DAW Latency Factors. Facts?

[theblue1]

Originally Posted by jazzymike

This is a purely hypothetical question as no 2 systems are alike.
Let's take a closed DAW-system such as an Apple mac and for simplicities sake use the CoreAudio protocol with the built in card and not introduce 3rd party solutions/drivers or get into the USB/FW war.

Let's further assume that we have 3 similar systems such as 3 MBPs or 3 iMacs (They usually release 3 from the same line for different price segments with upgrade options) that only differ in CPU clock speeds and maybe the ability for one of the CPUs to overclock.

So all things being equal, will the highest clocked CPU deliver the lowest latency?

If so, let's introduce another variable.
one CPU being a quad-core i7 with 2.0 GHz and the other CPU a quad core i5 with 3.3
which would win?

Any realtime FX plugs in the monitor cue? That would cause some additional differentiation from slower CPU to faster, possibly. And, of course, the ADC/DAC processes will account for roughly 1.5 to 2 ms of your total time, no matter what converters you're using or what data bus they're on.

Also, keep in mind that results might be different with a different OS. OS X has traditionally had issues when trying to deliver super low latencies, as can be seen in benchmarks like these: DAW Bench : DAW Performance Benchmarking

[jazzymike]

Originally Posted by theblue1

Any realtime FX plugs in the monitor cue? That would cause some additional differentiation from slower CPU to faster, possibly. And, of course, the ADC/DAC processes will account for roughly 1.5 to 2 ms of your total time, no matter what converters you're using or what data bus they're on.

Also, keep in mind that results might be different with a different OS. OS X has traditionally had issues when trying to deliver super low latencies, as can be seen in benchmarks like these: DAW Bench : DAW Performance Benchmarking

Again for simplicities sake let's take all non-native hardware factors out of the equation. No FX in the way, and let's treat the DA latency as if it were uniform or non-existent across all systems in this setup.

I'm sure there's a possibility of an OS other than OSX delivering lower latencies, but that would require at least 3 test specimens custom built with the same components which is of course doable but even more complicated than our hypothetical setup of 3 MBPs or iMacs or MacPros.

[jcschild]

Originally Posted by jazzymike

This is a purely hypothetical question as no 2 systems are alike.
Let's take a closed DAW-system such as an Apple mac and for simplicities sake use the CoreAudio protocol with the built in card and not introduce 3rd party solutions/drivers or get into the USB/FW war.

Let's further assume that we have 3 similar systems such as 3 MBPs or 3 iMacs (They usually release 3 from the same line for different price segments with upgrade options) that only differ in CPU clock speeds and maybe the ability for one of the CPUs to overclock.

So all things being equal, will the highest clocked CPU deliver the lowest latency?

If so, let's introduce another variable.
one CPU being a quad-core i7 with 2.0 GHz and the other CPU a quad core i5 with 3.3
which would win?

yes GHZ is still king assuming you are even hitting the CPU hard.

3 identical subsystems systems would all do the same low latency with 1 instance of a sample and 10 tracks audio even if the processors were 2.4GHz 2.8GHz and 3.3GHz.

now lets say 60 tracks audio and 40 instrument tracks (samples) with mulitple instances, intruments and software.

the 3.3GHz would be fine @ low lantecy the 2.8GHz would need higher buffer and posible bouncing and the 2.4 would need a very high buffer and a lot of bouncing.

again this has to be kept within the same processor architecture.

this is confusing

one CPU being a quad-core i7 with 2.0 GHz and the other CPU a quad core i5 with 3.3
which would win?

there are no 2GHz Quad i7 (or i5) unless you are talking previous gen laptop and there are no Quad 3.3GHz i5
but yes GHZ will win. a few yrs ago a faster dual core would out perform a slower Quad but thats changed now.

Scott
ADK

[JalenRawley]

Originally Posted by jcschild

there are no 2GHz Quad i7 (or i5) unless you are talking previous gen laptop and there are no Quad 3.3GHz i5
but yes GHZ will win. a few yrs ago a faster dual core would out perform a slower Quad but thats changed now.

Scott
ADK

I think he might mean the mobile version of the i7? My new machine has an i7 2670qm running 2.2ghz with turbo boost up to 3.1ghz.

[Jason West]

Originally Posted by jazzymike

So all things being equal, will the highest clocked CPU deliver the lowest latency?

The most important variable is the workload.
A workload of 96kHz is different than 44.1kHz.
A workload of 100 simulataneous tracks is different from 10 tracks.
A computation workload of adjusting the gain by 1 dB is different than the computations of a convolution reverb.
Basically, the workload is the total amount of data and what processing you're doing to that data.

If the total workload is "light" enough, it can be handled by all cpu speeds you mentioned, both desktop and mobile to deliver the same latency. The lowest DAW buffer you can set is 32 samples. Because of this minimum limit, this negates any speed advantages of faster processors if the slower processors can get all their computations done within that 32 sample window. Therefore, a future 10.0TeraHertz supercomputer with liquid nitrogen cooling towers could process raw data faster BUT that speed is rendered irrelevant by the 32 sample buffer.

If a cpu is clocked at 2.0GHz, that's saying it cycles its state 2 billion times per second. To simplify discussions (leaving out subtleties of cpu architectures and memory fetches), let's say it executes 1 instruction per cycle. When you set the buffer to 32 samples, another way of thinking about it as that you're telling the DAW to limit its ongoing processing window to 32/44100th of time (0.7 ms) which is enough to complete about 1.4 million cpu instructions. (2,000,000,000 * 32 / 44100 == 1,451,247).

Is a 0.7 millisecond window of 1.4 million instructions enough to calculate 1 dB of gain added to 2 tracks? Yes. Is it enough cpu breathing room to add convolution reverb to 100 tracks? No. You have to adjust the DAW to a higher buffer (give it a bigger window of instructions -- say 2.8 million cpu instructions, or more) ....or... get a faster computer that can complete more instructions in the same 0.7ms of time.

[jazzymike]

Thank you theblue1, Scott, JalenRawley and Jason West.
That was enlightening and precise!
Thank you very much!

[jazzymike]

One more question, if you look at the current lineup of Mac Minis.
We have a:

• 2.3 GHz Core i5 (I5-2415M)
• 2.5 GHz Core i5 (I5-2520M)
• 2.7 GHz Core i7 (I7-2620M)
• 2.0 GHz Core i7 (I7-2635QM)

Now it's quite self explanatory, that the Quad Core 2.0 i7 is the most "powerful" CPU on this list and will be put to good use by a DAW such as logic (HT, 2.9 TB). The Benchmarks speak for themselves and so on.
But what if we were to look at it from the perspective of a secondary computer such as a Vienna Ensemble Pro slave designed to run 16 tracks of samples only.
Wouldn't the 2.7 i7 or even 2.5 i5 be more suited in terms of delivering 16 channels with lower latency than a super powerful 2.0 GHz Quad Core?

I appreciate your opinion on this one!