Latency Values

The latency i have on reading the values from Cubase 12 are 14ms in and 15ms out.
Are those values Ok ?
Everything I read about latency has the usual vagueness.
What are the cut off values between good and bad ?
Simple exact figures would be good instead we get high latency or low latency.

This is not possible since it all depends on the use case and the person listening to the signals.
To avoid latency issues during recording, use direct monitoring or a hardware mixer.

Yes, for recording direct monitoring is preferable. In case this isn’t possible, say you want to play through a guitar amp simulation, such a latency is considered very high and you should probably aim for a round trip latency of under 10 seconds (input and output summed together)
For live playing from a MIDI keyboard only the output latency is relevant. Lower doesn’t hurt here either.
If you just work with loops or edit the piano roll manually, the latency is irrelevant.

OK Thanks. Im using a pc at home in the spare bedroom.
I record audio and MIDI but am thinking of what the Cubase Studio setup values should be to give you OK results without stuttering or drop out etc.

Are you doing overdubs?

Yes. I’m doing overdubs.

Depends on your audio interface and the power of your pc. I use a rme Babyface pro fs and my latency is about 2ms with 64 sample buffers on an i9 9900. If you are using the inbuilt audio then you are not going to get close to that without glitches although you could improve if using asio4all. You don’t mention which interface though.

I’m wondering what are the limits that Steinberg work to.

10ms as far as I know is the recognised latency you need to get to or under so you don’t notice latency. Depends if you mean for recording audio and monitoring with an effect or triggering midi from a controller of some type.

The latency i have on reading the values from Cubase 12 are 14ms in and 15ms out. Are those values Ok ?

Sounds good to me. I’ve got an RME FireFace UFX running at 44.1khz with a 512 sample buffer, and my latency is 12.1ms input/13.9 output. I also have ASIO-Guard set to High, which results in ASIO-Guard latency of 92.9ms!

Of course I always use direct monitoring through the RME TotalMix software when recording vocals or external instruments, so the latency values are meaningless to me. I use a large-ish sample buffer and High ASIO-Guard settings because once my tracks near the end of the mixing stage, all the FX plugins tend to result in audio glitches with a smaller buffer/lower ASIO-guard. I am a set-it-and-forget-it type, so I just set everything high and never worry about it again.

I can’t set high and leave it, as I play edrums to trigger sd3 and It feels best with the lowest latency I can get. I leave mine at 64 samples (2ms) and find it works even on quite large projects with plenty of effects. For audio I too use TotalMix so no need to bother with latency.

1 Like

Thanks for all the help guys.
I’ll have to set up my system properly.

I have an i9 18 core, 36 threads, overclocked 60% with the fastest RAM 64 gigs, all M2 and SSD drives, liquid cooling, fast video card… and my latencies at 96k/24 with 64 bit float are 11.3 ms input and 21.3 ms output with a 128 buffer. That may sound good but, as soon as I load Ozone 9 with a few mastering modules, my performance (press F12 on Cubase) leaps to 25% average, with 50% peaks. Not exactly stellar. I have found that latency depends on sample rate, ASIO buffer settings, and CPU load, because if you got lot’s of piggy VSTs and lots of effects going, it takes longer for the CPU to handle it so you have to increase buffer size and the latency goes up.
I’m using (not by preference) a Dante network along with the Virtual Dante Card which is software. I bought a Dante PCie Card (a Yamaha), but have never installed it. Supposedly, you can get super low latencies. But, it all depends on how much loading you put on the processor. If you got lots of VSTs, and processing going on… the buffer will have to eventually be increased to keep the CPU load down.

One thing I have noticed and found surprising… is that some of the loading has to do with how many outputs your mixing to. So CPU load does affect latency because you have to increase buffer as you increase VSTs and effects, but I noticed when I mix from Cubase to 8 sets of stereo outputs, my CPU load drops significantly from what it would have been had I mixed to one 64 bit stereo out.

Think of it… if you have ten mono tracks at 24 bits, you are throwing maybe 240 bits at the mix buss. But Cubase only has a 64 bit summing buss. Does that mean that 75% of the bits from the tracks or VSTs get thrown out? And if so, think of all the calculations that have to be made to accommodate that.

I would like to know any tricks as to how to get Cubase to lower latency. You can’t very well have direct monitoring of VSTs and I’m not using any vintage gear right now.

