Hardware required for Diva, what are you using?

[arachnaut]

Well, if I set it to 44.1kHz and 1024 sample latency I can easily get 20 Diva's running at Divine with only 7% CPU.

And if I kill off some processes and really try hard I could do a lot better.

Plus I could overclock to 4.8 GHz if it mattered, maybe more if I want to run hot.

But... I usually use what I wrote before - 48 kHz, 256 samples.

[Howard]

We need a reference patch to test this. Any suggestions?

[arachnaut]

Sonar must do some type of optimization, if I Clone a Diva track.

I just made 50 cloned tracks set to Divine and I still only see 7% CPU.

We do need some sort of standard, but the point is you won't go wrong getting a Core i7-2600K or similar CPU.

{I've only had Sonar for a few weeks, so take that into consideration, usually I use Live - but it's 32-bit.}

[atraf11]

Sonar must do some type of optimization, if I Clone a Diva track.

I just made 50 cloned tracks set to Divine and I still only see 7% CPU.

We do need some sort of standard, but the point is you won't go wrong getting a Core i7-2600K or similar CPU.

{I've only had Sonar for a few weeks, so take that into consideration, usually I use Live - but it's 32-bit.}

you dont seem to get it... not how many you can have open, but the amount of notes being able to be played at the same time. now, if your machie could handle 20 DIVAs at divine setting all playing at the same time with many voices per instance, then were talking...

[VitaminD]

We need a reference patch to test this. Any suggestions?

Reference patch and MIDI file.

I can tell you I load 10 instances at 10% CPU... but that might be 10 instances of single, quarter notes.


someone else might be using long, 4-5 note chords and claiming only 3 instances at 90% CPU...

[arachnaut]

Sonar must do some type of optimization, if I Clone a Diva track.

I just made 50 cloned tracks set to Divine and I still only see 7% CPU.

We do need some sort of standard, but the point is you won't go wrong getting a Core i7-2600K or similar CPU.

{I've only had Sonar for a few weeks, so take that into consideration, usually I use Live - but it's 32-bit.}

you dont seem to get it... not how many you can have open, but the amount of notes being able to be played at the same time. now, if your machie could handle 20 DIVAs at divine setting all playing at the same time with many voices per instance, then were talking...

Now be nice, I think I do get it.

I had them all playing at the same time, all playing the same MIDI note, live...

You can see the channel meters were all pegged...

Actually I cloned 20 Diva's so the total was 21. They were all playing the first HS preset in the Bass section.

But as has been said, this is probably a bad test because I think Cloning doesn't really generate a new VST instance on tracks... I don't know what it does... but it doesn't increase the CPU usage even though it adds audio samples...

We don't want this to be a CPU war game, do we?, we need a standard test setup to apply apples with apples and PC with PCs. Something like the Live Performance test. But in this case, with multiple hosts.

I don't know if anyone is really interested in 32-bit results either, are they?

So that leaves Live out.

[VitaminD]

I don't know if anyone is really interested in 32-bit results either, are they?

Why would we be interested in 64-bit results? The bonus to using 64-bit software is largely associated with memory addressing. A synthesizer requires tons of calculations not a ton of memory addresses.

[arachnaut]

I don't know if anyone is really interested in 32-bit results either, are they?

So that leaves Live out.

Don't get me wrong, 32-bit computers were my favorites for more than a decade, but now they are ancient history.

Any new computer is going to be 64-bit (assuming you are serious about audio, anyway). And why emulate a 32-bit environment in a 64-bit architecture and sacrifice all the extra registers and optimization?

Let's make 2012 the year of 64-bit, the year of SSD, and stop promoting 32-bit OS's and companies that make 32-bit products exclusively.

If you have a small notebook or telephone - fine, use an 8 bit microcontroller if you want, I'm talking about serious audio work.

[ford442]

I've got Win 7 x64 and i am running Cubase 6 32bit with 5 Diva in draft as well as one ACE, two ADM, one ABL2 comfortably.. it gets hectic if i want any one of the Divas in accurate or great without muting everything else..

Q9650 = LGA 775 quad core 3.0 ghz Intel / 8gb DDR2 1066

[arachnaut]

I don't know if anyone is really interested in 32-bit results either, are they?

Why would we be interested in 64-bit results? The bonus to using 64-bit software is largely associated with memory addressing. A synthesizer requires tons of calculations not a ton of memory addresses.

