PDA

View Full Version : 8.5 - Render Speedups in some situations...



Bog
10-10-2005, 03:28 PM
It really is. Oh my giddy aunt.

I love you guys, NewTek folk, floating in your nutrient tanks and living on a thin protein broth. I love you very, very much. You gave me something free, and it's made out of Quick, just when I'd whumped into a job which desperately needed Surface Effectors, which don't render in FPrime, you flounced into my inbox and gave me the speed I need.

I owe you beer.

Thank you.

Twice. As fast. Oh, lordy, I haven't been this giddy in a long time. It's all made out of fast!

byte_fx
10-10-2005, 03:35 PM
No speed change seen here.

Modeler has problems.

Went back to 8.3 .... might try 8.5 again later.

byte_fx

Bog
10-10-2005, 03:39 PM
With something as complex as 3D modelling software, there's good and bad with every revision. Just with what I'm working on today, I've totally lucked out, and will therefore be dancing a small jig.

Tiger
10-10-2005, 03:44 PM
:i_agree: :thumbsup:
Here`s a small testrender with Hw at it`s best quality settings.
Rentime:41sec :thumbsup:
:lightwave

Bog
10-10-2005, 03:45 PM
<Jefferson_Clay> Full burn for the Blocker. </Jefferson_Clay>

hrgiger
10-10-2005, 03:48 PM
Is the renderer faster in this version? I thought the speed improvements weren't supposed to come until 9.0? Anyone do any benchmarks between 8.3 and 8.5?

Bog
10-10-2005, 03:51 PM
Is the renderer faster in this version? I thought the speed improvements weren't supposed to come until 9.0? Anyone do any benchmarks between 8.3 and 8.5?

Yeah, an hour or so ago. 2 mins 30 - odd for PLD 4- Pass and a Mitchell (Soft) reco in 8.3. Then 1 min 9 s for the exact same frame.

Unless I'm missing something, that's extra warp speed. I suppose, theoretically, I should have bounced the box to make sure that nothing was stowing away, but I shut LW down fully, upgraded, wallopped F9 and those are the numbers.

*shrugs* Seems hella fast to me, dude.

:I_Love_Ne

Captain Obvious
10-10-2005, 03:54 PM
Are you sure you didn't mess with the thread count or something? ;) Anyway, what is it that renders that much faster? Standard raytracing?

Bog
10-10-2005, 03:57 PM
Four threads going in, four threads coming out, high-res Dispmapped subpatch scene with high-res (4k) textures for Colour and Bump, ten Surface Effectors, ten 200-patch models sub-D'd to three. Rendering to 1024x720, square pels.

Dunno what else to say, mate.

UnCommonGrafx
10-10-2005, 04:23 PM
Yeah, they've been in the code...

:I_Love_Ne

Scene info attached. :chicken: Pic attached.
I had an area light in the scene and it rendered at 14:38.00 under 8.3. Some went up as high as 17 mins @ 720 x 486. Changed it to a spotlight and got the times down to around 9 mins.
Now, with 8.5, and more hair, the Spotlight time was a little more than 5 mins. The area light time... (drumroll) 8 mins 42 secs!

Something's been done and it's been good.

Bog
10-10-2005, 04:25 PM
Glad it's not just me!

Intuition
10-10-2005, 04:35 PM
We have to wait awhile to upgrade at work since we are in the middle of a few projetcs but when I get home i'll be busting out some renders.

:thumbsup: :lwicon:

Can't wait. :devil:

ACLOBO
10-10-2005, 04:39 PM
Or even better, I wouldn't mind if newtek posted some scenes to see if what the average speed increase with 8.5 is. All i can say is that it does seem faster and by an impressive amount! IF 2x speed increase is the average then that is darn impressive. Anyway, a clean install seemed to make all the diference in how it was behaving. When I first updated, it was running sluggish as heck - let alone trying to render, but after the clean install of lightwave and then the upgrade, it seems to be running well!

-Adrian

Dodgy
10-10-2005, 05:05 PM
I've done some renders and it does seem about twice as fast. Can't be sure how much faster (I don't keep render times written down!) but one scene which I'm fairly sure was about 40 secs is now 21.7 secs. It definitely rendered a lot faster. Good job!

EmperorPete
10-10-2005, 05:19 PM
8.5 is DEFINITELY a lot faster. I did a test render of a model I'm working on, and it took about half the time it did to render the identical frame yesterday. THANK YOU NEWTEK...!

meanlebh
10-10-2005, 05:28 PM
I just rerendered a scene which took 8 min and 30 sec last night, and it just rendered in 5 min 43 sec....definitely quite a bit faster...this is good....

Fraust
10-10-2005, 05:33 PM
And this version wasn't the super speedy update renderer either... this makes me VERY excited to see what they consider worthy of mentioning as a feature for 9 if this kind of speed is not worth mentioning for 8.5... did they disable something and forget? hehe

Surrealist.
10-10-2005, 05:45 PM
OK where is the "faster" button? I read this thread. Got a woody, did some test renders on a scene with a character that had Bk Drp radiosity and 3 area lights, one with raytraced shadows, saverd scene and installed 8.5 without a hitch did the render and it is now 20 seconds longer.