I think you’re getting a bit mixed up with the bits and where they apply. …
The Cubase mix engine is either 32bit or 64 bit (floating point). Let’s stay with 64bit, which means that all internal calculations of the summing engine are done with that precision, and the result on the output is also a 64bit value.
Now if you import say 2 audio tracks with 24bit (integer) in Cubase. Those get converted to 64bit float in the summing process, because the engine cannot work with mixed bit types.
The summing engine does exactly that, summing. Say your two files at a given point in time have the 64bit values 0.3 and 0.5 (float values in audio are usually between 1.0 and -1.0), the sum would be 0.8 as a new 64bit value, and that’s what you get as the output of your master bus. No bits get “thrown out”.

Anyhow, bit depth has nothing to do with latency. That depends, as you remarked, mostly on your buffer settings, your CPU and the quality/performance of the ASIO driver.
Your latency values seem very high to me. I have 3.2ms in and 3.7 out at 88.2Khz and 256 samples buffer, on a i7-8700K@4.4Ghz with an RME UCX. Of course, if I overload the CPU with more and more plugins, I have to up the buffer size eventually.
I don’t know Dante, but I guess that your Virtual Dante Card driver isn’t very performant, and if you have a dedicated PCIe Card, I’d try installing that and the corresponding driver.

Install the PCIe Dante card. DVS is not made for low latency use.
Latency depends on the audio interface and it’s drivers.

I think the answer is to buy an RME unit.
Think i’ll have to stick with the cheap stuff though.

Thank you fese and all. I appreciate your knowledge.

The reason for my use of Dante… well, I tried the new Aurora and Burl converters. I have two Apogee converters I used with an RME Digiface system in the early 2000’s. I then moved to using a Prism around 2012. I was looking for an upgrade, hoping the Aurora might sound as good as my Lynx Hilo which I use for live piano. I loved the way the Aurora worked, so easy and intuitive to use, so convenient, fit right in with my workflow. But the Burl had the sound, detail, and clarity. I built my PC around around TB3 and picked an ASUS motherboard with TB3 built in. But, Burl doesn’t support TB3, so I had to deal with Dante which for live feeds to the stage would be great… but in a creative work studio, it’s a just adding another very annoying layer of complexity that I don’t need.

So I have the Yamaha Dante PCie card (Yamaha AIC 128-D) and as soon as I can reach a stopping point, I will install it.

I tried uploading a couple of jpegs but I’m not sure this forum allows for it. So, just for comparisons using the same Cubase 11, 96/24 file with Ozone 9 on the output, and buffer set at 512, here’s what I get for latency with Multi Processing checked, 64 bit float checked, and no ASIO GUARD:

Lynx Hilo USB input - 9.333 ms
Lynx Hilo USB Output - 9.333 ms

Dante DVS input - 15.333 ms
Dante DVS output - 25.333 ms

I can knock 5 ms off the DVS output by lowering the DVS latency app down 5 ms to 20.333 ms. Not sure if ASIO GUARD would make a difference. I tend not to use it, nor do I use the Steinberg Audio Power Scheme, or adjust for Record Latency.

ASIO Guard adds an additional bigger buffer, but only for playback tracks that are not record enabled or monitor enabled. It can help optimize CPU usage on plugin heavy projects. Worth trying out if it makes a difference, it might well be that you’ll then be able to lower your regular buffer a bit. But very much workload dependent.
I would actually also advise trying the Steinberg Power Scheme (unless you have created your own one already or modified the default). The default windows power scheme is very much biased towards energy saving, including cpu throttling, and that can in a worst case scenario actually lead to dropouts and performance problems.

1 Like

fese, thanks again. STEINBERG POWER MANAGEMENT is not checked so I checked my power settings, it was set to BALANCED but after checking those settings and clicking APPLY and OK, it still says BALANCED. Maybe I need to restart so it refreshes… I don’t know but rechecking the settings, they look correct. CPU is minimum and maximum 100%, no display, SSD, sleeping. I will experiment with the ASIO GUARD when I get a chance. Right now, I’m harvesting from about 100 DATs and about 400 Cassette tapes. If I use the ASIO GUARD, does that additional buffer cause any latency issues when recording new tracks?

No, because ASIO guard is disabled for tracks you record on, they have the standard latency you set with the buffer size.