rendering and memory

Started by scott8933, August 31, 2009, 10:53:39 PM

Previous topic - Next topic

scott8933

An oddball question actually:

I recently got an Aspire-One netbook and naturally one of the first apps I installed on it was Terragen 2; For the sake of amusement, I ran a head-to-head against my aging desktop. The Acer has the Intel Atom 1.6 ghz proc, and 1gb ram. The desktop has a 3.4ghz P4, with 4gigs installed ram, 3 gigs "active" under XP (using the 3 gig switch).

The two scenes actually start out pretty much even. Roughly, dividing the screen into quads, they both run through the first 3/4's about the same speed. But hitting the last quadrant, all of a sudden the Acer just grinds to a stop. End times are about 8 minutes on the P4 and 50 minutes on the Aspire.

So I'm curious - that last quadrant has about the same amount of visible geometry as the adjacent quadrant of the screen, so it's not like its hitting an especially complicated part of the render. Is it running out of ram at a certain point? FWIW, the Aspire can be upped to 2 gigs for pretty cheap. Not that I'd use it as my only render machine. I was just surprised that it stayed neck and neck, then all of a sudden took a huge nosedive in render speed. Also, another oddity I noticed was that when I rendered with a beta (1.9.04.1) it finished in 45 minutes. Rendering the same scene under the current version (2.0.3.1) it increased to about an hour.

Any thoughts?

Hetzen

#1
That benchmark test, (asuming that's the scene you're using) always grinds down when you get to the chrome ball and water area due to all the reflection calcs.

R3igN

hi

I always get the blue screen error when i render..

My damn computer always crashes when i render images bigger than 600 pixels..

Why  is that? Is terragen bugged or something?

thnx

scott8933

I didn't even realize there was a benchmark scene; for my test I just opened a blank TG project, added an Alpine Terrain fractal terrain, rotated the camera a few degrees for a better view and hit "Full Render" at the default settings.

So the scene roughly rendered in 4 sections, more or less - the top two being mostly atmosphere and the bottom two being mainly mountains. Both machines got through the first three sections pretty fast, it was the last 1/4 that bogged down the slower computer. Just wondering if adding an additional gig of ram would change that, or if its more a factor of it being a slower processor.



Quote from: Hetzen on September 01, 2009, 04:11:42 AM
That benchmark test, (asuming that's the scene you're using) always grinds down when you get to the chrome ball and water area due to all the reflection calcs.

Oshyan

The only way to check would be to monitor TG2 memory usage and total free memory while rendering. Also if you have a hard drive (as opposed to a solid state disk) then you may hear "disk thrashing" if it's having to use the hard drive for swapping virtual memory. RAM is so cheap anyway, it's worth having just for multitasking IMO.

By the way, how many cores does TG2 detect on the Atom netbook? It should be dual core if I remember correctly. If it's not detecting that, try forcing it to 2 threads. Also check the task manager for CPU use. Does it show 2 CPU graphs?

- Oshyan

scott8933

#5
Hmmm. Retested it, and payed closer attention this time to HD usage and cores/threads:

1 core detected, set it to minimum 2 threads. No change in render time, also its not hitting the HD much during the render. Almost not at all - minimum disk activity, all could be accounted for by the few remaining background tasks that I let stay open, i.e. just the obvious system ones. Everything else I had turned off, and also had a utility that was forcing it to stay at 800mhz instead of dynamic (running on battery power right now).

Same results: about 1 hr render time.

Oh well. Conclusion: its no speed demon. But then its not meant to be, it does get about 5 or 6 hours battery run time, and I've heard of some aftermarket batteries that bump it up a few more hours even more.

What's confusing me is how an earlier test with one of the tech previews clocked in 15 minutes faster. Makes me wonder if I've got a rogue process in there somewhere slowing things down some. The Acer shipped very light on the pre-installed junk, almost nothing really (especially compared to the insane bloat I've seen on other laptops like my Sony, which took a couple hours just to rid of the excess "value added" stuff).

But for sure another gig of ram will be getting ordered soon. No reason not to; and I'd even be tempted to try an SSD once the prices come down a little.

Edit: Some quick testing since I already had it open & running... Under TG 2031, minimum threads still set to 2, only other change was to set the "Size of subdiv cache in Mb" from 400 to 800. Also I reset the CPU control panel to "Dynamic" while under battery power (which it was). But this simply allows cpu throttling, so if anything it would end up slower than 800mhz but I'm not sure how the control panel correlates CPU demand with throttling. Presumably it stayed at 800mhz during the render.

Anyway, the major change was just the memory setting in the render box - and this brought the render down to 31 minutes. I guess some more testing would be in order, but now we're getting into a whole new thread probably well covered somewhere else.



Quote from: Oshyan on September 01, 2009, 10:38:17 PM
The only way to check would be to monitor TG2 memory usage and total free memory while rendering. Also if you have a hard drive (as opposed to a solid state disk) then you may hear "disk thrashing" if it's having to use the hard drive for swapping virtual memory. RAM is so cheap anyway, it's worth having just for multitasking IMO.

By the way, how many cores does TG2 detect on the Atom netbook? It should be dual core if I remember correctly. If it's not detecting that, try forcing it to 2 threads. Also check the task manager for CPU use. Does it show 2 CPU graphs?

- Oshyan

Oshyan

Interesting tests. As for older versions being faster, in some cases that's true. This is largely due to rendering issues or inaccuracies in older versions that have since been corrected. In general performance improves version to version as optimizations are made, but sometimes there are issues that need correcting, and the fix can slow things down in some cases. I believe you may be seeing the results of some work on GI or clouds (I don't remember the details at the moment ;D).

- Oshyan