JamesCurtis
10-10-2005, 07:20 PM
I believe the fellows who had render speed increases had first uninstalled ALL of the earlier LW installs and did a FRESH install of LW 8.0 and then 8.5.

Haven't updated mine as of yet. Plan on doing it next weekend [LW8.5 AND Newest GL drivers].

Still will have a 5600FX card tho' till I can afford a newer card.

SplineGod
10-10-2005, 07:25 PM
At siggraph I heard told that youll really notice the speed increase on larger scene files. :)

Surrealist.
10-10-2005, 07:27 PM
At siggraph I heard told that youll really notice the speed increase on larger scene files. :)


That makes a bit of sense.

JML
10-10-2005, 08:32 PM
I do not see any speed increase either...
is this speed increase for pc, mac?

can you guys try to take a scene in LW8 content that renders faster in 8.5 and then tell us for we try on ours ?

thanks.

JML
10-10-2005, 08:38 PM
I believe the fellows who had render speed increases had first uninstalled ALL of the earlier LW installs and did a FRESH install of LW 8.0 and then 8.5.


is that true ?
can the people that had the speed increase confirm that ?

fabmedia
10-10-2005, 09:02 PM
I would like to know as well. I had it crash on me today, but that's another story. I'm still going to keep 8.3 around. I've learned my lesson.

Auger
10-10-2005, 10:47 PM
Did some tests before installing the 8.5 update. On my machine it's taking longer to render the same scenes.

Other than that, 8.5 is working great.

Put me down for wanting to know the secret for the 2x render speed increase. :help:

WizCraker
10-10-2005, 10:51 PM
Using LW Autobench 102 for Raytace benchmark I have a 22 second less from 8.3. Gained 9.7 seconds on Variations, .7 less on zbuffer, 10.3 seconds less on Tracer - no radiostiy.

Did not do the others as I didn't want to wait any longer. Will do more later.

Processor specs:
http://www.imagereality.com/Images/DellComputer/ProcSpec/procSpec.jpg

holeycow
10-10-2005, 10:56 PM
My first test was on a scene with a large area light and losts of sasquatch fur happening. What took 8 min 15sec in 8.3 has just rendered in 2min 53sec. I couldn't be happier!............unless it had come last week when I really needed that speed increase!

Thank you Newtek!

Auger
10-10-2005, 11:01 PM
My first test was on a scene with a large area light and losts of sasquatch fur happening. What took 8 min 15sec in 8.3 has just rendered in 2min 53sec. I couldn't be happier!............unless it had come last week when I really needed that speed increase!

Thank you Newtek!

Did you update over the top of 8.3 or did you install clean?

toby
10-10-2005, 11:36 PM
I am still having problems with it loading TIFF files - I get errors which don't affect the scene but you have to hit enter MANY times for many tiffs!
Change your Alert Level to Expert ( general options ) and you won't get pop-up alerts.
And if you know how to create actions in photoshop, you can batch-process a conversion from all the tiffs to pngs.

Dodgy
10-11-2005, 04:04 AM
I did an install over the top of my old 8.3 and I seem to get the speed increase. It's altogether possible that the speed up only occurs when certain functions are utilised, depending on what exactly has been optimised.

Bog
10-11-2005, 04:11 AM
Mine's a culiminative install... I think this box started out with 7.5c on it and has had about every possible update installed.

Then again, I'm minded of a time when I had an Athlon and a PIII box sat side by side - the Athlon whupped the Pentium for render times, until I switched Traced Reflections and Refractions on, at which point the Pentium gallumphed ahead.

Given the knee-watering math involved in 3D Stuff (technical term, there) Your Mileage May Vary depending on what it is you're actually rendering.

Steve McRae
10-11-2005, 04:48 AM
edited - edited

Bog
10-11-2005, 04:50 AM
Your "clear polygon trick" intrigues me.

What does it do?

Captain Obvious
10-11-2005, 04:51 AM
How about we compile a database of the speed up (or down)? If you've seen a change in speed, please list the following, if you want to contribute:

*time it took before 8.5 (approximately if needed, but preferably not too far off)
*render time in 8.5
*Mac or Windows
*quick rundown on hardware (CPU number, kind and speed, mostly)
*thread count
*which raytracing options you're using (please list all)
*radiosity
*number and kinds of lights in the scene
*total number of polygons
*anti-aliasing settings
*resolution
*volumetrics and HyperVoxels
*other "special features" that effect render time

You could post it here or email me at [email protected] and I'll try to make something useful of it. Feel free to provide as little or much information as you want.

Captain Obvious
10-11-2005, 04:55 AM
Your "clear polygon trick" intrigues me.

