KVR Audio

headquest · Post by **headquest** » Tue Jun 27, 2006 2:54 pm

Thanks again!

Okay I redid the test in Sonar with dithering and 64-bit mix engine OFF. This time the results were actually similar to before: Overall CPU usage ranged from 22% - 32%, so still within the limits I reported in the OP above. Because of the range involved I could not obviously tell whether making those changes made a tangible difference.

I also ran the test in Ableton again and noticed the following regarding CPU meters:

Sonar's CPU meter was around 5% less than the WTM meter... but the Ableton CPU meter was showing significantly higher CPU use than the WTM. I guess this shows (among other things) that the WTM is more trustworthy in terms of making this sort of comparisson?

Any other comments would be much appreciated. At this stage I am kinda glad I did this test because the results are intertesting (although inconclusive). At least they seem to dispel the myth of Live being a CPU hog... for smaller projects at least (which in fairness is what I normally do myself) it clearly compares well.

ObsoleteAcc99 · Post by **ObsoleteAcc99** » Tue Jun 27, 2006 3:28 pm

[DELETED]

kuniklo · Post by **kuniklo** » Tue Jun 27, 2006 3:54 pm

ttoz wrote: 5.8ms is fine for my needs and many others' here. but yes the engine needs optimization.

Keep in mind though that Live's engine is subject to different constraints than the others. It has to be absolutely gapless for almost everything so it can be used *Live*, unlike Sonar's "gapless" engine that gaps out all over the place. I doubt Live will ever match Logic's or Sonar's efficiency because it's not doing the same work.

I'm glad to see Live becoming so popular and capable but I worry that people coming over from hosts like Logic are pushing to have all their favorite features ported over without considering carefully what makes Live appealing in the first place. Personally I'd rather see Ableton focus most on making Live an even better tool for composition and experimentation and performance and make it easier to dump the result as individual tracks for import into Pro Tools or Logic or Sonar for final mixing/mastering instead of adding all the high-end daw features where they really don't fit.

That said, some performance optimization and dual-core support and a more flexible freeze function would definitely be welcome.

headquest · Post by **headquest** » Tue Jun 27, 2006 4:00 pm

ttoz wrote: 256 is the magic number.

Yes, that's what I'm using in the test.

spaceman · Post by **spaceman** » Tue Jun 27, 2006 4:29 pm

kuniklo wrote:
ttoz wrote: 5.8ms is fine for my needs and many others' here. but yes the engine needs optimization.
Keep in mind though that Live's engine is subject to different constraints than the others. It has to be absolutely gapless for almost everything so it can be used *Live*, unlike Sonar's "gapless" engine that gaps out all over the place. I doubt Live will ever match Logic's or Sonar's efficiency because it's not doing the same work.

exactly.. Cakewalk must be using some exotic ancient definition of the word 'gap' because gapless in this reality is definitely is not. It doesn't have to be and Sonar is a great host but why they insist on calling the audio engine gapless is beyond me.. maybe the marketing guys have vetos

WillieJenkins · Post by **WillieJenkins** » Tue Jun 27, 2006 4:31 pm

This is a great thread.

Maybe I'll try creeping up to 256, I've been using 128. I'm using Live as my DJ tool, so I can't have ANY crackles, or else the dance floor will notice, so more stability is key for me.

Thx! I'll let you know what I find.

-Eric

whyterabbyt · Post by **whyterabbyt** » Tue Jun 27, 2006 5:09 pm

Well Ive just done some tests, and I think I sort of have an explanation of what headquest was seeing.

I loaded up a Sonar project then ran it in both single-processor and multi-processor mode (Im using an X2 Athlon 3800+ dual-core processor), and the CPU usage reported in the Windows Task Manager was pretty similar in both cases, varying around 32-35%... but the Sonar CPU meter showed a jump to 50% or more CPU usage when running single-processor. Most interesting.

So I found a CPU-usage application which reports the usage per CPU (or core)(*) and reran the test again, and that gave me some insight into what was happening. In multi-processor mode, the CPU usage was fairly closely even across both cores, and relatively close to the WTM figure. However, in single-processor mode, the CPU usage was shown to be quite heavy on one processor, and negligible on the other (other tasks on the system, no doubt). However averaging the two CPU usage percentages gave a figure pretty close to what WTM was reporting...

So I suspect what's happening is that the WTM is displaying the CPU usage as an average of all cores... which makes sense; its not reporting per core or processor, its reporting overall usage of the total system resources.

Thus, if there are two cores and one core is being taxed 60% by Live (and the other 10% by other stuff), the average usage is reported as 35% over both cores. And if both CPUs are being used by Sonar efficiently at (say) 35% usage each, then the same average processor usage will be reported.

And by implication, if WTM shows the Live task taking 45% of a dual-core's CPU, then its actually running one core at 90% and real close to flat-out...

(*) See http://www.nospaceleft.com/bicpu.html

headquest · Post by **headquest** » Tue Jun 27, 2006 7:06 pm

Thanks for that feedback (and doing the tests) - very interesting and I think you've hit the nail on the head. This would definitely explain the difference between the internal CPU meters and the WTM one.

Thanks again.

whyterabbyt · Post by **whyterabbyt** » Tue Jun 27, 2006 7:10 pm

headquest wrote:Thanks for that feedback (and doing the tests) - very interesting and I think you've hit the nail on the head. This would definitely explain the difference between the internal CPU meters and the WTM one.

