Have you seen this behavior in CPU usage?

larry-peter wrote on 12/15/2011, 10:30 AM

Excuse the long post, but I haven't seen this addressed before and wondered if this was particular to my system.

I mentioned in an earlier post that I found I was getting better playback performance in Vegas 10 when a track FX was applied (in particular CC or Color Curves). I have finally installed Vegas 11 and immediately noticed the same thing.
I do not have a GPU that supports acceleration, so I realize this may all be moot when I get one. But this behavior is identical in Vegas 10 on my system.

Source media is AF-100 AVCHD 1080p 23.97. Monitoring through AJA Kona. Preview RAM set at 2 gig (6gig total in system). Monitoring at Best/Full.

When playing one long clip with no FX applied, it will play at full frame rate for about 45 seconds - CPU usage at 12%, 4 cores active, RAM usage 2.5 gig - then drop off to 21 fps as RAM usage slowly increases, CPU use remains the same.

If I apply a CC to the track, active but with nothing changed from default, it will play full frame rate indefinitely. CPU usage 15%, 4 cores show same activity but now 3-4% activity on the 4 other cores, RAM usage 2.95 gig and never increases.

So I started doing this with multiple video tracks simultaneously all at 50% opacity. With 2 tracks - no FX applied - I get full frame rate best/full indefinitely. CPU at 48% with all cores sharing equal load. RAM remains at 3.1gig used. Apply track CC and playback is full frame rate indefinitely, CPU at 53% all cores equal, RAM at 3.2 gig.

With 3 tracks playing simultaneously, things happened the way I expected. With no FX, frame rate dropped to 22fps. When I applied track CC, frame rate dropped to 19fps.

So, is it best practice to put a load on the CPU to the point where you're forcing all the cores to work? Obviously - at least on my system - there's a minimal point where Vegas is not sharing the load to all cores.

Anybody else seen this?

Larry

Comments

ingvarai wrote on 12/15/2011, 1:51 PM

>Anybody else seen this?

I am not quite sure what you have seen, actually..
On what kind of device is ur media stored? Do you restart Vegas for each experiment? There is certainly some chaching going on, somewhere.

larry-peter wrote on 12/15/2011, 3:13 PM

To be brief, (and I thought I was being clear before, but perhaps not) I see that there is a point, at least on my system, where if you make the CPU work hard enough to engage all processing cores, performance actually increases even though more media or effects are being used. The fact that a single clip with no FX applied would lose fps over time but a ten minute cross dissolve (50% opacity) would play back perfectly or a single clip with CC applied would play back perfectly seems very odd.

My storage is a G-Speed 2TB RAID0. I reproduced this consistently both with restarts and without restarts. The reason I was using 10 min. clips was so that any caching would be a minimal issue. I would not restart play from a previously played section of the clip. And I could watch the performance drop when playing a single clip and REMOVE the CC fx from the track. 4 cores would go dormant and RAM usage would increase.

ritsmer wrote on 12/15/2011, 4:18 PM

atom12: what you have found sounds very interresting.
Right now it is late in this part of the world - but I will test it tomorrow and as you have explained.

I have seen and reported similiar strange things - but most often nobody understand it - or they are not interested in testing it on their machines.

Will be back tomorrow, first thing.

dxdy wrote on 12/15/2011, 4:30 PM

I agree that some unexplained things can happen in this arena.

Shortly after starting to use VP 11, I installed a GPU monitor, GPU Caps Viewer, to monitor GPU usage. I found on my i7-950 that my Nvidia 560ti would run under 25% during renders. (Win 7 64 bit)

One day I dragged the GPU monitor window off my primary monitor and onto my secondary monitor. GPU usage leaped up into the 50% range and stayed there for the duration of the render.

I haven't been able to duplicate that performance, although I tried.

larry-peter wrote on 12/15/2011, 4:39 PM

Thanks to all replies and for looking into it. I also want to confirm that I was not using copies of the same clip when adding new video tracks. All were unique ten minute clips.

And I haven't used proper terminology re "cores". If you look at my specs, I obviously have a quad-core showing 8 from hyperthreading.

ritsmer wrote on 12/16/2011, 3:26 AM

As promised I did the tests on my machine (Mac Pro 2 x Xeon Quads, Win 7 64 bit Ultimate, Vegas 11 B425) and found that adding or removing GPU assisted Color correction did not have any noticeable impact on preview speed nor on CPU or GPU usage nor on number of cores used.
Likewise there were absolutely no change around the mentioned 45 seconds mark (nor at any other time).
Strange.

larry-peter wrote on 12/16/2011, 9:19 AM

Thank for testing. I don't have a GPU that supports acceleration, so what impact that may have is unknown. Perhaps it's just the configuration and CPU speed of my system that causes the addition of processing threads to fall precisely at the point between playing one AVCHD clip and the addition of a track effect. System is several steps behind the optimum today.

I've been reproducing this for months without looking at what the CPUs and RAM were actually doing. It hasn't presented a problem, just been a head-scratcher.

Again, thanks.

Rich Parry wrote on 12/16/2011, 10:18 AM

This may have absolutely nothing to do with your post, nevertheless, I thought I would mention that my machine has 24 cores (dual Xeon 12 cores each) and performance is terrible (i.e., render, preview, etc.). This is especially true with VP11.

I know I have not given details so this post may be useless, but I have spent many hours doing benchmarks and it seems to me, VPxx doesn't like a lot of cores (i.e., 24).

Again, apologies for lack of detail, but I am still pretty upset over bugs in VP11 and error messages I get.
Rich