What does it do?
Put a flat and entirely transparent polygon in front of the camera. Turn on raytraced transparency. Since that's the polygon closest to the camera, and it obscures everything else, Lightwave will render that polygon and that polygon alone, starting at the top left pixel and working its way down to the lower right pixel, instead of the standard one polygon at a time way. Sometimes Lightwave renders a polygon that will later become obscured by another polygon (for example, if you have a single large floor polygon, it will sometimes render before all the chairs and tables and what have you). This is a waste, since it calculates a lot of pixels you won't see in the final image. By putting that polygon in the front, you can do away with this.

If NewTek did something to reduce the risk of this happening, it could definitely account for a large speed-up, as well as that this trick doesn't work anymore.

-EsHrA-
10-11-2005, 05:06 AM
http://www.funnyfarm.tv/rendertrick.htm


Your "clear polygon trick" intrigues me.

What does it do?



mlon

Steve McRae
10-11-2005, 05:10 AM
it looks like, with admittedly only inital quick tests, that the clear poly trick still has *some* affect on the speed of the scene

* edit: more testing - the clear poly trick still works - cut my more intensive radiosity tests by half with 8.5

Bog
10-11-2005, 05:51 AM
See, I love stuff like this. Strange, esoteric things that would only make sense to a fellow practitioner of the Dark Arts. Erm. Yeah.

Very cool - cheers, fellers!

Some more details on the Little Scene that Could:

Single high-complexity Sub-D surface, wearing a 4k displacement, colour, bump and clip map, two Distant lights, ten small Sub-D objects, Prem's Surface Effectors.

Render Res 1024x720, Square Pels, 7 pass anti-A, Mitchell Soft reco.

No tracing or GI at all.

3.2GHz P4 Hyperthreading, 1Gb RAM, almost a default Compaq Presario.

Times posted at the beginning of this thread.

mav3rick
10-11-2005, 05:56 AM
it would be easyer if some1 from newtek just say a word on this speed phenomena

holeycow
10-11-2005, 06:31 AM
I installed over 8.3.

The scene I used for comparision had a dog taking up one third of the screen.

There was a large area light and 5 passes of normal motion blur. The biggest speed increase seemed to be with Sasquatch, Which seems to be more than twice as fast.

Will 9 give us similar increases in speed?

JML
10-11-2005, 06:52 AM
looks like lot of people are having speed increase with sasquatch..
I don't use sasquatch so I'm going to instal 8.5 at work, keeping 8.3 and do
some test..
I will let you know.

kcole
10-11-2005, 07:32 AM
I ran a quick benchmark on my system with the Lighting\Parfume scene, camera set to single pass:
7.5 - 7m2s
8.2 - 6m19s
8.5 - 5m51s

Newtek, Intel called and asked that you stop speeding up your app. Something about you screwing up the demand for processor upgrades. :)

byte_fx
10-11-2005, 08:18 AM
Still haven't seen any speed increase - maybe because I'm using benchmark scenes from both Classic and 8.0 cd's.

But at least doing a complete uninstall, including dongle drivers, followed by reboot and re-install 8.0-->8.3-->8.5 (parallel dongle .... have to install 8.3 so 8.5 can find dongle) fixed the problems with Modeler.

It'll be a few days but I'll run some tests comparing 8.3 vs. 8.5 for some interpolated radiosity scenes when time permits.

byte_fx

Bog
10-11-2005, 08:27 AM
Yeah, I haven't actually used Modeler 8.5 yet, been tied up with rendering.

Really quick rendering! :D :D :D :D :D

(Sorry, I'm still jazzed about it)

Auger
10-11-2005, 08:39 AM
Has anyone running Athlon 64 processors seen the render time improvements? That's what I'm running and I am not getting the speed increase. :confused:

JML
10-11-2005, 09:09 AM
some scenes.
(P4 3.2 2GB 1thread)
(image rendered at 720x486)

the room is made of at least 2 area lights, some spot lights..
8.2 : 7 mn 05
8.5 : 6 mn 38

the highway cars were made with hd_instance (volumetric plugin)
8.2 : 4 mn 12
8.5 : 4 mn 04

perfum in lw8 content
8.2 : 6 mn 31
8.5 : 6 mn 18


the room seems to be the fastest one.. it's good for us

be careful people, I just saw and confirmed a bug in the bug workshop that 8.5 eat even more memory than before (at least on XP)

Kaiten
10-11-2005, 09:26 AM
Has anyone else noticed the glitch with transparent surfaces in modeller? Some no longer appear transparent in wire/texture modes????

Bog
10-11-2005, 09:27 AM
Has anyone running Athlon 64 processors seen the render time improvements? That's what I'm running and I am not getting the speed increase

I shall be properly benchmarking this once LW64 arrives here - I've got a shiny new twin Athlon 64 all waiting and ready. I'll light some candles and open a bottle of wine as well. *twitch*

pixym
10-11-2005, 09:34 AM
PC - P4 3,06ghz HT off 1.5 GB ram
Clean install ATI latest drivers
Clean Install from LW 8.0 CD then updated to 8.5

Archi Viz scene
poly count: 1 million (999 673 exactly)
extra RT optimization off
1 raytraced spot light
small part with reflection (windows)
no radiosity
1 thread
720x576
AA low enhanced, no adaptive sampling
reconstruction filter: classic

8.2: 3mn 38sec
8.5: 3mn 57sec

Same scene with an area light instead of the spot light:
8.2: 31mn 19sec
8.5: 36mn 27sec

I do not see any speed improvement but the contrary
:thumbsdow

byte_fx
10-11-2005, 10:09 AM
For some reason the problems with modeler came back after a shut down and restart

So render speed increases, which I have not seen on my P4 anyway, don't mean anything to me.

