Mac Pro 8 core

Post your testing results with HandBrake.
davidahn
Posts: 7
Joined: Tue May 22, 2007 5:25 pm

Something to try?

Post by davidahn »

You may want to try 320x240 at Baseline profile (rather than Main profile) set up manually instead? Though I suspect it's not just the preset but the baseline profile.

I've been doing X x 320 with the baseline profile for my iPhone, and it's pretty fast on my quad G5, but I haven't had a chance to try it yet on the Mac Pro.

David

baggss
Moderator
Posts: 886
Joined: Tue Jan 02, 2007 8:21 am

Post by baggss »

I recommend going to the main page and pulling the latest svn and building it. You'll get much better performance than you will from 8.5B1 or even the svn from a month ago.

Monstermac
Posts: 5
Joined: Sat May 05, 2007 9:49 pm

Screamin Update

Post by Monstermac »

Wow, mad props to the development team... I just encoded Dodgeball using the iPod setting. This would normally have encoded at 22fps, but with the new build it was 90fps! Sweet!

Fredster
Posts: 11
Joined: Mon Mar 26, 2007 9:24 pm

Re: Screamin Update

Post by Fredster »

Monstermac wrote:Wow, mad props to the development team... I just encoded Dodgeball using the iPod setting. This would normally have encoded at 22fps, but with the new build it was 90fps! Sweet!
Now imagine when the 45NM Nehalem 8 core is out, and double that on a Mac Pro! :shock: Sweeeet!:)

Number6
Posts: 2
Joined: Sun Aug 12, 2007 8:25 pm

Post by Number6 »

I'm a Mac Pro owner and I'm not getting full CPU utilisation. This has seemed to fluctuate over the last few SVN releases, but I've never seen Handbrake use all 8 cores! Here's what I'm getting:

Image

and I see this CPU usage:


Image

or

Image

Almost always, it will just peg one core and not run on all eight.

Cheers,
Ron

eddyg
Veteran User
Posts: 798
Joined: Mon Apr 23, 2007 3:34 am

Post by eddyg »

Hi Ron,

I bet you used "make" to compile and not "./jam" right? Make is using the precompiled libs, and they haven't been recompiled to include ffmpeg multithreading until today.

If you recompile today you should get a different story, I get 300fps on my Mac Pro 4 core.

Cheers, Ed.

Number6
Posts: 2
Joined: Sun Aug 12, 2007 8:25 pm

Post by Number6 »

I trashed my source tree and re-SVNed a new one. Then I used jam to do a complete build of all the contribs and Handbrake. No go. I'm still getting this:

Image

I'm off to do some digging in the development forums...

eddyg
Veteran User
Posts: 798
Joined: Mon Apr 23, 2007 3:34 am

Post by eddyg »

Using ffmpeg at 1500 abr, I get 350% CPU on a 2.66Ghz 4 core mac pro - which gives me round about 200-300fps.

So there is room for improvement, however I see nothing like as low frame rate as you are seeing. I assume you are reading from a ripped version of the DVD on HD right?

Try a chapter using one of the presets, say iPod Low and see what you get.

Cheers, Ed.

eddyg
Veteran User
Posts: 798
Joined: Mon Apr 23, 2007 3:34 am

Post by eddyg »

Actually - maybe you are reaching the limit of what the pipeline can handle in that configuration. Given that on a 4 core I had reached the limit, you probably wouldn't see any improvement on what I am seeing.

I need to do some further performance enhancements. I think I'll have the pipeline tuned according to the number of cores. We've reached a stage where one size no longer fits all. Not when you have a 1Ghz G4 at one end and an 8 core 3Ghz Intel at the other end.

Cheers, Ed.

remyhelsinki
Posts: 23
Joined: Mon Apr 16, 2007 1:35 am

Post by remyhelsinki »

Just in the process of running a test to see how this Mac Pro 3.0 Ghz Dual Quad core can do (5GB of RAM).

My source is the dvd "National Treasure." I ripped with MTR onto my HD. This test was first run in HB .91 using the iPod High-Rez preset (changing it to 2-pass with a turbo first pass). The second test is using the standard Deux Six Quatre Preset.

iPod High-Rez (changed to 2 pass) - 640 x 272 px

1-pass (Turbo): 342.07 avg fps
2-pass: 128.89 avg fps

Deux Six Quatre - 853 x 358 px (anamorphic)

1-pass (Turbo): 203.27 avg fps
2-pass: 61.53 avg fps

Note that I was watching a movie in Quicktime during these encodes. However I never hit 100% processor load over all 8 cores according to MenuMeters. I would assume if I were to only run HB after a fresh restart I would get slightly higher encode fps, probably a trivial difference though.

eddyg
Veteran User
Posts: 798
Joined: Mon Apr 23, 2007 3:34 am

Post by eddyg »

I had a look at HB using Shark. And I saw that the x264 library spawns a number of processing threads that are connected via semaphores. If any of these threads can not run, e.g. because some other program is using the CPU then they all stall.

Therefore running other programs has a proportionally bigger impact on HBs speed than maybe expected, especially on multi-core computers, as all processing stops on all cores every time just one of the cores is used for something else.

In my case I was running a temperature monitor program that was polling every 2 seconds. Every time it ran it took up two CPUs for a period of time. HB totally stopped processing during this period, the other cores were idle.

Cheers, Ed.

remyhelsinki
Posts: 23
Joined: Mon Apr 16, 2007 1:35 am

Post by remyhelsinki »

eddyg wrote:I had a look at HB using Shark. And I saw that the x264 library spawns a number of processing threads that are connected via semaphores. If any of these threads can not run, e.g. because some other program is using the CPU then they all stall.

Therefore running other programs has a proportionally bigger impact on HBs speed than maybe expected, especially on multi-core computers, as all processing stops on all cores every time just one of the cores is used for something else.

In my case I was running a temperature monitor program that was polling every 2 seconds. Every time it ran it took up two CPUs for a period of time. HB totally stopped processing during this period, the other cores were idle.

Cheers, Ed.
That's interesting to know. Thanks eddyg. I may try running these tests again without anything running after a restart, however if I do then I wouldn't run as long of a sample (would only do a few chapters instead of the whole movie).

Of course my first results are probably more representative of what I would usually do because I'm either encoding while I'm doing something else that doesn't need a lot of processor usage, or I'm asleep (some of my apps stay open almost all the time though even then).

--------------------

Update: I just tried another movie (The Bourne Identity) without any apps running after a fresh restart. Only used the iPod High-Rez (with 2 pass) this time though.

iPod High-Rez (changed to 2 pass) - 640 x 272 px

1-pass (Turbo): 371.21 avg fps
2-pass: 122.37 avg fps*

*On the second pass I had to start working on some photoshop files. The second pass had been going for a good amount of time averaging around 122 fps before I had to open Photoshop and do some work. The final result was 122.37 avg fps so it really didn't change from what I was seeing before I started opening other apps besides HB.

I'm still not seeing a significant performance hit from running other apps and encoding using HB though at least with my limited tests.

Post Reply