Completely baffling and amazing CPU behavior with Bitwig 3!

[pdxindy]

What Quality setting are people using for Diva too...

Which OSX version? Etc.

[machinesworking]

What Quality setting are people using for Diva too...

Which OSX version? Etc.

In my ramblings at the beginning there I posted it's latest version of Mojave,

The benchmark is a four bar cycled continuous eighth note run at 120 BPM. All DAWs are set to 256 buffer. The exception, I let Reaper cheat like it does by not having a set buffer.

The Quality setting in Diva is "Great", no multicore etc. It's the initial blank INIT patch that comes up with Diva when you instantiate it.

[anoise]

I know that CPU benchmarks are important but if I recall correctly, ( and there's plenty of people on here more informed than I am on this to correct me if I'm wrong ), it's all broken down into different types of operation and various CPUs can score better or worse on various operations etc. At that site you mentioned cpubenchmarks.com it scores the 3.33 at 8354 and the 2.7 at 8402, when clearly the 2.7 is a better CPU per core, just not at four cores VS 12.

I guess you mean CPU instructions like AVX, AVX2 etc. In our comparison it does not matter, because both of our CPU's are roughly from the same generation with the same instruction set and same core counts, yet still, your laptop CPU with a 8402 score can handle 14 instances of Diva, while my desktop CPU with 9856 score can only 7-8. How is that possible? (using: Diva 1.4.4 rev.9603, Bitwig 3.1.1)
https://www.cpubenchmark.net/compare/In ... /2027vs899

[machinesworking]

I know that CPU benchmarks are important but if I recall correctly, ( and there's plenty of people on here more informed than I am on this to correct me if I'm wrong ), it's all broken down into different types of operation and various CPUs can score better or worse on various operations etc. At that site you mentioned cpubenchmarks.com it scores the 3.33 at 8354 and the 2.7 at 8402, when clearly the 2.7 is a better CPU per core, just not at four cores VS 12.

I guess you mean CPU instructions like AVX, AVX2 etc. In our comparison it does not matter, because both of our CPU's are roughly from the same generation with the same instruction set and same core counts, yet still, your laptop CPU with a 8402 score can handle 14 instances of Diva, while my desktop CPU with 9856 score can only 7-8. How is that possible? (using: Diva 1.4.4 rev.9603, Bitwig 3.1.1)
https://www.cpubenchmark.net/compare/In ... /2027vs899

AVX etc. are part of it in some tests but not this one I believe. Different chips get different scores with plug in performance based on other things as well. Anecdotal, but years ago I bought a 667ghz powerbook that underperformed a 500mhz powerbook specifically on plug in count, because the L2 cache was smaller. There's also varying floating point scores for the same Ghz chips from different companies or different models.
Other odd things might play into it. What's your RAM speed? amount?
The MacBook has 16GB DDR3 at 1600mhz. It's also a propriety m.2 SSD 512GB drive in these laptops. graphics card - NVIDIA GeForce GT 650M 1GB.

[anoise]

Same RAM amount and speed. SSD's and GTX970, but these don't matter much in this case. There is something else going on.

[apoclypse]

Seems to me like Bitwig values core/thread count versus clock speed. It makes sense since Bitwig is running each plugin within its own process so it basically spinning up threads depending on the amount of plugins you are running but also the threading model you choose in the preferences (Plug-in Hosting Mode).

I did the test on my end using Bazille and I was able to run 11 instances of the plugin at 256 buffer before it broke up on my iMac (it's an older Late 2012 model 3.4 Ghz, 32GB RAM). In Ableton I was able to run 14 instances.

What's really interesting is the difference between Mac and Windows here. I have an older Haswell-E X99 6-core PC here and will do the same test on it to see what happens.

[machinesworking]

Same RAM amount and speed. SSD's and GTX970, but these don't matter much in this case. There is something else going on.

Anoise. Yeah that's weird? Again, I suspect some of the reason they're not publishing this is because of the huge variance between machines. I suspect it's something like floating point operation or L2 cache. Chip manufacturers are bastards and a chip might perform well on a test but not in real life. The fact that a "faster" chip doesn't perform as well isn't that surprising. I doubt it's an OS thing? Bitwig specifically worked with Microsoft on touch surfaces so they I'm sure are in touch with them about optimizing the OS. <<< ---- 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.

[machinesworking]

Seems to me like Bitwig values core/thread count versus clock speed. It makes sense since Bitwig is running each plugin within its own process so it basically spinning up threads depending on the amount of plugins you are running but also the threading model you choose in the preferences (Plug-in Hosting Mode).