Looks like LW evolution stops at 8.3 for me at least until I can get 8.3 installed on one box for modeling and 8.5 on another for rendering (if I ever see any actual render speed increases).

Man - am I glad I didn't pre-order 9.

byte_fx

JML
10-11-2005, 10:53 AM
8.5 was not really supposed to see any speed increase.

pixym
10-11-2005, 10:55 AM
8.5 was not really supposed to see any speed increase.
Well said JML :agree:

Bog
10-11-2005, 11:00 AM
8.5 was not really supposed to see any speed increase.

Ah, right... best I start complaining that it's buggy and ruining my "Render Days" then ;)

Rabbitroo
10-11-2005, 11:14 AM
8.5 was not really supposed to see any speed increase.

Other than those in the press release?

http://www.newtek.com/news/releases/08-01-05b.html

:confused:

-K

tektonik
10-11-2005, 11:24 AM
anything new with regards to FPRIME ?

any more hooks activated in the sdk ?

kcole
10-11-2005, 11:33 AM
Other than those in the press release?

http://www.newtek.com/news/releases/08-01-05b.html

:confused:

-K
The discussion is about render speed increase, which is not stated in the press release.

KRoo2
10-11-2005, 11:35 AM
The discussion is about render speed increase, which is not stated in the press release.

Ah--get it. ;)

-K

sonofmickel
10-11-2005, 11:52 AM
Most of my test renders are about the same a little faster or slower. On the other hand, Radiosity Box was 100 sec faster! Hmmm... I guess that means that my Blanos tests are inconclusive.

JML
10-11-2005, 11:56 AM
Other than those in the press release?
http://www.newtek.com/news/releases/08-01-05b.html
:confused:
-K

I did not see any speed increase information for LW 8.5 32bit in that press release.

--------

here is another test I did during lunch...
it's a little bit faster.. but it's better to be a little bit faster than a little bit
slower.. right ?

720x486 enlow , 6 area light, each quality 3

lobby
8.2 : 16 mn 31
8.5 : 15 mn 18

JML
10-11-2005, 12:13 PM
I already posted that one but look at the reflection on the bottom of the bottle..
the one in 8.2 looks grainy,
the one in 8.5 looks much better.. (and tiny bit faster)

perfum in lw8 content
8.2 : 6 mn 31
8.5 : 6 mn 18

(only 1 Pass of antialiasing.)

mkiii
10-11-2005, 12:33 PM
I haven't had chance to test the render times yet, but I have noticed that the OGL in Modeler is slow.

For example, a 50k poly model with a number of largish (2000x2000) textures takes around 17 seconds to draw a textured perspective window, with the OGL set to 1024 sized textures. This is pretty much identical to the speed of 7.5d However, the screen display is corrupting in the way 6.0 used to. with badly shaded polys, and transparent sections appearing all over the mesh. This wasn't there before I updated yesterday.

Now I realise that modeler has been pretty much ignored once again, but I somehow expected *some* improvement.

ACLOBO
10-11-2005, 01:12 PM
Well, we did get multishift in modeler, plus a lot more config buttons. Sure I would have liked to have seena lot more. However, 8.5 is a free update and realistically, we don't have to get new features at all. I could see Newtek just fixing bugs for these freebies.

As far as teh speed increases go and doing a fresh install, someone mentioned that they had to install 8.0, then 8.3 and then 8.5. They were mentioning that they didn't see the speed increase. I installed 8.0 then right to 8.5 (parallel dongle) and have noticed that it is a bit zippier. I haven't had a whole lot of time to really take 8.5 throughs ome paces as I start up teaching again tomorrow and have been planning/preparing for that. However, I might have a little time to goof around on Thursday and the weekend (maybe).

-Adrian

hrgiger
10-11-2005, 02:57 PM
I'm finding about what JML found as far as the speed increase goes. I picked 3 scenes at random from the content directory and did some benchmark renders between 8.3 and 8.5. Here are the results:


8.3 8.5

Benchmarks/Teapot.lws 2min44sec 2min32sec (2.6% faster)

Digital Confusion/BlueToy.lws 7min50sec 7min30sec (4.3%faster)

Lighting/Cobra.lws 10min26sec 10min10sec (2.6%faster)


This is rendering on my 2.4GHZ Pentium 4, 1.5GB RAM, GeForce 6200

So the average speed increase that I've found from these is around 3%. Hey, at least it's something.

