Hive 1.2 and CPU

[tony10000]

It looks like Hive is moving from a light CPU synth to a medium CPU one.

I am using a pretty minimal-spec notebook with an 8th gen i3-8130 (dual-core). Unlike older i3s, the 8th gen has turbo boost and it runs at 3.32 GHz. As a result, it outperforms some older i5s and even i7s.

I am noticing that when playing 3 note pads, many of the factory patches get up well over 20% and go up to 30% CPU and even higher. It is very close in performance to Diva in great mode with the multi-core switch engaged on my system.

By contrast, I tested Dune 3 with its factory patches and observed that most were in the 10-20% range, even when multiple layers are employed. Some of the pads were even in the single digit CPU range.

My question is: Will Hive 2 undergo any CPU optimization or at least have multi-core and quality options to improve CPU performance?

[Urs]

Wavetables take a little more CPU than standard waveforms (but still much less than any other WT synth we tried, particularly with oscillator unison). If you take the most popular wavetable synth today, sweeping through a wavetable at 8x unison Hive uses less than half the CPU, maybe even a third only.

Another thing: Displaying the wavetable graph takes considerable CPU, but does not block the audio processing.

Other than that, Hive has become even faster because of changes in the ModMatrix. It actually uses less CPU than before for any preset from Hive 1.1.

That said, can you name any presets that were fine in Hive 1.1 but take more CPU in Hive 1.2?

[MorpherX]

What tony1000 wrote is correct !

But I recognized that Hive 1.2 takes 2% of CPU power with 44 instances in idle state with a 16 thread CPU at 3.7 Ghz.
This is the case with Reaper x64 on Win 8.1 x64 with "CPU reducing function for vstis in idle state" enabled.
My speculation is that Hive 1.2 prevents this function to work correctly because with other VSTis this function works.

Could you please optimize CPU reducing in Hive 2 in idle state, so that there is no CPU usage when no note played.

Thanks !

[Urs]

2% with 44 instances... that's an amazingly low value. I would have expected more.

It is a feature that we do some processing in the background "while not used". Our plug-ins still need to monitor UI changes, parameter automation, MIDI and what not. Also, effects need to be able to settle, e.g. compressors and stuff. Otherwise there will be glitches when the plug-in "wakes up". You'd see threads going "clicks and pops after some idle time", and I really don't want that.

[tony10000]

Wavetables take a little more CPU than standard waveforms (but still much less than any other WT synth we tried, particularly with oscillator unison). If you take the most popular wavetable synth today, sweeping through a wavetable at 8x unison Hive uses less than half the CPU, maybe even a third only.

Another thing: Displaying the wavetable graph takes considerable CPU, but does not block the audio processing.

Other than that, Hive has become even faster because of changes in the ModMatrix. It actually uses less CPU than before for any preset from Hive 1.1.

That said, can you name any presets that were fine in Hive 1.1 but take more CPU in Hive 1.2?

I am going to see if I can do some comparisons using new vs. old presets and to test different ways of reducing CPU via programming (for example, envelope release times).

[tony10000]

That said, can you name any presets that were fine in Hive 1.1 but take more CPU in Hive 1.2?

My quick tests show that changing presets back to the analog shapes has a pretty drastic effect on reducing CPU. The use of Wavetables + Long Envelopes creates the higher CPU usage I have observed. Unison also adds a bit more CPU usage but not as much as I would have thought.

Thus, to make it perform like Hive 1.1, you need to change back to analog shapes. I wish there was a "no wavetable" filter that you could use when you need lower CPU patches. Or perhaps access to the 1.1 patch set.

[Urs]

That said, can you name any presets that were fine in Hive 1.1 but take more CPU in Hive 1.2?

My quick tests show that changing presets back to the analog shapes has a pretty drastic effect on reducing CPU. The use of Wavetables + Long Envelopes creates the higher CPU usage I have observed. Unison also adds a bit more CPU usage but not as much as I would have thought.

Thus, to make it perform like Hive 1.1, you need to change back to analog shapes. I wish there was a "no wavetable" filter that you could use when you need lower CPU patches. Or perhaps access to the 1.1 patch set.

You can select all wavetable presets in the preset browser (tag: wavetable) and move them into a separate folder. Then they're out of the way.

(Note that installing an update will bring them back though, unless you untick the preset install option)

[tony10000]

That said, can you name any presets that were fine in Hive 1.1 but take more CPU in Hive 1.2?

My quick tests show that changing presets back to the analog shapes has a pretty drastic effect on reducing CPU. The use of Wavetables + Long Envelopes creates the higher CPU usage I have observed. Unison also adds a bit more CPU usage but not as much as I would have thought.