I don't want to be confrontational, but I think it's much more than that.

If you are using a 32-bit CPU, ignore me, but if you have a 64-bit OS (and I'm only talking about PCs that I know something about) you are dealing with more than just memory.

Why do you think people have to port to run on 64-bit? Different CPU architecture, more operation codes, more registers, more of lots of stuff. I don't really know what goes on down there, but it's very different.

The OS is involved too - the 32-bit app gets all it's system calls remapped into a virtual 32-bit machine. Who needs all these extra layers? There's even a virtual Registry and some virtual system folders.

But don't take my word for it read this:

Benefits of x64 for Audio Workstations
http://www.cakewalk.com/Press/release.a ... rkstations

Naturally they are trying to sell Sonar, but the general facts are still true.

[VitaminD]

I don't know if anyone is really interested in 32-bit results either, are they?

Why would we be interested in 64-bit results? The bonus to using 64-bit software is largely associated with memory addressing. A synthesizer requires tons of calculations not a ton of memory addresses.

I don't want to be confrontational, but I think it's much more than that.

If you are using a 32-bit CPU, ignore me, but if you have a 64-bit OS (and I'm only talking about PCs that I know something about) you are dealing with more than just memory.

Why do you think people have to port to run on 64-bit? Different CPU architecture, more operation codes, more registers, more of lots of stuff. I don't really know what goes on down there, but it's very different.

The OS is involved too - the 32-bit app gets all it's system calls remapped into a virtual 32-bit machine. Who needs all these extra layers? There's even a virtual Registry and some virtual system folders.

But don't take my word for it read this:

Benefits of x64 for Audio Workstations
http://www.cakewalk.com/Press/release.a ... rkstations

Naturally they are trying to sell Sonar, but the general facts are still true.

Perhaps so However, I wouldn't trust anything from cakewalk.

Besides.. much of their argument is in processor architecture improvements.. not in direct 64-bit abilities. CPUs are becoming more efficient and including more instruction sets.. I suspect that is the bulk of the performance increases. I'm sure Urs or Clemens could definitively answer this for us.

[xh3rv]

Test everything, the point of benchmarking is to sort out the bits and the bumps

[arachnaut]

To satisfy my own curiosity, I did the following set up.

I list the track number followed by the preset name (all by HS so I left that off) followed by the voice count in the patch.

I did not change the patch voice count or any patch parameter except to check 'Divine'.

I picked the patches from different categories at random.

I set up my sound card to use 44.1 KHz samples at 2560 sample latency (to exagerrate the results in a positive way).

I activated ALL channels, played 5 simultaneous MIDI notes and held them all for about 20 seconds.

The total CPU during this time maxed at 41.9%, there were no audio glitches.


Track 1 - Bass Nine - 2 voices
Track 2 - Albert Hall Mini - 2 voices
Track 3 - Big Old Polly - 6 voices
Track 4 - Ganymede - 6 voices
Track 5 - Metal Major - 4 voices
Track 6 - Fluffle - 6 voices
Track 7 - Hippie Bubbles - 4 voices
Track 8 - ananoised - 2 voices
Track 9 - Ursa Major - 3 voices
Track 10 - String Ensemble - 6 voices

Total Diva voices = 2+2+6+6+4+6+4+2+3+6 = 41 voices
Total notes = 5 * 41 voices = 205 audio channels.

The presets had a variety of effects on and off, I didn't change them.
The output was really noisey and not at all useable. I set the faders so nothing clipped.

I could not swear that there were no glitches, because it was awful noisey, but there were no dropouts or unusual audio symptoms.

No one would ever use this many things at once.

I could reduce latency even more and kill some processes, but I didn't. I had all my normal processes running - email, browser, background stuff, etc. - nothing disabled. I don't think it would make much difference.

[arachnaut]

I repeated the exact same setup, same channels, same audio settings using Live and the 32-bit Diva.

I got 43.8% CPU used, no glitches.

Not a significant difference.

[arachnaut]

I should say that my motherboard uses dynamic overclocking - it could run at a few GHz on a few cores or all cores at max GHz, but complex things happen and some cores get their voltages reduced for a variety of reasons too complicated to keep track of.

That's why I held the same note for a while and used the worst case CPU. It didn't really change much during the test anyway and I assume that all the cores were being maxed out.

For the CPU percentage I used a neutral third party reporter - TaskInfo's average over all cores.