Bog
10-11-2005, 03:00 PM
I think I was way lucky that the first scene I happenned across landed right in the middle of Opto Country, and therefore slammed me back in my seat cushions from the speed-boost. I'm semi-tempted to edit the thread title - especially with some of the raw negativity that this release has engendered :(

However, my numbers are good for my system. I'm under the Deadline Gun at the moment (frankly, the speedup's been a life-saver) - once this gig's done, I'll set up a similar scene and chuck it on the web if anyone wants to have time-trials with it.

fabmedia
10-11-2005, 03:10 PM
...a 50k poly model with a number of largish (2000x2000) textures takes around 17 seconds to draw a textured perspective window, with the OGL set to 1024 sized textures. This is pretty much identical to the speed of 7.5d


That's been the issue with LW for quite sometime now. The OpenGL performance of Modeler has been horrible. You shouldn't have to have a smoking fast machine to have basic functionality.

CAClark
10-11-2005, 04:23 PM
I have a scene I have been working on at work, which has a ground plane, a silver IMI Desert Eagle, 2 clips, and 4 rounds of ammunition. So bascially we got some decent feflections on the go, plus an area light. HDRI for env, though not lighting.

LW8.3 rendered it in 1m 17s
LW8.5 rendered it in 1m 39s

So figure that out.

A totally different scene with glass objects and caustics

LW8.3 : 36.4s
LW8.5 : 39.7s

So it's not that I am bothered I haven't any blinding render speed improvements, cos I wasn't expecting that. I wasn't execting the times to get slower consistently though.

Cheers!

Chuck
10-11-2005, 04:26 PM
I think I was way lucky that the first scene I happenned across landed right in the middle of Opto Country, and therefore slammed me back in my seat cushions from the speed-boost. I'm semi-tempted to edit the thread title - especially with some of the raw negativity that this release has engendered :(

However, my numbers are good for my system. I'm under the Deadline Gun at the moment (frankly, the speedup's been a life-saver) - once this gig's done, I'll set up a similar scene and chuck it on the web if anyone wants to have time-trials with it.


New title work for you? :)

Bog
10-11-2005, 04:46 PM
If it does for you, Chuck - I really don't envy you your job, mate!

Thing is, the very first scene I loaded, a real-world project for the BBC, was hammered out by 8.5's geometry core in half the time that 8.3's did it in. I quite like getting excited about things, and stuff, but if it's turning into a bone of infinitely-qualified contention then perhaps you're correct to tweak the topic.

It's still going twice as fast for me, though, and I can't wait to see people's faces in the studio tomorrow when I unlimber the laptop and pin 'em to the wall :D

3dworks
10-12-2005, 04:07 AM
I already posted that one but look at the reflection on the bottom of the bottle..
the one in 8.2 looks grainy,
the one in 8.5 looks much better.. (and tiny bit faster)

perfum in lw8 content
8.2 : 6 mn 31
8.5 : 6 mn 18

these seem to be rendered not at the default settings, but without any antialiasing, right? would it be possible to rerender the scene with default settings on your PC and post it (possibly with a JPG compression quality level at 100%)? could be there is a considerable difference in the OSX version and PC renderings here. please have a look at here:

http://vbulletin.newtek.com/showthread.php?p=307139

cheers

markus

Chris S. (Fez)
10-12-2005, 05:12 AM
OpenGL in Modeler 8.5 is wonky and slightly slower than 8.3. I am getting all sorts of strange transparency artifacts.

**** the Mr. Fancypants GL display. Just make her FASTER. Mush! Mush!

mkiii
10-12-2005, 06:22 AM
Sounds exactly the same as I'm getting with anything more complex than a realtime character.

[Rant mode]
UV map still has the same bugs as before, and I get the feeling that modeler has been ignored again.

This is not a problem for all you speed seeking render monkeys out there, but rather than having one new toy added, to modeler, I would have preferred the update to have actually fixed bugs that have been there for ages.

Admittedly, one modeler crash bug introduced in 8.3 seems to have been fixed. but for anything heavy, it looks like I might still need to keep a copy of 7.5d handy.

I would like to suggest (again) that Netek do just a little bit more testing on high poly (and I mean a million polygons+) & heavily textured models. In modeler, not Layout.

My last few jobs included a 1.5 million poly ship, that could not be loaded into 8.3 (it loads in 8.5 thankfully), and a 1 million poly+ object for a certain recent space related film. These objects were both making modeler creak at the seams, and had to be broken up into separate objects, and neither of them had much in the way of textures.

Both of these loaded into 8.3 layout but killed modeler. Is this the way to stop people moving to other apps for modelling?

LW modeler was always the best IMO, but it now needs a serious overhaul if it is going to keep up.

Flame away....
[/Rant mode]

JML
10-12-2005, 06:31 AM
these seem to be rendered not at the default settings, but without any antialiasing, right? would it be possible to rerender the scene with default settings on your PC and post it (possibly with a JPG compression quality level at 100%)? could be there is a considerable difference in the OSX version and PC renderings here. please have a look at here:

http://vbulletin.newtek.com/showthread.php?p=307139

cheers

markus

yes, the scene was rendered with just 1 pass.
that's the only change I did. I did not want to wait 30minutes.. and still can't because I'm at work.
I will try at home.

