u-he

Lbdunequest · Post by **Lbdunequest** » Fri Dec 23, 2022 3:48 pm

I figured it might be worth to continue discussion from a previous topic because i believe there might be some room for optimisation

As mentioned in an original thread I played each instrument 16 notes at the same time, on both i set oscillator Unison at 16, 2 oscillators each.

Im using 4790k intel processor, with 32gigs of ram.

Original thread: viewtopic.php?t=591044&start=45

Urs · Post by **Urs** » Fri Dec 23, 2022 5:04 pm

Have you tried "fixed buffer size" with Hive, and set buffer size to, say, 64, 128 or 256 samples? This might give a more realistic readout.

Background: FL might process a single sample. Hive however always processes multiples of 16 samples. So, in FL, sometimes Hive processes nothing, and just accumulates samples until it has more than 16 (or a multiple thereof). Now, sometimes Hive might process 16 samples when FL asks it to process just a single sample. At this moment, FL sees "OMG, this plug-in uses 16 times the CPU that it normally would" and cranks up its CPU readout.

So there. Always processing multiples of 16 samples is a phantastic way to reduce CPU usage. But sometimes it may look as if it was the opposite.

Anyhow, it would be great if you had the presets. Maybe there's more to it, and we can have a look at the actual settings.

As I said elsewhere, all our measurements of Hive agains Dune were rather close, I'd be surprised if there was a setting that was off by this much. And if there is, I would like to know what I have missed (but just posting a screenshot does not tell me that).

Urs · Post by **Urs** » Fri Dec 23, 2022 5:12 pm

It's been a while, 1.5 years maybe, but this is what I found:

Sawtooth Osc, 8x unison, 8 voices:

<snip - no need to post this anymore, see below>

Sawtooth Osc, no unsion, 8 voices (Note: Advantage @ Dune 3):

<snip - no need to post this anymore, see below>

Sawooth Osc, 8x unison + resonant filter:

<snip - no need to post this anymore, see below>

That's in Reaper, Mac Pro.

Urs · Post by **Urs** » Fri Dec 23, 2022 5:36 pm

(also noteworthy that I can't know what's going on in the other synths. Apparently, switching filters on or off has made a difference, as did switching oscillators on or off. But who knows, with Hive at least when 1 filter is on, both filters compute. So there's a good chance that other people reach other findings. I did my best to level the field, e.g. by switching off oversampling wherever possible, switched off multithreading where applicable and set modulation quality to something that feels like what we do in Hive)

Urs · Post by **Urs** » Fri Dec 23, 2022 5:39 pm

Also, as my colleague just points out: Certain UI features in Hive can cost extra CPU. E.g. the built-in CPU meter itself, and most obviously the Scope. Whenever Hive's UI is open with the scope visible, there's extra demand on the CPU. Any such comparison should be made with Hive's UI tabbed to something else, as Hive's scope is a rather unique feature.

Lbdunequest · Post by **Lbdunequest** » Fri Dec 23, 2022 6:53 pm

thank you for reacting to this thread. But in my case setting fixed size buffers or turning off gui has no effect on CPU

I played in this test 16 notes btw

i added the preset

tasmaniandevil · Post by **tasmaniandevil** » Fri Dec 23, 2022 7:01 pm

Btw, as other elements, like the oscillator and filter visualization will also affect the CPU, it's best to always close the plugin GUI completely when trying to do feature performance tests.

Oh, and do not record arm the tracks you want to compare.
Some hosts will have an increased CPU demand if the track is armed for real-time playing/recording.
Instead, create MIDI clips and play those clips.

Lbdunequest · Post by **Lbdunequest** » Fri Dec 23, 2022 7:09 pm

tasmaniandevil wrote: Fri Dec 23, 2022 7:01 pm Btw, as other elements, like the oscillator and filter visualization will also affect the CPU, it's best to always close the plugin GUI completely when trying to do feature performance tests.

Oh, and do not record arm the tracks you want to compare.
Some hosts will have an increased CPU demand if the track is armed for real-time recording.
Instead, create MIDI clips and play those clips.

There is nothing like that activated and gui was closed

opened gui makes no difference

Urs · Post by **Urs** » Fri Dec 23, 2022 7:17 pm

Well, thank you for the preset.

You got filters on, and not cheap ones, for starters. The Hive preset has spread, the Dune one doesn't. I think that made a difference, not sure.

Would be cool to have the Dune preset as well.

Urs · Post by **Urs** » Fri Dec 23, 2022 7:28 pm

Your preset was made with Hive 2.1.0, revision 11037. That's the one before we implemented aggressive resource balancing. This version always processes all modulators and everything, simply because that's faster with real world patches. It does however compute less nicely for such performance tests, which is what led to Hive 2.1.1 (apart from M1 support).

Hive 2.1.1 has, among other things,

- Better resource management for improved CPU performance

https://u-he.com/products/hive/releasenotes.html

It was part of our "performance update" which is the version I was talking about in the other thread and which was used for my own measurements.

Lbdunequest · Post by **Lbdunequest** » Fri Dec 23, 2022 7:30 pm

Urs wrote: Fri Dec 23, 2022 7:28 pm Your preset was made with Hive 2.1.0, revision 11037. That's the one before we implemented aggressive resource balancing. This version always processes all modulators and everything, simply because that's faster with real world patches. It does however compute less nicely for such performance tests, which as what led to Hive 2.1.1 (apart from M1 support).

Hive 2.1.1 has, among other things,

- Better resource management for improved CPU performance

https://u-he.com/products/hive/releasenotes.html

It was part of our "performance update" which is the version I was talking about in the other thread and which was used for my own measurements.

I had the newer version installed, but i asked support to recieve older version installer because newer one started to cause Crashes inside FL Studio.

Lbdunequest · Post by **Lbdunequest** » Fri Dec 23, 2022 7:31 pm

Urs wrote: Fri Dec 23, 2022 7:17 pm Well, thank you for the preset.

You got filters on, and not cheap ones, for starters. The Hive preset has spread, the Dune one doesn't. I think that made a difference, not sure.

Would be cool to have the Dune preset as well.

What other 24 db filter would be easier on CPU?

I have added 2 dune presets, one with CL 24 (this one barely uses any cpu) and other with the heaviest saturn filter which still uses couple times less cpu then Hive

Urs · Post by **Urs** » Fri Dec 23, 2022 7:36 pm

So, you were aware the whole time that you were using a version from before the performance update?

I mean it's a pity that the newer version would crash for you, that's normally not the case.

But asserting that the optimisation we made wasn't effective simply because your experience is based on a version that does not have this optimisation... I must say, that's pissing me off. What a waste of time.

Lbdunequest · Post by **Lbdunequest** » Fri Dec 23, 2022 7:46 pm

Urs wrote: Fri Dec 23, 2022 7:36 pm So, you were aware the whole time that you were using a version from before the performance update?

I mean it's a pity that the newer version would crash for you, that's normally not the case.

But asserting that the optimisation we made wasn't effective simply because your experience is based on a version that does not have this optimisation... I must say, that's pissing me off. What a waste of time.

Im sorry, didnt ment to do that. But i installed the latest version and i must say thats quite an improvement. Still not CPU efficient as Dune 3, but really significant improvement (both guis closed)

thank you

just there is a need something to do about those crashes...

Lazarus451 · Post by **Lazarus451** » Fri Dec 23, 2022 7:56 pm

Merry Christmas! Both are fantastic synths.

u-he

KVR Audio

U-He Hive 2 cpu usage vs Dune 3 (FL Studio 21)