Completely baffling and amazing CPU behavior with Bitwig 3!

[lunardigs]

Everything is critical, the best computer is the fastest one you can buy with all the RAM, CPU cores, speed and SSDs etc.

I'm quite fond of server hardware. In the past I would cobble together prosumer/gaming stuff, trying to get the biggest bang for the buck. Although, I've found server hardware to be the best with respect to total system stability--and therein realized performance.
Down side is server stuff comes at a premium. Also, you need to match everything properly. I've found it's worth it though.

Right now I'm able to get 24 instances of Diva, with default patch, multicore enabled; 21 instances with multicore off. That's with a 4 bar clip, eight note arp, repeating.
I didn't close any other apps. 28 out of 64GB RAM used, 559 processes running, docker containers, nextcloud server native, 200+ firefox tabs, software RAID, full disk encryption on all disks, etc.--quite a bit of overhead.

System: Xeon E5-2630 v4 @ 2.2GHz (10 phy cores), Supermicro mobo, ECC RAM, NVMe SSD, Nvidia Pascal, Fedora 31 Linux (performance mode during test, X display server, KWin compositor, Nvidia driver 440.44, 3 x 1080 displays). 2016 build.
Audio interface running at 2048 frames, by 4 periods (set by JACK); USB 2.0, RME Fireface UFX.

... So, USB 2.0, KWin and the 3 displays don't help. The USB 2.0 interface hurts. Nevertheless, best performance I've owned as a workstation.

[anoise]

<<< ---- OK in the Windows of OS X's activity monitor showing percentages of programs using CPU what are you seeing? In OS X I can clearly see Bitwig runs all plug ins regardless of sandboxing as separate process's. I'm wondering if this is true in Windows or it's something that's OS specific? like how touch screen controls are much better in Windows obviously.

In windows, separate processes show up only when Plugin host mode in Bitwig, is set to "individually". Regardless of this setting, the result is the same, 7-8 instances max out the system.

[anoise]

Another weird thing I observed, it has probably nothing to do with the topic of this thread.
Set plugin hosting mode to anything but Individually.
Use the init patch in Diva and detune OSC 2 by 0.10, so the meter in the mixer is moving up and down. Now duplicate this track a couple of times and watch the mixer levels. They are fluctuating separately like expected.
Now deactivate the project, set plugin hosting mode to Individually, and re-activate the project.
Now all the mixer levels are fluctuating in sync, thus the audio is louder too.
So plugin hosting mode is influencing the audio... This should not happen.
Don't forget to restart engine/project whenever hosting mode is changed.
This happens with other VSTi's too, not just Diva. I also tried it with Hive and Hybrid3 with the same results.
Anyone else can replicate this?

[jarnold]

Yes having everything is nice but looking at specs it seems that the higher you go in core count the lower the CPU speed.

ie: Mac Pro 7,1
3.5Ghz 8-core
3.3Ghz 12-core
3.2Ghz 16-core
2.7Ghz 24-core
2.5Ghz 28-core

[anoise]

I see people use different "init" presets.
Select the real init preset by right clicking on the central data display that shows the name of the preset.

[stamp]

Yes having everything is nice but looking at specs it seems that the higher you go in core count the lower the CPU speed.

ie: Mac Pro 7,1
3.5Ghz 8-core
3.3Ghz 12-core
3.2Ghz 16-core
2.7Ghz 24-core
2.5Ghz 28-core

Xeon is not what I would reccomend for audio production since everything else out there has higher clock speed nowadays.
But if you 'need' to go with a Mac Pro you have no other choice.

[machinesworking]

Yes having everything is nice but looking at specs it seems that the higher you go in core count the lower the CPU speed.

ie: Mac Pro 7,1
3.5Ghz 8-core
3.3Ghz 12-core
3.2Ghz 16-core
2.7Ghz 24-core
2.5Ghz 28-core

Xeon is not what I would reccomend for audio production since everything else out there has higher clock speed nowadays.
But if you 'need' to go with a Mac Pro you have no other choice.

Kind of true, kind of not. Clock Speed is one thing, in the case of the 4 core 2.7 MacBook here it's actually performing like it's Turbo Boost speed (up to 3.7ghz) is legit, but the Xeon also should boost to at least 4ghz, thereby eliminating any different, but you can see that the 2.7 gets 3.5 instances per core VS the 3.33 at 3. Pure clock speed isn't the only thing, and these brand new architecture Xeons are probably faster per core than anything I have by a big margine, including the 2.5 there.

