It’s quite possible the clarett drivers are not working well with Cubase… just to rule that out and be 100% sure, did you try with in built sound? The mac mini 6 core is powerful enough to do what you are asking of it.
Sure, will do some more tests. The Mac mini 2018 does not have built-in audio input, so I will have to test with external devices.
I must say I find this topic very interesting. I never knew that DAWs differed much in performance, but more in the feature sets.
The audio interface I am using now for this test is my new Slate VRS8.
I’ve had minor issues with pops and clicks on the Clarett before in Cubase, but at the time I thought it was caused by wordclock issues, since two UR824s have been hooked up to it via ADAT. It was running at 44.1k and higher buffers…
Now, I will test the Clarett in Cubase again on its own, with 96k and 32 buffer, and of course compare it in Logic.
Thank you for sharing your knowledge on this stuff! 
Yes it’s possible your WC chain had a kink, but then why was it OK in Logic?
Everything you have said specifically points to an issue with clarett and Steinberg.
I have to be honest here… for me personally, focusrite have always had some of the worst interface drivers in the industry… I have been burned by them twice and will never touch any audio interface of theirs again.
The apollo pretty much has the best interface drivers on mac, period. But it’s main strength is it’s DSP, and you want to avoid that.
Still, the AXR4 would probably work brilliantly with Cubase at low latency…
I have to investigate it more to see if it offers “hybrid monitoring”.
Let me give you an example… Apogees latest interfaces have this feature with Logic… they have their own DSP on board but you can still add native effects to the monitoring path in Logic’s mixer and combine it with the DSP monitoring.
I really wonder if AXR4 does this.
Honestly, even though I love pro tools, I firmly believe with all my heart that Cubendo are the finest DAWs in the world… They do literally everything and are easy to use. Just that ONE feature that stops me from migrating to steinberg and using it full time, and that is the way clip volume is done.
It’s right a the start of my youtube video here:
If you think cubase clip gain should work like that, maybe you can support me in a topic I make requesting it.
But, back to latency, let me know how you get along.
Same experience for me. In Studio One, I tested a UA Arrow Thunderbolt interface. At 16 samples. Totally stable, with a RTL of 3.23ms. Really impressive, with no clicks or pops.
Then the same test with Cubase 10, but 32 samples was the lowest it would allow. The audio had continual clicks and pops, which were recorded into the audio.
I switched to 64 samples, which had less clicks and pops, but they were still present. This time the clicks were not present in the recorded audio.
Why is the low latency performance in Cubase so bad? Is there any way to improve it?
I’ve been a ProTools user since the very early days and it’s always been the hardware DSP systems I’ve used. I see a few different use cases with PT - as an audio editor it’s my favorite (and I own Cubase and Logic); as a multitrack recorder with very low latency it’s my favorite, as it’s very, very consistent and predictable. Where I’ve run into problems with PT (and why I’m using Cubase now) is when I use it to score to picture. I have nothing against PT’s MIDI sequencing - though it could have more features, it simply can’t be said that you can’t compose efficiently in it. I do a Warner Bros./John Williams animated score with it and getting through cues was never a problem from a sequencer standpoint - and using memory locations to show and hide orchestral sections with a few taps on the number pad is very easy and elegant - but where it’s clear that there are better ways to do things is when you use a very large composer template and need low latency for live VI input.
Perhaps on a faster system than my 2013 Mac Pro 12-core it would do better - but though I might start a cue with 256 buffer (because many tracks were turned off) by the time the cue was done I’d be at 1024. Also, HDX adds its own problems - if you want to take advantage of hardware DSP on VI’s, for example, there are added delays from needing to jump across the river from HDX to host and back. This was disappointing to me because I have various Plug-in Alliance plugs, for example, that I like to use but don’t want to drag the system down with. The two big tells in that area were: because I use Altiverb for two of my main reverbs, just turning them on caused my system delay (shown in the session setup window) to increase by something like 3000 samples -but the latency (@48k) was much greater than that. And using an HDX-powered plugin on the master bus, like MasterDesk, again increased the latency by more than just the amount of samples one might expect. And to put a bow on it, this may have only been my experience, but my system became quite unstable trying to do it this way - lots of refusing to go into play etc. with larger sessions. In desperation during one episode I switched the interface to the only other thing connected at the time that had an audio out - a CalDigit thunderbolt breakout box with a headphone out - and I was surprised to find that because the HDX card was no longer available, upon replacing the DSP plugs with native ones, I had better latency performance and far fewer issues. So for a time there I was sitting there with (I think) around $9k of audio hardware idle while I listened to virtual orchestra through $5 converters that worked better. I tried switching to DSP- based reverbs for composing but I have to say that was a frustrating experience, because the convolution reverb I had for HDX (Space) is vastly inferior, even when playing back IR’s from Altiverb, and typically because I mix as I go via dynamics and only small fader moves, I don’t have time to rebalance using a different reverb - and though I took the time to balance things so that the HDX verbs were the same in level, the difference was dismaying. Call me picky but I want to hear it as it will be, because I’m writing for virtual orchestra.