JML
10-12-2005, 07:09 AM
there, another one..

scene is radiositybox in content folder
150x150 (instead of 600x600)
1thread
same antialiasing
P4 3.2ghz

8.2 : 4 mn
8.5 : 3 mn 31

pretty good speed increase there..

hopefully, in LW9 it will be 30 second ;)

pilF
10-12-2005, 07:50 AM
Just installed 8.5 over 8.3 did a rerender, lo poly character, 800 x 600 4 pass aa mitchell, hidri light interpolated 4x12, 8 threads on dual opteron in 32 bit mode under xp

8.3 ---> 293 seconds
8.5 ---> 384 seconds


?

JML
10-12-2005, 07:58 AM
Just installed 8.5 over 8.3 did a rerender, lo poly character, 800 x 600 4 pass aa mitchell, hidri light interpolated 4x12, 8 threads on dual opteron in 32 bit mode under xp

8.3 ---> 293 seconds
8.5 ---> 384 seconds

?
pretty bad render time..
your computer was not doing anything else while rendering, right?

even though you have a dual machine, try to render with 1 thread the
radiosity box in 83 and 85 to see what happens..
and then do the same thing with 4 and 8 thread..

(that's a lot of test, but those test should not take too much time if you change the default resolution to 150x150)

pilF
10-12-2005, 08:06 AM
No, the pc only ran lw, I tested the secene in 8.s and it performed fastest with 8 threads. I'll test the settings you suggested anyway.

Steve McRae
10-12-2005, 08:11 AM
for those of you who have seen render times decrease - are you running athlons or opterons?

Chuck
10-12-2005, 08:31 AM
Sounds exactly the same as I'm getting with anything more complex than a realtime character.

[Rant mode]
UV map still has the same bugs as before, and I get the feeling that modeler has been ignored again.

This is not a problem for all you speed seeking render monkeys out there, but rather than having one new toy added, to modeler, I would have preferred the update to have actually fixed bugs that have been there for ages.

Admittedly, one modeler crash bug introduced in 8.3 seems to have been fixed. but for anything heavy, it looks like I might still need to keep a copy of 7.5d handy.

I would like to suggest (again) that Netek do just a little bit more testing on high poly (and I mean a million polygons+) & heavily textured models. In modeler, not Layout.

My last few jobs included a 1.5 million poly ship, that could not be loaded into 8.3 (it loads in 8.5 thankfully), and a 1 million poly+ object for a certain recent space related film. These objects were both making modeler creak at the seams, and had to be broken up into separate objects, and neither of them had much in the way of textures.

Both of these loaded into 8.3 layout but killed modeler. Is this the way to stop people moving to other apps for modelling?

LW modeler was always the best IMO, but it now needs a serious overhaul if it is going to keep up.

Flame away....
[/Rant mode]


The serious overhaul is in the works for v9, as we announced last August at SIGGRAPH. Check the LightWave v9 Feature List entries for Modeler:

LightWave v9 Feature List (http://www.newtek.com/lightwave/lw9featureprelim.php)

lardbros
10-12-2005, 08:45 AM
Chuck, Proton and Kurtis have such a tough job here sometimes. People are after everything in a simple point upgrade. I knew 8.5 was going to solve bugs and such, and so far have had only a couple of crashes, which happens with every piece of software under the sun.

Version 9 is hopefully (and promises to be) going to be the single most significant release Newtek has had, and as such i think people are being a bit harsh. Yeah, i've had a moan or two about previous issues, but i think Newtek are heading in the perfect direction. Give the guys who show their faces around here some credit, they are doing a brilliant job

T-Light
10-12-2005, 09:04 AM
Lardbros-

but i think Newtek are heading in the perfect direction. Give the guys who show their faces around here some credit, they are doing a brilliant job
:agree:
Full speed ahead and good luck with 9.0 guys. :)

Bog
10-12-2005, 09:06 AM
I think my feelings on the subject are crystal clear at this point.

CAClark
10-12-2005, 09:18 AM
The serious overhaul is in the works for v9, as we announced last August at SIGGRAPH. Check the LightWave v9 Feature List entries for Modeler:

LightWave v9 Feature List (http://www.newtek.com/lightwave/lw9featureprelim.php)

9's where it's at, and I surely hope it delivered beyond the current shortcomings (which we have wored with for a while, so don't get me wrong, I'm slating LW as total garbage).

Is opengl going to be much more responsive in modeller Chuck... you may not be in a position to asnwer that I know... but enquiring minds want to know ;)

Cheers!

CAClark
10-12-2005, 09:19 AM
No, the pc only ran lw, I tested the secene in 8.s and it performed fastest with 8 threads. I'll test the settings you suggested anyway.

I get slower times in 85 as well.

Cheers!

mkiii
10-12-2005, 10:03 AM
Chuck

I know you have great things planned for 9.0 & beyond.

