View Full Version : CS5 & AVCHD native performance


Randy Johnson
August 11th, 2010, 04:55 PM
I have been playing around with Premiere CS5 and thinking of editing AVCHD native but when I do I am getting really poor performance. It freezes when I scrub sometimes I need to wait for like 30 seconds before it "wakes up" after scrubbing. Is this normal? I dont have a fermi card but im not using any effects on the clips. Just scrubbing in the time line and maybe a 2 camera multi-cam. I am running a 8 core mac pro (under windows 7 764) with 4 gigs of memory and a i-7 920 with 6 gigs of memory.

Khalil Ismail
August 22nd, 2010, 12:22 PM
I do not know much about Mac but this is not normal. I have PC with I7, GTX285 card and I am able to place 5 layers with PIP and CC on the timeline and it plays smooth as silk. I am sorry I am not much of help to solve your issue. Hope some does.

Peter Manojlovic
August 22nd, 2010, 08:24 PM
Go to the audio track, and mute the audio...
If the timeline scrubs correctly, i'm afraid you've stumbled upon the same bug i found a week ago...

Let us know.....
It's an easy fix.

Olakunle Olanrewaju
August 23rd, 2010, 12:09 PM
Go to the audio track, and mute the audio...
If the timeline scrubs correctly, i'm afraid you've stumbled upon the same bug i found a week ago...

Let us know.....
It's an easy fix.

I am interested in the 'fix'

Jay West
August 23rd, 2010, 02:48 PM
I have this problem intermittently running PPRo CS5 on an I7/12gbRAM/GTX260 under Win 7.

However, I wonder if I read your post correctly: are you running PPro under Win 7 under a Mac OS with only 4 or 6 gb of RAM? I would seriously consider adding more RAM.

Apart from that, I've found three things help avoid this on my Win 7 system.

1. Use the AVCHD sequence presets within PPro CS5 rather than HDV or the like. I run with two computer monitors plus an external HDTV fed via a Matrox MXO2 mini which requires using a Matrox preset for the external feed, and the Matrox presets seem intermittently vulnerable to AVCHD overload. When things bog down -- say, when I've got four multi-cam tracks plus titles and etc. --- I switch over to an AVCHD preset and the problem often (but not always) goes away. Sometimes, the problem just goes away if I reboot.

2. Turn off automatic updating, turn off background anti-virus/security scans (schedule them for when you are not working), and be sure that Windows has not secretly restarted Windows Defender. Windows update seems to randomly turn all these things back on whenever I allow an update. On my system, Defender seems particularly insidious, getting turned back on and getting double anti-virus scans going and what seem to be conflicts with Internet Security. When there is a big Internet Security download, it can also slow the system to an absolute crawl.

3. Divide your project into shorter sequences and nest the sequences for a final timeline. This became a habit for me under CS4. The thing I've noticed under CS5 is that most of the slowdowns happen when I've got a longer timeline (say 30 minutes) with many edits. The more edits I make, the more things seem to slow down when they are slowing down.

What is most annoying about this for me is that the bogging down is intermittent and almost random. I'll do two projects and not have any problems. The, on a third project, it starts almost immediately. Start over with a new project, and no problem.

Also, there may be memory leaks that cause these kinds of problems so, sometimes, shutting down and rebooting seems to fix things. I think Steve Kalle also recently posted something about memory leaks in CS5 so you might try searching for his post..

Peter Manojlovic
August 23rd, 2010, 03:01 PM
I don't know if this is the same issue as the original poster, but there's many of us users, who have found out that Multicam simply bogged the system. It's not an AVCHD thing.
Ironically, once you mute the audio track, the video runs fine...
The culprit seems to be those of us who use the option "Audio follows video" in the flyout on the Multicam screen...
For some reason, this just bogs down the system..Uncheck this option.

The fix would be to save out the audio as a separate .WAV file, and reimport it into the main sequence..

I don't know if this was addressed in CS5...
Only certain individuals seem to be having this problem..I am one of them..