PT also made it necessary to use Vienna Ensemble Pro. I like VEPro for many reasons, including the ease with which one can have a modular template and just turn on what one needs. But the added latency via their buffer multiplier is a little bit of a drag.
So when I began to try out Cubase for this purpose, I bypassed that entirely - the template lives entirely (other than what’s on remote machines, which still use VEPro) in Cubase, on instrument tracks, disabled unless they are in use. I have found Cubase has no problem with this approach and it’s very powerful for so many other reasons - the customization and the facility to create custom remote controllers using things like TouchOSC that access specific functions being just two reasons. Also, using ASIOGuard has been great. I wish the AG settings went one more iteration past “high”, but it’s easier to use 512 as my baseline buffer size (for things like SampleModeling Strings that require that in a full-string-orchestra setting) than to be forced to use more than twice that in PT with HDX.
So what’s funny is that in Cubase I’m using the HDX with an HD IO connected to it as… a CoreAudio interface. For this reason I’m looking at the Focusrite 16Line - because it would be nice to have a presumably good-sounding interface that could use a low-latency thunderbolt driver or a Digilink connection to the HDX card, depending on what was required. Though I’m not convinced the 16Line will sound as good as my HD IO. Say what you like but that old Avid hardware is sonically terrific. And I’m not even remotely considering any of the non-Red Focusrite gear. Not saying one can’t get good results with them especially if one is all ITB, but I don’t like them much. Don’t mean to offend - just my taste.
In any event my main complaints with PT have to do with sequencing massive templates. Cubase wins that one for me. But I would rather track and edit using PT HDX any day of the week. No weird hardware monitoring setups - just plain low latency.
Latency is calculated by the speed of your processor. Thunderbolt will always have lower latency over USB because it connects directly to the PCIe bus.
I have a Focusrite Clarett 4 pre Thunderbolt running on an i7 6700K @ 4.4 GHz. I can play a set of V-Drums via midi in on the Clarett and trigger EZDrummer 2 at as low as 16 samples 0.4ms with no pops or clicks.
A PCIe card is also connected to the PCIe bus 
and the latency using the Avid hardware is great for tracking live performances at extremely low latency. My issue is with VI’s - and I agree that in that realm and also in strictly host-based systems clock speed is king. What’s funky about PT is that if you are using HDX, when you leave that land to use a host plugin there is even more latency added. Your system benefits from not going from one processing arena to another. The new Carbon can apparently bypass that somewhat.
As I was determined to move to another DAW having used Logic from the Emagic days I have done a test to compare the performance both in LPX and Cubase 11 pro.
My system is a Mac Pro 2013 12 cores with 64GB of ram.
I have loaded Pianoteq Pro (stereo) playing a midi file, followed by the Pro Q3 with 4 filters activated plus the analyzer, followed by the Waves CLA 76, then the Softube Console 1 with all modules activated (HP and LP filters, Gate, EQ with 2 filters activated, the compressor and saturation all activated. Then a TC reverb.
At 32 samples and with live recording activated, LPX seemed to perform smoother as the CPU meter was more stable but both DAWs were handling the load without any crash. When the Live recording was activated LPX seemed to perform smoother than Cubase. However when the live recording was deactivated they were very similar.
Then I was duplicating tracks to see which DAW was handling more tracks (all with the above plugins loaded and same midi file). Both DAWs were able to play Max 20 tracks at 128 samples without overloading the CPU. One more track at this buffer size was overloading the CPU in both DAWs.
Also both DAWs were able to play up to 24 tracks at 1028 without overloading the CPU. Cubase was not able to play one more track at 2056 without overloading the CPU. Both DAWs were overloading the CPU with the 24 tracks at 512 samples. So the performance for these plugins was absolutely the same for LPX and Cubase 11.