Thanks again.

No probs.

headquest · Post by **headquest** » Tue Jun 27, 2006 7:56 pm

So to summarise (and check I've understood correctly) :

Sonar seems to (possibly) use slightly more CPU than Ableton overall (performing the test task), but it does spread that load efficiently across the two cores, so avoiding overload. This is going to be more particularly important in larger projects.

Ableton (possibly) uses a little less CPU overall - contrary to its reputation! - but because the full load is placed on one Core only, CPU overloads/etc are more likely with larger projects. Multicore support in version 6 should redress this.

These observations relate to running these programs on a dual core setup.

Architect · Post by **Architect** » Tue Jun 27, 2006 9:05 pm

WillieJenkins wrote:headquest,

I'm not trying to push Sonar over Live here because they're both my main hosts, I'm only posting this stuff, because I know from real world usage that Live has way less runway to play with compared to Sonar. I often start projects in Live, then re-create the exact same project in Sonar and go back and forth. 100% of the time The project will become unusable in Live before Sonar.

Just my .02

-Eric

Hey Eric what do you mean by "The project will become unusable in Live before Sonar"?

WillieJenkins · Post by **WillieJenkins** » Tue Jun 27, 2006 10:14 pm

So, lets say I'm working on a dance track, by default I'm going to have

1xMinimonsta
1xTrilogy
1xRMX
1xMicrotonic
1xBattery 2
1xKontakt 2
5x percussion loops
4-5x vocal tracks

So I record a bunch of midi clips, and record vocals, and I'm digging the groove, and my cpu usage is about 50% at this point. Often times I will re-create this template in Sonar, and just import my midi/audio clips, and do my heavy arranging / mixing in Sonar.

At this point, for me, the cpu usage is ALWAYS lower, but beyond that, lets say I add 10 (4 compressors, 4 eqs, 1 reverb, 1 limiter) effects, pushing the sonar CPU meter up to about 80%.

If I were to then try and go into Ableton Live and add those same effects I had in Sonar, Live would be unable to play this project back, as the extra plugs I used in Sonar add too much CPU for live to handle.

It appears that headquests tests may refute this, but I really wish I could replicate his results, as that extra cpu would be handy on my powerbook

-Eric

headquest · Post by **headquest** » Wed Jun 28, 2006 8:12 am

WillieJenkins wrote: It appears that headquests tests may refute this, but I really wish I could replicate his results, as that extra cpu would be handy on my powerbook

-Eric

Eric, if your music PC is dual core then your own experiences look to me like they tie in with my test result and whyterabbyt's findings. If your PC is single core then there does seem to be a contrast there, which no doubt highlights how system-specific these types of results can sometimes be.

However, there's one more "wild card" in your example, which is that you seem to use a lot of RAM-hungry plugs. For example Trilogy loads every patch directly into RAM, and each patch can be 128MB. RMX is far worse than that. Once you get low on RAM, the CPU will struggle to keep up with everything - especially in real time (i.e. in Live). The solution is to freeze just those tracks that are RAM-intensive. In your example freezing the RMX and Trilogy will probably solve your problem at a stroke.

Another variable introduced in your example is Live's warping, which can use a fair amount of CPU grunt, especially if you select "Complex" mode. If your percussion loops are audio and you are warping them (which presumably you are by default if they are "loops") that will be using CPU. (Make sure they are set to "Beats" mode for warping by the way, if you didn't already know that).
Your vocal tracks presumably don't require looping or warping, so be sure to switch that off.

I hope that helps... even if it doesn't exactly answer the question! As was said before, it's hard to directly compare sequencers' CPU consumption because there are too many variables. That is why I kept my initial test so very simple (to minimise variables). Both are excellent programs which I regularly use, and CPU is not a significant issue for me in either. But I remain keen to see how Dual Core support in Live 6 will affect things.

advaya · Post by **advaya** » Wed Jun 28, 2006 9:02 am

Interesting thread headquest... I've often found that trying to determine a host's efficiency is a bit of a grey area...

When I first tried S5 with my dual-core athlon, I was kinda shocked to see the high CPU usage reported by both it's own meter, as well as WTM... One instance of ZeroVector (known to be a bit of a CPU killer) was shown to be using 26% of the CPU when I held down a 4-note chord. This was quite odd, because doing this in SX (also multi-processor aware), the CPU meter only showed roughly 10%.

Common sense told me that I'd not be able to run more than three instances of ZV without bringing my system to it's knees. But to test further, I opened up multiple instances of ZV, and had them all play 4-note chords simultaneously. The results were kinda strange.

Two instances of ZV playing at the same time, the CPU meter showed 40% (not the 52% I had expected). Three instances = 47%... Four instances = 50%...

I had a total of 7 ZVs all playing 4-note chords simultaneously, and the CPU meter leveled out at around 70% (pretty good for such a CPU killer synth)...

Makes me wonder about how apps 'scale' the reported CPU usage. Again, I found Cubase to be quite different from Sonar... Each instance of ZV (same patch, same 4-note chords) showed roughly 10% CPU usage. And as expected, seven instances of ZV in SX reported back roughly 70%...

Not sure what all this means, but it was an interesting finding none-the-less...

spuddle · Post by **spuddle** » Wed Jun 28, 2006 11:52 am

Any one feel that Ableton should work on a product that is firmly based around MIDI sequencing and video scoring?

Ableton vs. SONAR - CPU Shock!!