The problem is that I was hoping that 8.x might have had some of it's bugs squashed. The fact that optimisations are not there in this release is less of a problem than bugs like the UV problem I posted in the Bug thread, that still isn't fixed for example.

colkai
10-12-2005, 12:19 PM
AMD1800+XP
1GB RAM
FX5200 Geforce

Tested a scene with an area light and radiosity.
LW8.2 ... 202.8 secs
LW8.5 ... 217.6 secs

Gonna test a few more and see where it leads

JML
10-12-2005, 12:28 PM
looks like most people getting speed decrease are people with AMD, weird..

colkai
10-12-2005, 12:30 PM
Aye..always the poor cousins ;)

Captain Obvious
10-12-2005, 01:10 PM
Hmmm... It could be SSE3 optimizations. Does anyone have a very new Opteron or Athlon64 that has SSE3? Can the Pentium 4 users please report whether you have Prescotts or Northwoods? There are 3.2GHz models of both, and only the former has SSE3.

Bog
10-12-2005, 03:00 PM
Prescotts or Northwoods

*blank gaze*

I have no idea how to find out. How do I find out? And which one's the what now?

I play with crayons!

CAClark
10-12-2005, 03:06 PM
Hmmm... It could be SSE3 optimizations. Does anyone have a very new Opteron or Athlon64 that has SSE3? Can the Pentium 4 users please report whether you have Prescotts or Northwoods? There are 3.2GHz models of both, and only the former has SSE3.

I've a dual-core Athlon 4800+ less than a month old which has MMX, SSE, SSE2, SSE3 apparantly. I've seen a pretty consistent 10% loss on render times.

Cheers!

hrgiger
10-12-2005, 03:15 PM
I'm actually not sure what mine is but my pentium 4 system is over 2 years old now and is a 2.4GHz and I'm seeing about a 3% gain (meaning less time) on renders.

Captain Obvious
10-13-2005, 01:07 AM
I've a dual-core Athlon 4800+ less than a month old which has MMX, SSE, SSE2, SSE3 apparantly. I've seen a pretty consistent 10% loss on render times.

Cheers!

I'm actually not sure what mine is but my pentium 4 system is over 2 years old now and is a 2.4GHz and I'm seeing about a 3% gain (meaning less time) on renders.
Well, that kills the SSE3 theory. Hmmmm...

pilF
10-13-2005, 01:59 AM
don't know about the sse3 thing, did some more testing on my newest system:
sp 32 bits
AMD dualcore opteron 265

hidri lit scene sub d character

lw 8.5 (32 bits) lw 8.3
2 threads: 768 s 599 s
4 threads: 396 s 306 s
8 threads: 384 s 294 s

Think I'll wait for 9 and stick to 8.3 fttb.

Captain Obvious
10-13-2005, 02:32 AM
The dual-core Opterons all have SSE3, I think, so that definitely kills the theory.

mkiii
10-13-2005, 04:49 AM
*blank gaze*

I have no idea how to find out. How do I find out? And which one's the what now?

I play with crayons!

If it makes a noise like a pair of Jags, and all your meat pies keep disappearing, then you have a Prescott. :thumbsup:

FranK10
10-13-2005, 05:04 AM
I made some test benchmark using the autobench from Chris Blanos.

Here what I get:

Lower time Render:
.......................... 8.3...........8.5
Raytrace:...........131.6.........116.8
Tracer-No Rad.....386.7.........365.8
Tracer-Rad..........359.2........334.5
Rad.Refl...............73.4...........61.2
Teapot...............152.4.........141.5

Worse Time Render:

Nebulae...............141.2.........151.2
Variations............145.3..........148.5

And the parfum scene:

..........................517.5..........470.2

So I have a 3-20% gain, and 2-7% loss only in that 2 scenes (they have volumetrics, could it be volumetrics slower?)

Anway it's a general gain of about 10% :)

And I made it on a Pentium-M 1,6Ghz, so SSE-3 doesn't count.

CAClark
10-13-2005, 05:18 AM
don't know about the sse3 thing, did some more testing on my newest system:
sp 32 bits
AMD dualcore opteron 265

hidri lit scene sub d character

lw 8.5 (32 bits) lw 8.3
2 threads: 768 s 599 s
4 threads: 396 s 306 s
8 threads: 384 s 294 s

Think I'll wait for 9 and stick to 8.3 fttb.

On an Athlon X2 4800+ .... how many threads is it optimum to use when rendering? I always set it to 2?

Cheers!

Captain Obvious
10-13-2005, 07:03 AM
As far as I've seen, if you have more than one hardware thread (ie two processors, two cores, whatever), set the threads in Lightwave to eight for best results.

Auger
10-13-2005, 08:00 AM
This article makes mention of Newtek working with Intel: http://vfxworld.com/?sa=adv&code=91367ff7&atype=news&id=15098
I won't assume that that means they aren't optimizing for AMD or Mac though.

CAClark
10-13-2005, 08:07 AM
As far as I've seen, if you have more than one hardware thread (ie two processors, two cores, whatever), set the threads in Lightwave to eight for best results.