The Turbo Boost on the 28 core 2.5 Xeon there is to 4.4!
and I don't doubt that track count wise it would show up like that VS the 2.7 at 3.7Ghz, probably better than the numbers show.

[machinesworking]

Another Side note, in real world tests my old Mac Pros old chips are getting in the way of Bitwig's good processor addressing. Throwing three instances of Repro-5 at it with multicore on, cypher 2, Kontakt with a heavily scripted violin library, the MPC2 drum machine software, time stretching a sample, Pigments and Diva at it. it reaches it's limit. One of these is spiking the CPU and it's not able to get another instance.
This isn't bad and pretty much as many things as I ever throw at a song especially heavy CPU things, but it bites away at the lead it had over DP10, and probably Logic. I can get the same out of DP10.
I'm probably still going to pick up Bitwig though, ( it's $219 to upgrade from 8 track at AudioDeluxe right now) the fact that it can keep up with DP or Logic is impressive on it's own, and if they (or DP really) could solve their spiking issue, then it could easily surpass any DAW in any test.

[lunardigs]

It used to be, years ago, ~2005 I seem to recall, that Xeons were the best picked dies from the fab. Meaning, what might become 'run of the mill' CPUs or Xeons started from an early picking process. Almost like first-strike coins. Thus, the dies with the least defects and whatever else became Xeons. I'm not sure if this goes on anymore (as in Pentiums and Xeons share the same line), but I'd image some kind of grading still takes place and is reflected in the retail price.

[stamp]

The Turbo Boost on the 28 core 2.5 Xeon there is to 4.4!
and I don't doubt that track count wise it would show up like that VS the 2.7 at 3.7Ghz, probably better than the numbers show.

Good luck cooling that chip without an industrial chiller considering that it has a tdp of 250watt at base clock.
Considering that 4th gen Ryzen will bring a 50% increase in floating point operation and will still use am4 socket, the best choice is a Ryzen 9 3950x IMO. Buy a 3950x today and when 4th gen is out simply swap cpu. All this with tdps under 120watt and the possibility to keep your pc in the same room where you produce your music.
Ymmv of course.

[stamp]

All that said an 8-16 core cpu and a judiciuos use of resources will get you a long way this days.

[machinesworking]

All that said an 8-16 core cpu and a judiciuos use of resources will get you a long way this days.

That’s my thoughts exactly. My old 12 core mac here will survive at least another four years before some soft synth comes out that truly calls me to upgrade. Anyone getting a 12 core today would last 5-10 years without the need to upgrade IMO. Especially considering I can run a half dozen instances of Repro 5 in high quality mode with multi core on with this old mac. Any modern 12 core chip could probably pull off 20 instances!

[lunardigs]

Yeah, same here; I've got like 5 more years on my current build before I consider anything else. I'd like to get into AMD again, perhaps, but I'll only know when the time comes.

[machinesworking]

Just wanted to add in. Not knocking Bitwig entirely but as some have expected in different circumstances different things happen. Playing a sequence with five instances of Repro 5, Cypher 2, MPC2, Diva, Pigments and Falcon, along with a track with a stretched Apple Loop, and Amplitube etc. Bitwig seems to have a spike with Falcon, so it performs more on line with Digital Performer and other DAWs, crackling the audio past this plug in count.

The fact that it does as well as it does though, and does MPE, I still think I'll upgrade, and take it for a serious spin. Plus Möss did such a great job on the controller implementation for Push 2!

[antic604]

Use the init patch in Diva and detune OSC 2 by 0.10, so the meter in the mixer is moving up and down. Now duplicate this track a couple of times and watch the mixer levels. They are fluctuating separately like expected. Now deactivate the project, set plugin hosting mode to Individually, and re-activate the project. Now all the mixer levels are fluctuating in sync, thus the audio is louder too. So plugin hosting mode is influencing the audio... This should not happen.

Interesting! Perhaps Diva somehow is aware there are several instances being run and randomises each one ever so slightly to sound "analog" (by using different random seeds), but once you've separated them - by switching to individual - it doesn't know that anymore and uses the same random seed for each instance?