Does Turbo boost on CPUs actually help with REAL-TIME audio?

[Max Oepen]

When trying to push latency down to its absolute minimum, good old-fashioned clock-cycles are still the key ingredient with audio. As has been noted on this forum before, multiple cores can help with parallel processing of multiple tracks, but from A/D to D/A, so-called "round-trip latency" is best cut down by the fastest processor.

Since Intel's Turbo Boost was introduced, what are people's experience with how it handles low latency audio? Since it is by design a reactive technology, I just don't see how it could be useful in minimizing latency. Wouldn't it be better to be able to "lock in" the highest possible speed? And if so, can that be done with these chips? My current laptop, purchased over four years ago, has a 3.06 GHz CPU. Why is it that today I can hardly find anything over 2.8 GHz?

Continuing to cut down on real-time audio latency is a huge concern for me personally, and my guess is it will become more popular as more people start fooling around with live-processing of their instruments like I do. What is everyone's pick if they had to buy the most blazingly fast machine (laptop)? Whether PC or Mac...

Secondly, how much does the type of operating system figure into this, if at all?
The fastest I've been able to tweak a setup is still on XP.

Thanks.

[fishmonkey]

max CPU speeds are limited by heat and power usage factors, that is partly why CPUs with more cores tend to have lower clock speeds. there is a pretty good explanation here:

http://www.anandtech.com/show/2832/4

generally speaking you can't peg Turbo Boost to it's max frequency indefinitely because the processor would overheat. in fact, Turbo Boost can only kick in when some of the cores are inactive, and the processor is cool enough (has enough thermal overhead available)...

[Max Oepen]

Thank you for the link.

[Max Oepen]

understanding the facts about core count vs clock speed, as well as other considerations such as on-dye cache size,
what would currently be the mobile processor of choice for lowest-latency real-time audio?

To limit variables, let's say this involves using only four tracks in Live with effects running on each.

[kitekrazy]

When trying to push latency down to its absolute minimum, good old-fashioned clock-cycles are still the key ingredient with audio. As has been noted on this forum before, multiple cores can help with parallel processing of multiple tracks, but from A/D to D/A, so-called "round-trip latency" is best cut down by the fastest processor.

Since Intel's Turbo Boost was introduced, what are people's experience with how it handles low latency audio? Since it is by design a reactive technology, I just don't see how it could be useful in minimizing latency. Wouldn't it be better to be able to "lock in" the highest possible speed? And if so, can that be done with these chips? My current laptop, purchased over four years ago, has a 3.06 GHz CPU. Why is it that today I can hardly find anything over 2.8 GHz?

Continuing to cut down on real-time audio latency is a huge concern for me personally, and my guess is it will become more popular as more people start fooling around with live-processing of their instruments like I do. What is everyone's pick if they had to buy the most blazingly fast machine (laptop)? Whether PC or Mac...

Secondly, how much does the type of operating system figure into this, if at all?
The fastest I've been able to tweak a setup is still on XP.

Thanks.

If you are relying on onboard sound you will probably not reach low latency. With laptops heat and power consumption are big factors when they are designed.

The Turbo Boost and Amd Overdrive are sort of marketing gimmicks. They boost CPU power 10-15%. It's doesn't always make a difference.

[kitekrazy]

understanding the facts about core count vs clock speed, as well as other considerations such as on-dye cache size,
what would currently be the mobile processor of choice for lowest-latency real-time audio?

To limit variables, let's say this involves using only four tracks in Live with effects running on each.

Effects are part of the variables.

[Max Oepen]

If you are relying on onboard sound you will probably not reach low latency. With laptops heat and power consumption are big factors when they are designed.

No, I am assuming the use of a high-end audio interface. I have an RME Fireface800.

Effects are part of the variables.

That's right! But as long as they sit on one track, Live processes them sequentially (as opposed to parallel), so multiple cores won't help on a single track.

Still looking for recommendations on mobile CPUs.

[3dot...]

understanding the facts about core count vs clock speed, as well as other considerations such as on-dye cache size,
what would currently be the mobile processor of choice for lowest-latency real-time audio?

To limit variables, let's say this involves using only four tracks in Live with effects running on each.

you're forgetting the return bus and the master
imo don't overthink it.. get the strongest/fastest for the $ you have

[H20nly]

understanding the facts about core count vs clock speed, as well as other considerations such as on-dye cache size,
what would currently be the mobile processor of choice for lowest-latency real-time audio?