Mmm, thanks, I'll try that tonight.

Cheers!

umstitch
10-13-2005, 11:47 AM
well i run an x2 3800, and i can confirm that running 4 threads is the most optimal in most cases.

8 threads goes slower

eg: the teaset sample from the original lw8 install cd-

4 threads =56 secs

8 threads =58.6 secs :hey:

CAClark
10-13-2005, 01:21 PM
Cheers umstitch, gonna try it in a very short while.

Cheers!

toby
10-13-2005, 11:30 PM
You never know what thread setting will be fastest! For the longest time, 2 threads was usually fastest for me. Then after some OS and LW versions later, 8 was more often the better choice.

You can make educated guesses though. Multi-threading divides the frame top to bottom, so with 2 threads, one cpu renders the top half and the other renders the bottom. If you're rendering an ocean view with ray-traced reflections on a heavily sub-divided and displaced ocean object, but then a simple backdrop sky, the sky will be done in an instant while all the hard work, the bottom half, will be rendered by one cpu. In this case 4 threads will usually go faster, because it allows both cpu's to share the hard part.

Frames that already render fast (a few seconds ), or that only have elements in the bottom third of the frame may very well render fastest with 1 thread, because it takes extra time just to divide the frame into more threads - the more threads, the more time it takes.

lots
10-14-2005, 09:48 AM
4 threads also seems to be fastest on my dual Opteron 246. Now, if I had dual dual cores, 8 would most likely be faster...

Of cource this is the average for me. And the post above shows that it depends on what you are doing and where the geometry density is highest..

Weepul
10-15-2005, 04:53 PM
You can make educated guesses though. Multi-threading divides the frame top to bottom, so with 2 threads, one cpu renders the top half and the other renders the bottom. If you're rendering an ocean view with ray-traced reflections on a heavily sub-divided and displaced ocean object, but then a simple backdrop sky, the sky will be done in an instant while all the hard work, the bottom half, will be rendered by one cpu. In this case 4 threads will usually go faster, because it allows both cpu's to share the hard part.

Frames that already render fast (a few seconds ), or that only have elements in the bottom third of the frame may very well render fastest with 1 thread, because it takes extra time just to divide the frame into more threads - the more threads, the more time it takes.

That's what I've found. The more imbalanced your composition is in terms of hard-to-render areas, the faster it'll go with more threads. The more uniform the composition in terms of render difficulty, 2 threads will be fastest. Or if it's really simple, 1 thread will save the time it takes to break up the scene for the processors.

FWIW I've never seen 4 threads go faster than either 2 or 8 depending on scene, on my dual 1.25 GHz G5.

Steve McRae
10-16-2005, 09:16 AM
I have seen my opteron box likes 8 threads and my P4 box likes 4 threads on the same scene.

toby
10-16-2005, 02:09 PM
btw

It seems that multi-threading is never any benefit for single proc systems, even though 8 threads works great with only 2 procs...

WilliamVaughan
10-16-2005, 03:11 PM
It's great to see these speed increases. I've been running 8.5 for awhile now and have been quite happy...we are about to upgrade the entire studio at the school and use it on the current production. I'll let you guys know how it works for us when we finish in 3 months.

Keep the trest results coming..thanx~!

CAClark
10-16-2005, 03:11 PM
well 2,4 or 8 yeald render times within seconds of each other, so I'll eave it on 2 for now, but I have ditched 8.5 for the time being as it is consistently slower than 8.3 with every scene.

Cheers!

Chris S. (Fez)
10-16-2005, 04:08 PM
I am moving back to 8.3.

JML
10-16-2005, 04:27 PM
I'm staying with 8.5 and going to install 8.5 on all the machine at work now
I know things are fine.

(we have 8.2 installed right now, we had problems with some things in 8.3, but now those things were fix in 8.5)

Matt
10-17-2005, 02:10 PM
You can make educated guesses though. Multi-threading divides the frame top to bottom, so with 2 threads, one cpu renders the top half and the other renders the bottom. If you're rendering an ocean view with ray-traced reflections on a heavily sub-divided and displaced ocean object, but then a simple backdrop sky, the sky will be done in an instant while all the hard work, the bottom half, will be rendered by one cpu. In this case 4 threads will usually go faster, because it allows both cpu's to share the hard part.

Good tip Toby, never really thought of it like that, I always use 8 threads at the moment.

Cheers
Matt

papou
10-17-2005, 06:12 PM
somes bench with threads too:
http://www.frenchwave.org/Forums/viewtopic/t=1446/start=20.html

groove660
10-20-2005, 09:17 PM
Couldnt be bothered reading the 8 pages of posts, but is there a problem with LW Auto Bench and XP64 or LW64??? I ask becuase it will not run on my PC.

Anyways since i couldnt get that to work i did a test render of the moon scene supplied on the content CD with 8.3. Under LW8.3 and XP Pro(32Bit) it took 14mins and 14secs to render. Under LW64 and XP64 it takes 12min 01secs. So a saving of 2mins!!!! Not bad indeed.

Cheers