I did the test on my end using Bazille and I was able to run 11 instances of the plugin at 256 buffer before it broke up on my iMac (it's an older Late 2012 model 3.4 Ghz, 32GB RAM). In Ableton I was able to run 14 instances.

Yeah that's crazy? Live by far on every machine I've had is the biggest CPU pig.
Bitwig and Live before the latest updates would be at almost exactly the same level. I couldn't Get anywhere near Bitwig's performance with Live? Are you running all VSTs as sandboxed? I'm not.

[jarnold]

So if you were buying a new computer to run Bitwig, what would you look for; processor speed? number of cores? L2 cache?

[stamp]

So if you were buying a new computer to run Bitwig, what would you look for; processor speed? number of cores? L2 cache?

Depends on the average number of tracks in your projects, if you use vsts or sampled instruments, if you record external hardware/instruments and your workflow.

[machinesworking]

So if you were buying a new computer to run Bitwig, what would you look for; processor speed? number of cores? L2 cache?

It seems to really respond really well to multi core machines. Some people with four core machines seem to not be getting good performance, but my 12 core and others seem to be doing really well with it.

My four core even had spikes at first that limited it to 4 instances, the next day it didn't occur and it got fourteen. I'm super surprised it seems to do really well with an older Xeon without AVX etc.


These days any machine with an SSD is going to do well streaming tracks, IMO a larger SSD is a GREAT idea! both of mine are 512GB and that's not near large enough to host Komplete Ultimate let alone my other samplers etc. IMO getting a laptop if that's what you want, I would go with a larger SSD or HD.

[JHernandez]

I get 31 instances (2bar 8 note chord) without crackling on Ryzen 7 2700x. 32 if I turn off performance graph.

I have to be doing this wrong. I'm getting 82 instances before it croaks.

Cool!
Hey would both of you toss out more system specs? name and number of CPU, how many cores? RAM?

2010 Mac Pro 12 core 3.33ghz 16GB ram. Crappy ATI Radeon HD 5770 on High Sierra. The thing is that when I load up Diva it's not really an init patch, it has reverb and sound pretty good on it's own. With that patch I could get 10 instances of a repeating 8th note triad @ 120 BPM, 256 buffer. The 82 instances was a single repeating 8th note with a true init. When I actually init it from the menu I could go way up into stupid territory. I didn't bother checking how far I could take the triads on a "true init" because when it comes down to it I can only get about 2-4 instances of a great patch before it's time to bounce, not that I'd wait that long.

[machinesworking]

I get 31 instances (2bar 8 note chord) without crackling on Ryzen 7 2700x. 32 if I turn off performance graph.

I have to be doing this wrong. I'm getting 82 instances before it croaks.

Cool!
Hey would both of you toss out more system specs? name and number of CPU, how many cores? RAM?

2010 Mac Pro 12 core 3.33ghz 16GB ram. Crappy ATI Radeon HD 5770 on High Sierra. The thing is that when I load up Diva it's not really an init patch, it has reverb and sound pretty good on it's own. With that patch I could get 10 instances of a repeating 8th note triad @ 120 BPM, 256 buffer. The 82 instances was a single repeating 8th note with a true init. When I actually init it from the menu I could go way up into stupid territory. I didn't bother checking how far I could take the triads on a "true init" because when it comes down to it I can only get about 2-4 instances of a great patch before it's time to bounce, not that I'd wait that long.

The part in bold, yeah that's the INIT patch, it has the plate reverb on it.
I haven't tried a triad, I was doing a single note end to end F F# run.
We will get different results with different settings, so it's not surprising that you got more instances.

[jarnold]

Hmm Ok so a higher core count processor for VSTs/VSTi and a fast SSD for track count. Actual clock speed not critical?

[machinesworking]

Hmm Ok so a higher core count processor for VSTs/VSTi and a fast SSD for track count. Actual clock speed not critical?

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

My own discovery is that Bitwig uses multi core systems really well, better than the others. When you get a computer for any DAW plug in count equals CPU speed and core count, disc streaming is going to improve with SSDs but the terabytes will go down, if you use a lot of heavy sample libraries a good amount of RAM helps.

I have older machines, but with decent specs even for todays standards and Bitwig is shining here. To be fair though other DAWs are not, and Massive X won't even load on the Mac Pro since it doesn't support AVX like the new Mac Pros do.