Thus, to make it perform like Hive 1.1, you need to change back to analog shapes. I wish there was a "no wavetable" filter that you could use when you need lower CPU patches. Or perhaps access to the 1.1 patch set.

You can select all wavetable presets in the preset browser (tag: wavetable) and move them into a separate folder. Then they're out of the way.

(Note that installing an update will bring them back though, unless you untick the preset install option)

OK...good idea! Anyway to get multi-core support to reduce CPU with wavetables or would that help?

[Urs]

Anyway to get multi-core support to reduce CPU with wavetables or would that help?

I think Multicore support would cost more overhead than the benefit would be worth. As long as a single instance is still fully playable I'm sure you're better off letting the host handle the distribution of plug-ins onto CPU cores.

That said, I can't imagine Hive being that dramatic on any recent system. Sure, it uses more when wavetables are involved, but it shouldn't be a CPU hog ever. What samplerate and buffersize are you using?

[tony10000]

Anyway to get multi-core support to reduce CPU with wavetables or would that help?

Extremely conservative settings: 44.1K and 4096 samples.

And my notebook is a little more that a year old. The CPU was introduced in Q1 of 2018:

https://ark.intel.com/content/www/us/en ... ssors.html

I think it would be beneficial to do some comparative tests between Hive and other VSTs.

[EvilDragon]

Are you watching the CPU clock speed as you use Hive in your DAW? Does it oscillate a lot? DAWs usually don't like this. I see there's a pretty big difference between base clock and TurboBoost speed, 2.2 GHz dualcore is definitely pretty slow by today's standards no matter how you spin it. Only if you can make your CPU to be pegged all the time at that 3.4 GHz and without any oscillations in frequency (disable SpeedStep) you might see some benefits, I think.

But there's a reason why it's an i3 and not an i5 or an i7.

[tony10000]

Are you watching the CPU clock speed as you use Hive in your DAW? Does it oscillate a lot? DAWs usually don't like this. I see there's a pretty big difference between base clock and TurboBoost speed, 2.2 GHz dualcore is definitely pretty slow by today's standards no matter how you spin it. Only if you can make your CPU to be pegged all the time at that 3.4 GHz and without any oscillations in frequency (disable SpeedStep) you might see some benefits, I think.

But there's a reason why it's an i3 and not an i5 or an i7.

No, my CPU speed is pretty much locked at 3.28 GHz when I am running the DAW. I have dual fans under the notebook to help with cooling and to keep it from throttling since it is a fanless design. The 8th gen core i3 is no slouch and is much faster than the older Core i5 I used to use. And I could not ask more of a $350 notebook!

https://www.cpubenchmark.net/cpu.php?cp ... Hz&id=3225

[tony10000]

Here are some benchmarks using Ableton 10:

Patch = Ambient Soundscapes
Number of Notes = 3
Latency = 1104 samples
Peak CPU = 32%

Patch = Ambient Soundscapes
Number of Notes = 3
Latency = 2128 samples
Peak CPU = 32%

Patch = Ambient Soundscapes
Number of Notes = 3
Latency = 2128 samples
Peak CPU = 32%

Patch = Ambient Soundscapes
Number of Notes = 3
Latency = 3152 samples
Peak CPU = 32%

Patch = Ambient Soundscapes
Number of Notes = 3
Latency = 4096 samples
Peak CPU = 32%

[tony10000]

Here are some benchmarks using Ableton 10:

Patch = Ambient Soundscapes
Number of Notes = 3
Latency = 4096 samples
Peak CPU = 32%

By contrast, using the same test loop and using a three layer Dune 3 pad patch, I get a 10% better result:

Patch = Dark Choc ARK
Number of Notes = 3
Latency = 4096 samples
Peak CPU = 22%

However, if I turn off multi-threading in Dune 3, the performance is similar to Hive 1.2:

Patch = Dark Choc ARK
Number of Notes = 3
Latency = 4096 samples
Peak CPU = 33%

So, multi-threading does make a huge difference, at least in the case of Dune 3.

[Teksonik]

Here on my internet computer which is an old weak i3 running at 3.1 GHz I'm getting 21% average with 3 notes (#36,18, and 60) with Hive "AS Ambient Soundscapes".

512 Samples (12ms) 44.1 Hz FL Studio ASIO. FL is spreading the load across all 4 logical processors quite well.

EDIT: Examining that patch further it's probably towards the more extreme end of CPU demand as it uses 21 total Unison, one WT, almost all the FX and a healthy dose of Release on both Amps (and it sounds awesome).