Why does Pro Tools perform better when it is using fewer cores?

[meech]

I've been trying to figure out if I have the Playback Engine settings optimized ... and I've noticed that on my system Pro Tools uses less CPU and I get fewer related errors when I set the Host Processors setting to 3 out of 12 processors (I have a six core Mac Pro).

The most common advice I've seen here on the subject (and on Avid's own tech articles) is that a good starting point is to set it to one less processor than is available (for me that would be 11 out of 12). On my system it seems that Pro Tools can't take advantage of anywhere that number of processors.

For example, at 48khz, I can pull up 20 UBK-1 plugins (ten on an audio track through ten more on an aux) , and Pro Tools Native CPU meter hovers around 70%, and Pro Tools in Activity Monitor is around 225% cpu. This is when I select 3 out of 12 processors.

When I select 5 out of 12, Pro Tools still shows 70% on its own meter, but Activity Monitor shows Pro Tools using 325% cpu, with the exact same session! 7 out of 12 shows 76% PT meter, and 445% in Activity Monitor, and the more processors I select, the higher the cpu get until it gets jumpy and I start getting error messages.

It's almost as if when you allocate more CPU for Pro Tools, it immediately uses more CPU just for fun! The same exact session becomes less efficient, instead of showing less CPU and allowing you to pull up more plugins. It appears as though if Pro Tools took advantage of my CPU properly, I could set to the host processor setting to 9/12 or more, and triple my available plugin processing power! Why doesn't this seem to work?

I'm starting to wonder if I should move to another DAW such as Logic since I rely so heavily on VIs. Vienna Ensemble Pro sometimes works as a solution, but adds latency so it's not always ideal for tracking.

Does anyone have some insight into this setting? This is all with a buffer of 1024 and CPU usage limit of 99%. I noticed the same basic CPU patterns with different host processor settings with other buffer sizes and CPU usage limit settings. Does anyone else have a six core Mac Pro that sees similar performance? Do you notice better performance with more or fewer cores available for Pro Tools?

Thanks for the help :)

[albee1952]

I'm not a Mac user(and all cores seems best on my rig), but the "usual" setup is 1 core less than the total(now with hyperthreading, each core shows double, so I would start at 10 cores for PT). Experimentation to find the best performance is always a good idea

[Chief Technician]

I set the number of cores to half of the maximum. It works for me and the systems I maintain.

[meech]

Have you guys tried fewer cores and checked Activity Monitor? For me, fewer than half doesn't lower the CPU meter of Pro Tools, but drastically reduces the CPU of Pro Tools shown in the Activity Monitor ... but doesn't increase the number of plugins I can pull up without errors by much (maybe just a few more).

[Airton Silva]

When you increase the CPU share to Pro Tools it will for sure use more CPU, that's why the percentage increases at Activity Monitor. Pro Tools uses all the CPU you give him. It uses all the cores but decreases the load on each core when you add more cores.

[robo6675]

I think it may be more like the computer as a whole can run better with fewer cores dedicated to PT. background tasks need some of that CPU power and if you give it all to PT, things start to slow down.

[Mista Min]

Nehalem and Westmere hyper-threaded CPUs perform poorly when using hyper threading for audio apps. The hyper threading technology used is similar to the Pentium 4 HT and Pentium D.

New ivy and sandy bridge hyper threading is much better and doesn't have this problem.

[John_Toolbox]

Originally Posted by Mista Min

Nehalem and Westmere hyper-threaded CPUs perform poorly when using hyper threading for audio apps. The hyper threading technology used is similar to the Pentium 4 HT and Pentium D.

New ivy and sandy bridge hyper threading is much better and doesn't have this problem.

I have seen this posted online before, but it has not been my experience. Both my i7-860 and i7-920 systems work great with hyper threading enabled(Same architecture as nehalem/westmere). I'm not saying that these are faster or better than the newer i7 generations, just that I don't think the hyperthreading "limitations" really are as relevant to real world performance as some of the tech articles make them out to be. I think it's one of the hypotheticals like the difference between putting only 3 RAM sticks in a mac pro for triple channel performance, where if you use 4 sticks it reduces it's performance to 2-channel. Theoretically, there is better performance using a triple channel configuration, but in the real world more RAM(4 sticks equaling 16GB in dual channel mode) will always be better than more memory channels (3 sticks equaling 12GB in triple channel mode). YMMV.

[Bob Olhsson]

I've been told that this depends entirely on exactly how each application and plug-in has been coded in addition to the particular graphics processor and driver.

It would be great if there were a simple formula but apparently it's really a moving target.

[sunburst79]

Originally Posted by Bob Olhsson

I've been told that this depends entirely on exactly how each application and plug-in has been coded in addition to the particular graphics processor and driver.

It would be great if there were a simple formula but apparently it's really a moving target.

This is as good as any. And the original explanation was that some plugs follow the RTAS spec better than others. And that by leaving at least one whole core avaiable to processes outside protools its able to run the unoptimised code better. So you want to be one full core down in many cases.