2x1.8 G5 powermac (OSX 10.3.9), not much slower...

btw. i did test the firewalk scene on my 2x1.8 ghz G5 too: with 8 threads it rendered at 11,2 s at frame 0 and 9,4 s at frame 45. with the render threads set to 1, the rendertimes where 1,5 s and 3,9 s respectively!

i'm using LW version 8.3. which version of LW did you use?

best regards

markus

Andres_cg
06-04-2005, 03:23 PM
3dworks ,

Using 8.3 here. But I have not upgraded to osx 10.4 yet.

Is it really worth it perfromance wise?

toby
06-04-2005, 04:33 PM
NinjaSample.lws, nothing touched except the threads :
8 threads - 66.7 sec
2 threads - 72.5

dual 2.0 G5
OS 10.3.9
LW 8.2

turn off AA and it's still 37 sec.
I'm assuming you must have changed something in the scene -

3dworks
06-09-2005, 04:20 AM
3dworks ,

Using 8.3 here. But I have not upgraded to osx 10.4 yet.

Is it really worth it perfromance wise?

andres,

i cannot see any big advantage for tiger for using LW. maybe in opengl there is a slightly better performance, but nothing immediately 'visible'.
of course spotlight is a nice thing, but there are also many small glitches, so if you are in production, i would stay away from updating to tiger until the next bug fix update. of course my new G5 wants tigers only, so i will have to stick with it... .-)