Steve Kalle
August 23rd, 2010, 03:11 PM
Hey Jay,

My thread title was incorrect as Pete pointed out. I have 24GB of ram and Premiere is programmed to use up to 75% of available ram, which is why I was seeing 18GB being used during multi-cam edits and lots of scrubbing, which, I think, Premiere is caching the frames that were scrubbed.

With an 8 core computer, 4GB is wayyyyyy too little. For optimum performance, you need 2GB of ram per 'physical' core, and for the best performance, 4GB per core under CS5 really gets things moving. Adobe really kicks some a$$ with MPEG2 encoding speed - going from 2GB per core to 4GB cuts MPEG2 encoding time in half or more, which why I am adding at least 24GB more ram to have 48GB total for my 12 cores. An important caveat to this MPEG2 speedup is that MRQ is not enabled; so, I am waiting to see how much of a speedup occurs when having 4GB per core AND MRQ checked. For everything BUT web video, MRQ needs to be checked as it seriously increases scaling quality when decreasing pixel resolution.

Harm Millaard
August 23rd, 2010, 04:06 PM
Steve,

One very important correction or addition:

ALL encodes, rendering and scaling, with hardware MPE enabled are done with maximum quality and using linear color, so in essence the same as MRQ without MPE. One caveat: AME follows the PR settings, so if you have MPE enabled when closing PR, the maximum quality applies, if you closed PR with MPE off, you have to set MRQ manually.

Steve Kalle
August 23rd, 2010, 04:12 PM
Steve,

One very important correction or addition:

ALL encodes, rendering and scaling, with hardware MPE enabled are done with maximum quality and using linear color, so in essence the same as MRQ without MPE. One caveat: AME follows the PR settings, so if you have MPE enabled when closing PR, the maximum quality applies, if you closed PR with MPE off, you have to set MRQ manually.

This is not entirely correct as there is a major difference when MRQ is checked and not checked and hardware acceleration enabled on both.

Harm Millaard
August 23rd, 2010, 04:32 PM
Steve,

I forgot one important aspect. You are still on 5.01.

Randall Leong
August 25th, 2010, 01:39 AM
Steve,

I forgot one important aspect. You are still on 5.01.

That is correct. 5.02 has not been officially released in North America yet.

Jay West
August 25th, 2010, 02:12 AM
I don't know if this is the same issue as the original poster, but there's many of us users, who have found out that Multicam simply bogged the system. It's not an AVCHD thing.
Ironically, once you mute the audio track, the video runs fine...
The culprit seems to be those of us who use the option "Audio follows video" in the flyout on the Multicam screen...
Only certain individuals seem to be having this problem..I am one of them..

Hmmmm. I've never used "audio follows video." And, with CS5, I'm only seeing the bogging-down when I've got AVCHD tracks in my multi-cam projects (which is most of them since I got an NX5). I went back to a multi-cam project that had bogged down with 2 AVCHD and 2 HDV tracks and tried muting the audio tracks. Still had the same problem. Tried enabling "audio follow video." Still had the same problem. Tried removing the audio tracks. Same problem. Went back to a current project --- where I've actually got six video tracks (3 AVCHD and 3 HDV, put together in my multi-cam tracks plus two additional cut-away cams for different parts of the wedding venue) plus multiple audio tracks. Tried enabling audio follows video. No bogging down. Go figure.

Steve Kalle
September 2nd, 2010, 11:29 AM
Well, 5.02 has been released and MRQ is the same; so, what are you talking about Harm?

Harm Millaard
September 3rd, 2010, 01:56 AM
From the horses mouth:

Maximum render quality specifies that software should scale using a bicubic filter and blend in linear colour. For exports scaling and blending on the GPU will always process similar to software’s max quality, even without the flag set. Even with CUDA enabled there are cases where setting max quality can make a difference, namely any scaling or blending done not on the GPU but in software as part of an accelerated export. Setting maximum render quality will have no penalty and end up doing nothing if you do not fall into such a case. If you are seeing an increase of render time then you are falling into one such case and there will be a quality difference.

