BIG/odd CPU Usage for Abassi and Plini, and it's not my computer (full stats/workup are below)

[traviscollins]

New customer to both Abassi and Plini (love em). I’ve noticed that the CPU load for both plugins is HUGE, and behaves strangely. Ranges from 20% to 35% for one instance in an empty session, and changes depending on whether the plugin GUI is open! Weird. Relative to my system’s capabilities, that % massive. I own Komplete 12 Ultimate Collector’s Edition, Serum, and the Arturia V Collection, and the Neural plugins blow them away in CPU usage by a lot. The runner up is NI’s Modern Drummer at 16%, with internal eq/comp/effects active on 9 “mics”/stereo Kontakt outputs.

Forgive the length here, but I tried to be comprehensive. My system is a brand new (3 weeks old) purpose built custom PC by avadirect.com, and more than exceeds Neural’s recommended system specs -

• DAW is Abelton Live 10 Suite, latest version
• Buffer size is 32MB, sampling at 44,100, 32 bit audio in
• 8 core i7 9700
• Gigabyte Z390 Gaming X motherboard
• 64 GB of DDR4 RAM @ 3,200 mHz
• 500GB NVMe boot drive (where the Neural products are installed along with my DAW, currently at 40% capacity)
• 2TB SDD working drive (where my Abelton projects/sound libaries are stored, currently at 40% capacity)
• 6TB storage drive, not used for audio
• Interface is a Focusrite Scarlett 2i2, but that’s not relevant to the issue.
• The only non-OS 3rd party background software running are Malware Bytes, NVIDIA graphics drivers, and the Focusrite driver
• All native windows settings (power, etc) referenced in your articles have been optimized

Given the fresh/more than capable PC and the fact that I have a hard time believing that the minuscule background software I have running accounts for the CPU drain, I think it’s safe to rule out my computer as the cause.

Specific numbers for each plugin below. They were measured in an empty Abelton session after a computer restart with one audio track only for the neural plugin, which is running stereo out with oversampling at normal.

Abassi, with the plugin GUI open inside Abelton:

• Abassi, Gnarly Rhythm (Nolly) preset, reverb/delay on - 22-24%
• Abassi, Gnarly Rhythm (Nolly) preset, amp OFF only - 15-17%
• Abassi, Gnarly Rhythm (Nolly) preset, cab and amp OFF only - 12%
• Abassi, Gnarly Rhythm (Nolly) preset, cab, amp and delay/reverb OFF only - 6-7%
• Here’s the weird part - with the plugin GUI minimized, empty track with no active inputs or recorded audio - Abassi, Gnarly Rhythm (Nolly) preset, reverb/delay on - 2%
• With the plugin GUI minimized, AFTER audio has been recorded - Abassi, Gnarly Rhythm (Nolly) preset, reverb/delay on - 26-28%

Plini behaves almost identically +/- a few %, with one notable exception. On an empty track with no active inputs or recorded audio the CPU usage does NOT drop with the plugin GUI minimized like Abassi does. It hovers around 20-22%.

Both plugins can spike into the low 30% range when messing with mic placement, delay times, etc.

I ran LatencyMon as your other threads suggest, and these are the numbers. This is over my head, so maybe you can let me know if there’s an issue -

In sum, you’ve got some killer plugins here. But, it looks like there may be some room to streamline some stuff under the hood to make them less system intensive. I’m lucky to have a pretty powerful computer so far as an amateur apartment setup goes. Based on how my other plugins perform (including CPU heavy synths), I never thought I would have to start watching CPU usage on this computer with only an amp sim and drum VST loaded. Abassi + Modern Drummer crack into the 40% range. I can’t imagine how someone running these on a laptop with at/just above your recommended minimum specs deals with it.

Let us know if this is a known issue, and if any solution is expected beyond encouraging us to mitigate its effects.

Thanks!

[Gonzalo]

Hi @traviscollins. Thank you for reaching out and I’m glad you like our Plugins.

There are some topics that should be clarified before moving forward. Real-time performance is not the same as “raw” CPU performance. The usage shown by the ASIO meter will vary greatly between each DAW, and it never corresponds to the actual CPU usage of your system. Moreover, if you duplicate an instance of a Plugin inside a DAW, the system usage shown by Ableton won’t duplicate because it may be using just one thread per track/signal path [1], this meter shows an approximate real-time performance of your system. However, some of the numbers you mention do seem higher than usual. For instance, an instrument track with NI’s Modern Drummer (which is a sample library) shouldn’t use more than 5% on a quad-core processor, but amp sims and similar Plugins will always require more system power due to the nature of their design.

The analysis made by Latencymon can help you to optimize your system to handle real-time audio applications more efficiently. You should keep in mind that at low buffer sizes, only the very low execution times of DPCs and ISRs become acceptable to achieve a consistent workflow without audio dropouts.

Either way, a buffer size of 32 seems unnecessarily low, considering that the achievable buffer size and it’s stability will depend on your system, your audio interface, and the ASIO driver. Usually, a buffer size between 64 and 128 samples is good enough for recording sessions.

Looking at your Latencymon screenshot, the readings given by nvlddmkm.sys and dxgkrnl.sys, both graphics-related processes, seem higher than usual. These readings (combined with the short buffer size) could explain one of the issues you’re encountering (Increment in CPU usage when the GUI is visible). High readings related to graphic drivers are usually solved by enabling msi mode on your graphic driver, this can be done with a free application called msi utility v2. Before making any change, I would verify if the problem is mitigated by increasing the buffer size to 64 or 128 samples.

Lastly, there’s an issue with the current build of Archetype: Abasi that causes some elements of the signal chain to show CPU usage despite being turned off/bypassed under certain circumstances, this could also explain part of the problem. Our development team is already aware of this issue and it should be fixed pretty soon.

Please let us know if you have more questions. I understand that some elements of audio production in Windows systems could be frustrating, but we will be more than happy to help you to get the results you’re looking for. Remember that you can send an email to support@neuraldsp.com if you wish to attach more screenshots, audio files, and videos.

Multi-Core CPU handling in Ableton
CPU Performance vs Real-time Performance

[traviscollins]

Thank you for the very detailed response! I’m new at getting under the hood with this stuff, so the explanation is appreciated. I run the 32 buffer size to keep latency down on my audio interface. It’s an 1^st gen Scarlett 2i2 and I’m running just under 10ms RTL which is about the limit before it gets annoying. I just ordered the 3^rd gen, so hopefully the new drivers will drop that number down enough where I can reduce the buffer size. I’ll take a look at msi mode as well.

Thanks again!

Thanks again.

[Gonzalo]

No problem @traviscollins.

Remember that you can always set a low buffer size for recording projects and then use a bigger one for mixing.

[ray.s]

I noticed the same thing when using Cory Wong in MainStage. Especially when switching patches, cpu usage jump up to 200% +. I guess it’s due to the oversampling in Archetype?
For me, any settings with sample rate higher than 48Khz, or buffer size lower than 64 samples will crash MainStage because of CPU overload.

![Screen Shot 2020-12-03 at 5.34.06 PM|690x420](upload://2M

DlV4jU5qusjHQBKpOE0OXbcq7.png)