Sonar performance compared to Bandlab

[David Baay]

On 11/4/2024 at 6:37 AM, Sergei Pilin said:

Here it is.

FWIW. I don't have the FX plugins but downloaded the trial of Spire, and found the project behaves pretty much identically in CbB and Sonar. It ran pretty cleanly in both at 96 samples in ASIO mode (Roland Duo-Capture EX) on my I7-11800H laptop with Engine Load peaking at around 60% and exactly two late buffers per iteration of the loop - one at Note On and one at Note Off. Staggering the start times of the identical clips in the 54 tracks cured that.

Switching to WASAPI-Exclusive, CbB played with a lot of distortion at 132 samples (3ms @ 44.1 kHz) using the Roland interface, but at least did not drop out.

Using the onboard Realtek for output in WASAPI-Exclusive mode at 3ms, I could not even get the engine to start in CbB. I had to raise the buffer to something on the order of 18ms to get it playing with an Engine Load comparable to ASIO, and even then it was having audible hiccups with Engine Load spikes. Seeing this, I didn't bother trying Sonar in WASAPI mode.

Conclusion 1: Sonar's performance seems equal to CbB in this particular project scenario in ASIO mode.

Conclusion 2: WASAPI just doesn't have the towing capacity to pull this kind of load, especially using onboard Realtek for output.

 

[Tim Elmore]

After hearing about the performance improvements of Sonar, I decided to try it out for myself. My configuration is as follows.

Processor: Intel Core i7 4th generation at 3.4GHz with 16MB of RAM. It has 4 cores with two threads each.

Audio Interface: Roland Quad-Capture, running with native ASIO drivers.

ASIO buffer size: 64 samples.

Sonar version: 2024-09-MEM (Build 105, 64 bit). It is not activated, as I want to try it before buying a license.

The test project consists of 50 tracks. Each track has a pink noise clip and three instances of the TDR Nova compressor. No other plug-ins are used.

CbB results:

All 50 tracks active: CbB stops after a fraction of a second due to long dropouts.
49 tracks active (one track is archived): Playback has frequent drop-outs, but continues to play.
48 active tracks: Playback is steady.

Sonar results:

All 50 tracks active: Sonar has frequent dropouts, but continues to play.
49 active tracks: Playback is steady.

In this test, Sonar plays one more track than CbB, or about 2% more. Is this the expected amount of improvement for Sonar over CbB?

Could my processor or audio interface keep me from being able to take advantage of the Sonar performance improvements?

[Glenn Stanton]

for me - when recording - i can set the latency really low (say 8-32 samples = ~4-5ms roundtrip)  (if needed, some things like not using the track echo but rather the live mix for recording let's me set it much higher). for mixing, i set it to a minimum of 2048 (= ~100ms) and add read caching (2Mb) as well. this relieves a lot of stress on the CPU and other system components.

not sure of the value in mixing with 64 samples when it doesn't really matter...

Edited Monday at 08:38 PM by Glenn Stanton

[Tim Elmore]

For the experiment, I used 64 samples to make it easier to reach the breaking point with a simple test project. Normally I stay at 96 samples for both recording and mixing, and that works pretty well for the plug-ins I use. The latency is a tolerable 9.8ms, and by keeping it the same all the time, there is one less setting to keep track of when switching between recording and mixing.

Still, I wanted to explore the performance of Sonar. But my system sees negligible improvement while others are obviously enjoying much more impressive gains. I wonder if something about my configuration keeps Sonar from performing better than CbB.

[Xoo]

Less scientific but with my old MOTU Ultralite III, ASIO,  I can run Sonar at 64 samples buffer with a reasomable project, but CbB drops out on the same project almost immediately.

[Noel Borthwick]

On 11/11/2024 at 12:51 PM, Tim Elmore said:

After hearing about the performance improvements of Sonar, I decided to try it out for myself. My configuration is as follows.

Processor: Intel Core i7 4th generation at 3.4GHz with 16MB of RAM. It has 4 cores with two threads each.

Audio Interface: Roland Quad-Capture, running with native ASIO drivers.

ASIO buffer size: 64 samples.

Sonar version: 2024-09-MEM (Build 105, 64 bit). It is not activated, as I want to try it before buying a license.

The test project consists of 50 tracks. Each track has a pink noise clip and three instances of the TDR Nova compressor. No other plug-ins are used.

CbB results:

All 50 tracks active: CbB stops after a fraction of a second due to long dropouts.
49 tracks active (one track is archived): Playback has frequent drop-outs, but continues to play.
48 active tracks: Playback is steady.

Sonar results:

All 50 tracks active: Sonar has frequent dropouts, but continues to play.
49 active tracks: Playback is steady.

In this test, Sonar plays one more track than CbB, or about 2% more. Is this the expected amount of improvement for Sonar over CbB?

Could my processor or audio interface keep me from being able to take advantage of the Sonar performance improvements?

A lot of the performance gains come from optimizations  for repeated operations that were consuming unnecessary CPU at low latency.

Your system has only 4 physical cores so while you will see gains they won't necessarily be that dramatic as they are with lots of cores. The reason is that your system is gated by 8 concurrent processing threads out of which 4 are sharing cores. Your test use also artificial. So practically speaking, with 50 tracks and an  equal workload,  at best it will process the workload in 1/ 6 the time for each buffer. Whatever gain you see will be proportionate to this ratio as well. 

In the next update  we've shaved off even more cpu cycles, so you will see yet more improvements.  some testers reported up to 30 percent gains in their projects.