To limit variables, let's say this involves using only four tracks in Live with effects running on each.

you're forgetting the return bus and the master
imo don't overthink it.. get the strongest/fastest for the $ you have

this is exactly what I would say about it too.

don't buy into gimmicks designed to suck gamers in. just work with something that is strong enough to handle the task at hand. trying to squeeze some additional performance out of today's processors begs the question; why do you need it?

a dual core 2.8 GHz processor is more powerful than a single core 3.x and so on as you add cores. the reason why the GHz drops is mostly due to heat. a single core 3.4 GHz P4 is hot son of a bitch... stacking two or 4 or them just makes a recipe for meltdown. that isn't to say it can't be done... it's a best practices vs. cost type scenario.

if you're using onboard audio... that is your weakest link and should be more of a focal point than squeezing a few more 10ths of a GHz rating out of a more than capable chip.

[XSIMan]

Boost only works if only one CPU core is in use unless intell changed this in the lastest I7 cpus. Live runs each track on its own core.

[Max Oepen]

why do you need it?

Believe me, I need it. I process my drumming in real-time with Live. And in my 6-7 year experience of constantly dealing with this, I have found out that clock-cycles do matter, almost above all else.

you're forgetting the return bus and the master

I am very aware of the latency-from-routing issues regarding the return & master buses. It is a limitation I work around...

I am surprised after all this time that not more people are using Live for this purpose. I've conversed with Zach Danziger (Mr. Barrington) about this, but he is searching for answers as much as I am.
Whoever does have experience with this first-hand, and knows about the advantages/drawbacks of the last couple generations of CPUs over the preceding ones, is who I need to hear from.

Also want to say thanks for your quick input, you guys are some of the most knowledgeable people on this forum, and I appreciate your 2 cents very much.

imo don't overthink it.. get the strongest/fastest for the $ you have

You're right that I may be in danger of over-thinking it. However, mobile processors are built into their respective machines, and I'm not looking to make a ~$2000-3000 mistake, I'd rather err on being over-educated on the subject. Making a living playing music limits my budget somewhat

[gnurf]

Fast cores are only part of the solution to responsive audio, but it of course depends on what you do. Real-time processing is largely CPU speed and memory speed (which isn't an issue with anything but massive kits). If you work off round-robin sample alternations and RAM can't hold it all you have another latency problem. But having more than one core alone helps a ton, if the host is able to make use of them. While one core is handling the effect you need it to focus on, the GUI remains responsive on a different core.

Since turbo boost increases the frequency of ONE core when the others are dormant, this isn't much use for responsiveness between multiple tasks. If you're using Live to ONLY play one thing there is a chance it could help, though. But the safest bet is more and faster cores :)

Unless you wrote the host software/collection plugin and can definitely guarantee it all works off one thread for both instruments and effects ;)

[kitekrazy]

If you are relying on onboard sound you will probably not reach low latency. With laptops heat and power consumption are big factors when they are designed.

No, I am assuming the use of a high-end audio interface. I have an RME Fireface800.

Effects are part of the variables.

That's right! But as long as they sit on one track, Live processes them sequentially (as opposed to parallel), so multiple cores won't help on a single track.

Still looking for recommendations on mobile CPUs.

What kind of latency are you getting out of your RME?

[Max Oepen]

What kind of latency are you getting out of your RME?

running a 48-sample buffer at 48kHz. Live's settings claim round-trip is "4.94 ms", but in truth I'm closer to 20 ms. I have measured this with CEntrance's LTU "latency test utility". Fluctuations depend on what/how many plugins used, as well as routing (as noted by others above).

Ideally, I'd like to get below 10 ms, that seems to be the "magic" number; equivalent to the 20-25 fps it takes for the human brain to perceive motion in film.

At 20 ms, I still have to make compromises. As an example, a dotted 8th-note delay above a certain tempo will start to "swing". I've been learning to embrace these limitations. After all, it is quite amazing what is now possible. However, if I could just shave off those extra 10 ms or so....

[mooncaine]

Ideally, I'd like to get below 10 ms, that seems to be the "magic" number; equivalent to the 20-25 fps it takes for the human brain to perceive motion in film.

Fascinating topic. I'll read along because I sometimes play guitar & guitar synth thru Live, and latency can be a spoiler (mainly because of the guitar rig, not Live, but there's a minimum latency from sending the audio through the computer via FireWire, USB, or whatever, too).

I wouldn't assume that the visual detection time is related to the aural detection time, but it's probably a good start.

I can feel 13ms when I try to play fast runs on a guitar whose signal path has some latency; well, at least I know something feels 'wrong'.... I pretty much decided I had to fall in love with slower music, in a way.