Trying to get lower latency with higher buffer sizes : worth upgrading my Scarlett 2i2 (3rd gen) to a Clarett 2pre USB?


I am currently thinking about upgrading my Scarlett 2i2 (3rd) gen with a Clarett 2pre USB in order to get lower latency with high buffers sizes (512+) on Archetype Nolly 2.0.0.

I am using the plugin in standalone mode to practice 95% of the time.

The settings of the plugin are the following :

  • Audio Device Type : ASIO
  • Audio Device : Focusrite USB ASIO
  • Audio Output Channels : Output 1 & 2 enabled
  • Audio Input Channels : Input 1 enabled ; Input 2 disabled
  • Sample Rate : 44.1KHz
  • Oversampling : High on all the presets

My problem as to do with the high buffer sizes that I need to use in some situations in order to get rid of audio artefacts, all while keeping the oversampling set to “high”.

If I run the plugin in standalone mode without any other programs running, I need a buffer size of 256, which gives me a latency of 5.8 ms : no problem here.

If I disable the PostFX section, I can lower the buffer size to 128, which gives me a latency of 2.9 ms : no problem here again.

If I want to practice over some backing audio playback, or while I’m also using Guitar Pro 7.5, pops and glitches will not stop occurring until I set the buffer size to 512, which gives me a latency of 11.6 ms : at this point, the latency difference is noticeable and I start to find it annoying even if it’s still manageable.

Sometimes, I like to use my pedalboard effects in front of the interface, then into the clean amp of the plugin with preFX and postFX sections disabled. In mono, it works well, sounds very good to my ear and it doesn’t affect the required buffer.

The worst latency situation I face is when I want to connect my pedalboard to the interface while using stereo outputs of my Strymon pedals. With both input channels of the 2i2 are enabled and with the plugin set to stereo mode, I need a buffer size of approximately 880 samples to get rid of all audio artefacts, which gives me a 20ms latency. With this kind of lag, using the plugin in stereo mode isn’t worth it in my perspective.

I highly doubt that my PC’s hardware is the limiting factor here.

Specs :

  • OS : Win10 Pro, version 21H1
  • CPU : Intel 8700K @ 5000MHz 1.36v
  • RAM : Trident Z 16GB DDR4 @ 3200MHz
  • MOBO : Asus Maximus X Hero Z370
  • GPU : Asus ROG 1080 ti

The Scarlett 2i2 is connected directly into one of the USB 3.1 Gen 1 ports from the motherboard. Both TRS outputs of the interface are used to connect to my studio monitors.

It is worth mentioning that I followed every recommendation of the Optimization Guide for Windows.

In the Windows 10 settings :

  • Latest drivers and firmware installed
  • “High Performance” power plan selected
  • All audio devices disabled except for the Scarlett
  • USB power management disabled
  • Processor scheduling set to background services
  • Startup and background kept to the very minimum
  • Game mode disabled
  • Nvidia GPU power mode set to “maximum performance”

In the BIOS settings :

  • Latest BIOS version installed
  • Onboard Audio disabled
  • All C-states disabled
  • Intel SpeedStep and SpeedShift disabled
  • 100% stable overclock set to a fixed frequency and voltage on all 6 cores
  • No thermal throttling whatsoever

Also, I ran LatencyMon for multiple hours without any problems recorded in the stats.

All of this being said, I am currently wondering if an upgrade from my Scarlett 2i2 (3rd gen) to a Clarett 2pre USB would allow me see any meaningful reduction in latency with higher buffers sizes?

When using the standalone plugin over another source of audio playback, can I reasonably expect from such an upgrade to lower my current 512 buffer latency of 11.6ms to something closer to 5ms?

When using the standalone plugin in stereo input mode, can I reasonably expect from such an upgrade to lower my current 880 buffer latency of 20ms to something closer to 10ms?

Thanks in advance for your help with this matter!

No, this is not possible… The minimum latency is directly tied to the buffersize and Sample frequency. e.g. at 44100hz,
441000 samples take exactly 1s to be read. Buffering 512 samples takes 44100/512 seconds time= 0.010=10ms.(maybe effectively more due to other processing , but never less)
You could have stable lower latency (=lower buffersize) with better drivers, maybe try an RME interface.
You are asking for a time machine, basically:-)