PDA

View Full Version : LWSN slower in rendering than LW itself??



ruud
06-14-2005, 02:21 AM
Puzzle:

Same render:

LWSN: 3:42 minute
LW[8.3]: 2:41 minute

Now, why is it that Lightwave [8.3] itself renders so much faster than LWSN??

I can't recall that this happened with LW 7.5. And will it be fixed, now I'll think twice before dragging my scene into screamernet.

Did you have encountered this same problem or is it just the fault of my poor Pentium.

Greetz

Ruuter

SCS5
06-14-2005, 03:13 AM
I pointed this out in way back in LW 5.5. I had the guys in the California (Bob Hood, Etc..) office all scratching their heads! They named the Bug after me. The Sc… Sc…… Screamernet Bug. If you look at the task manager on ANY OTHER PC on the render farm, it is using virtually NO RAM, OR PROCESSOR RESOURCES! While the host PC’s Processor, & Ram usage are buried! And render times on the Render Nodes was sometimes 3 or 4 times slower than rendering the same project on a single pc! They focused on this for a couple of weeks, and when no one could figure it out they brushed it under the carpet! :confused:

ruud
06-14-2005, 04:02 AM
but it makes use of the same render engine...

That's wierd, but as I said, I can't recall that this same problem occured in LW7
or were the rendertimes also different with Seven?

SCS5
06-14-2005, 04:10 AM
This goes all the way back to 5.5! Just look at the Task Manager on any render node and you'll see it's hardly using it's Ram, or Processor!...

JML
06-14-2005, 06:16 AM
we have 12 pc render nodes and we did not noticed anything like that...
lwsn render faster or equal of lw8 for us...
are you sure both (lwsn lw8) are using the same plugins folders and same config files,etc .. ?

SCS5
06-14-2005, 06:54 AM
I believe it’s relative to the complexity of the scene. Are you timing a render on the host PC by itself (No Screamernet) then timing the same render on a Screamernet node render? Have you looked at the Task Manager and checked Processor, & Ram usage on the Render Nodes? Newtek verified this bug with me, and then sent it to the Development team in San Francisco, Bob Hood confirmed it IS A PROBLEM!!…Then squashed it when they couldn’t fix it!

ruud
06-14-2005, 07:47 AM
hope it gets fixed near LW[X] :(

I do think that on a dual processor it gets more advantage out of its memory, because it gets divided by two.

JML
06-14-2005, 08:23 AM
just to make sure, I tested rendering one of our scene with lwsn and lw8 on the SAME machine,

LW8 rendered the frame in 2mn55s
lwsn rendered the same frame in 2mn52s

the scene has 567 885 polys and a lot of volumetrics..
(508 objects)

how big is the scene you had problem with ?

JML
06-14-2005, 08:25 AM
we render a lot here we never noticed anything like that..

SCS5
06-14-2005, 09:03 AM
Try sending the scene files out to 5 or 10 OTHER PC's on the render farm, then time the render after it has been rendering for 30-to 70 minutes the times should increase dramaticaly On the RENDER FARM PC's and LOOK AT THE PROCESSOR, & RAM USAGE not the HOST PC! You should see almost no Ram, or Processor usage under Task Manager on the Render Farm PC's ............. this is the difference.

Lightwolf
06-14-2005, 09:58 AM
Weird, I never had that problem. 5 Dual CPU boxes in the farm, processor usage usually around 100%. (Using Spider)
What render controller are you using? What OS?
I do have nodes _crashing_ and not responding, but if they render they render fine and quickly.

Cheers,
Mike

JML
06-14-2005, 09:58 AM
rendering on the farm is pretty much fine...
the only time we had a weird problem is when most lwsn crashed after sending like 20 scenes. (and those ones had 3-4 millions polys +lots of image maps)

when it happened, it seemed to me that every time lwsn loaded a new scene,
it didn't totally empty from memory the previous scene.
So after a lot of scenes , little bit by little bit, it didn't have enough memory to render the rest..
to fix that, we just had to kill the node and relaunch it.. and everything was fine again..

so that was memory issue..
but renderwise, lwsn seem to render as fast as lw8.

on the same scene as before, I rendered on 12 nodes, it took:

6mn on each node on the renderboxxes (dual P4 3.0ghz 4gbram) (533fsb)
4min30s on dell P4 3.2ghz 2gbram (800fsb)
4min on dell P4 3.0ghz 2gbram.(800fsb)
(one of the dell has faster ram and motherboard..)
when rendering locally with lw8 , it took 4mn30.. so the same, just different computer performace...

and I all time get 50% in the system performace (not using multithreading)
(so really 100% of cpu power)
and the ram keeps going up... as it should..

SCS5
06-14-2005, 11:06 AM
Win XP Pro
LW's Screamernet Controller

Years ago, I wrote a batch file to activate all my Screamernet render nodes, then I'd send the scene out to all the nodes. That was it! But, for some reason the renders would eventually slow way down! We thought it was some sort of memory leak, but could never isolate the problem! Newtek could duplicate the problem, at least they could a couple of years ago, but I never got an answer back..I haven’t used Screamernet for over a year because of that.

Exception
06-16-2005, 11:24 AM
Using LightNet here, and no problems concerning this behaviour either.
Typical behaviour as expected, 50% cpu usage on a Hyperthreading CPU, and 100% on a normal CPU.

Tom Wood
06-16-2005, 12:43 PM
I use LW7.5c on a hyperthreaded dual Xeon, so there are four render nodes - two physical and two virtual. My animation scenes are pretty simple, so a LW render is about 9 seconds for a single frame. In Screamernet it takes about 12 seconds per frame. But the overall average is about 3 seconds per frame since there are four nodes working. Task Manager shows all four processors going to 100% utilization when each processor is rendering.

TW