MachFive and the DP8 Performance Meter (Solved)

[bayswater]

I'd appreciate some help understanding what's going on with MachFive 3.12 performance in DP 8.01. I'm running a simple sequence with bass and drum MIDI tracks in DP playing the Star Drums, and a Jazz Bass from the Universal Sounds in MachFive.

The audio performance meter in DP is peaking. The buffer size in DP is 512. I noticed if I change the buffer size up or down, the peaks disappear for about 15-30 seconds, and then slowly come back. Doesn't matter whether the buffer is 128 or 1024. I tried various combinations of M5 preload time and ring buffer size, and these make no noticeable differences.

I thought it might be a disk problem. I'm using an internal drive partition that holds only samples. Black Magic shows read and write speeds of about 100Mb / Sec for this drive. If the samples come from a FW800 drive that also makes no difference. (The FW800 drive has read speeds about half of that for the internal drive). Vm_stat shows no page swapping going on while the sequence is playing, so it doesn't look like the disk should be a problem for two instruments. I also tried turning disk streaming off on both tracks, and this also made no difference. (I assumed this would cause all samples to be loaded into RAM and stop any disk activity -- is this true?)

With streaming off Activity Monitor shows DP is using about 3G of RAM. (I have 16G). M5 itself shows it is using about 2G. Contrary to the DP Audio Performance meter, Activity Monitor shows that this whole thing is using about 12% of CPU capacity. M5 says it is using less than 10%.

Can anyone suggest what is causing the audio performance meter spikes? (I saw the advice to erase DP prefs, but this also made no difference).

[wvandyck]

On the 27" iMac?

My #s with 2 tracks: M5 3 Star Drums and J Bass.
Streaming Off
Act Mon Memory: 5.1 GB
I/0 1024 % cpu: 26
I/0 128 % cpu: 41


Streaming On
Act Mon Memory: 1.9 GB
I/0 1024 % cpu: 27
I/0 128 % cpu: 43

No performance meter peaking in either case. But I gotta tell ya, it took forever for DP to come back online after each I/O change and when activating/deactivating Streaming. Like about 120 seconds per instance. I don't think I'm being dramatic in calling that an eternity when that time frame in another DAW is 15 seconds or less.

2010 i5 MBP 8GB, 2x 7200 rpm drives (samples on the second internal drive via OWC Data Doubler)

[bayswater]

Thanks for this.

Yes, this is on the 27" iMac i7 quad core. Not the fastest Mac around but comparable to the earlier Mac Pros, and should be able to deal with this. Turning streaming on and off is taking about 10-15 seconds.

I tried the same setup but with M5 standalone linked to DP with IAC and Soundflower. Overall CPU use is about the same, but almost no Audio Performance meter movement in DP. CPU usage in M5 is still around 10%, and overall in Activity Monitor is about 15% at most. For a while, things sounded right, but after a while, I started getting digital distortion again. Right about the time I was ready to print the tracks in DP.

I noticed just before shutting down that the sustains on some of the Star Drums are very long. The ride goes on for a good 15-20 seconds. On top of that there are a lot of stacked layers. Maybe I need a lot more polyphony on a track with a 16th note ride.

What is your polyphony setting?

[bayswater]

The problem was solved by increasing polyphony, turning disk streaming off, using M5 only as a plugin, and ignoring the DP Audio Performance meter. The Star Drums presets have polyphony set to 10, odd for a drum VI that has things like a ride cymbal with a 30 second envelop, and nearly 2,000 keymaps. Setting polyphony to about 65 with no disk streaming uses about 2G RAM and is good enough for a simple drum track.

Solving this was complicated by the use of IAC and Soundflower. Even with the above settings, noise and dropouts were increasing after a few cycles through a MIDI track using standalone. Shutting down and starting up again made the problem go away, but it came back after a few minutes. Maybe the Soundflower/M5 combination has timing problems. This doesn't happen using DSP-Quattro and IKM VIs.

I still don't understand why the DP Audio Performance meter reading is so different from the Activity Monitor, and M5 CPU reading. I have no idea what it's telling me.

One quirk remaining: the polyphony meter when using Star Drums is not consistent. On some cycles of my drum track it reads in the mid 40s, while on other cycles it goes well into the 80s.

[wvandyck]

I was using M5 as a plug-in, not as SA via IAC.

As you mentioned, polyphony for Star Drums-10. JBass was 16.

Lowering polyphony for each instrument did not have any impact here with Streaming on or off.

The DP performance meter always showed minimal or no stress despite my previous mention of lengthy interval to process I/O changes etc. (Looks like you iMac has more processing power than my MBP).

In my case, Streaming On is the better choice as there was no DP Processing visual activity. With Streaming Off, there was maybe 5-10%.

Maybe it's time to look into Vienna Ensemble Pro? Or Plogue Bidule?

[bayswater]

It was running OK today as a plugin with a high polyphony count and streaming off. I got rid of the DP Audio Performance window and depended on the Actiivty Monitor readouts instead.

With the trouble shooting I did today, it's becoming apparent that timing errors had a lot to do with the original problem. I haven't figured out yet whether this is in DP 8 or the 828. But somewhere, I'm getting accumulating timing/sync drift.

Thanks for your comments.

[wvandyck]

You're welcome!.