Olakunle Olanrewaju
September 3rd, 2010, 04:00 AM
From the horses mouth:

If you are seeing an increase of render time then you are falling into one such case and there will be a quality difference.

I think that answers my question. Thanks

Steve Kalle
September 3rd, 2010, 12:27 PM
Thank you Harm, I appreciate the detailed info. I only notice MRQ engaged when going from 1080/720 to 480/406p. It didn't engage when increasing size from 480x360 to 640x480.

So far, I have noticed 2 things different with 5.02: Premiere has a problem when opening a project and the screen flashes and goes into non-Aero mode; and exporting H264 seems to be about 20% faster (a video took 90mins on 5.01 and 75mins on 5.02).

Harm, I know that I sound like a jerk sometimes, but you are the only one here that I can challenge when it comes to computer tech - for example, there aren't many here who can recall what C64 and Amiga were, let alone using them.

Also, any chance you can run the BlackMagic-Design disk test and tell me the read/write numbers for your R30? I can email you a serial# for one of my Decklink Extreme 3D cards so you can download thru their website. I ask because I am trying to decide between 6 2TB drives in R5 and 12 1TB drives in R30 for my home workstation in addition to getting another Areca or 3ware as I already have a 3ware in that PC.

Also, it seems that there is some funky stuff going on with my Z800 as you can tell from the PPBM5 disk test. For some reason, synthetic tests show almost 500MB/s read & write for my 6 drive R5 and 1680ix but the PPBM disk test is no where near that fast.

Harm Millaard
September 3rd, 2010, 03:11 PM
Steve,

Mail me the number and I will do the test for you. You have my e-mail address.

BTW, my best memories from the old days were with APL on an IBM 5520, far better than Fortran, RPG or Cobol on a DEC 2060 or IBM 370.

Look here: http://www.sigapl.org/whyapl.htm

Randy Johnson
September 3rd, 2010, 03:14 PM
Well im not sure which way to go I have 6 gigs of ram I may go up to 12 thats a pretty cheap upgrade. Part of me would love to see what the performance would be if I just went for broke and got a 6 core gulftown with a GTX 460. I dont really do effects not filters anyway I just do single track editing with a mutli-cam time line per wedding. i-7 920 just doesnt seem to be up to the task yet.

Harm Millaard
September 3rd, 2010, 03:25 PM
Randy,

My suggestion would be to hang on a bit. You can easily go for 12 GB and that will give you a nice performance increase, but for the rest I would wait a bit. Around this time next year you will have the option to go for an octo-core Patsburg with DDR4-3500+ memory with 32 GB and SSD's with double the current capacity and a nice performance boost over current models for prices comparable to current prices. Now that will give a real boost and by that time CSNext will be imminent and MPE support will have left the infancy stage behind. There are nice perspectives around the corner IMHO.

Randy Johnson
September 3rd, 2010, 06:27 PM
you know I just did a test with the new update and it was better. It seems more memory would help also the cpu maxes on transitions like dissolves. would the MPE help here? Also theres a bug with black magic & the multi-cam feature but I work around that by disabling the BM driver during multi-cam parts.

Randy Johnson
September 3rd, 2010, 08:16 PM
I just ran the hack on my GTS-250 Is there a cut off of older slower cards where its not worth it to run in MPE? Where its close to or even slower than just CPU?

Randall Leong
September 3rd, 2010, 08:35 PM
I just ran the hack on my GTS-250 Is there a cut off of older slower cards where its not worth it to run in MPE? Where its close to or even slower than just CPU?

For most CPUs, this cutoff is just below the GT 240. But some of the GPUs as low as the DDR3 version of the GT 220 might be an improvement over software-only mode if the CPU just barely exceeds Adobe's minimum requirements.

Randy Johnson
September 3rd, 2010, 08:45 PM
Cool thanks, I have a i-7 920 and a GTS-250 it looks like the only transition thats accelerated is a dissolve. Which is fine for now because its the one I use the most. Now that I can do the hack a low end fermi card may be in my future.