Presonus Quantum 2626 "zero latency" audio interface

[Terlisimo]

This is my personal impression/review of the Presonus Quantum 2626.

I bought it because I wanted to have lower latency than USB interfaces provide. My use case is playing SW synths with a MIDI keyboard.

USB latency is ~5ms + at least 1.5ms for buffers (64 samples @44.1kHZ) and I thought that wasn't bad until I tried playing a real HW synth. It felt so much better and immediate! But I wasn't prepared to start collecting analog gear just yet and I kinda like my SW synths. That made me look into ways of lowering latency.

So I figured, if USB is the bottleneck then maybe PCIe sound cards would do better.

Apparently they do, but they come in 2 flavors: gaming focused cheap cards like Xonar/Audigy that lack common features like digital output and ridiculously expensive/overkill (for my needs) proffessional gear like RME and Lynx.

I did a bit of research and bought a Presonus Quantum 2626. It's a $600 thunderbolt audio interface and because I don't have a Mac I had to get an add-on Thunderbolt card for my PC which was another $100 on top. And a freaking $50 Thunderbolt cable (still sore about that).

Features up to 24-bit/192kHz operation, analog and digital (spidf + adat) i/o, phantom power, midi. All analog inputs and outputs are TRS/balanced. See manufacturer site for full specs if you're curious.

The promise was <1ms roundtrip latency and the Quantum 2626 delivers. Well, the software says it does, and it sounds like it does, but I'll probably measure that with an oscilloscope just out of curiosity.

Tried playing a guitar through it and it felt pretty much real-time.

In the end it is still overkill for my needs and it didn't cost much less than some RME/Lynx thing might have but it has features that I'll actually use, like midi i/o, phantom power for mics and the ability to plug an electric guitar directly into it. Also a nice big physical volume knob

Lowest buffer size is 16 samples and the simple VSTis I tried played without pops/clicks (i7-12700 cpu). Heavier ones required a 32-sample buffer for smooth playback, which is still quite good.

The interface comes with its own ASIO driver for Windows.

I've read reports that Quantum 2626 can't be used as a regular sound card for Youtube, Spotify, gaming etc, but it works just fine (on Windows 10 at least). Maybe previous driver/firmware revisions didn't support that but as of 2023/03 they do. Basically each pair of line outputs is presented as a separate sound card.

I can't speak about the quality of DAC/ADCs. They sound good but I'm not an audiophile or a sound engineer so my bar isn't all that high.

I have a few grievances with the Quantum 2626.

1) you can't do pass-through of inputs to outputs of the interface. It has to go through the software stack. I.E. it is useless without a computer.

2) there is only 1 thunderbolt port on the device so you can't daisy chain anything to it, I.E. it has to be the last device in the chain. Not a big deal for me since I don't have any other TB devices but still.

3) doesn't work on Linux and probably never will

Also, there was a gut-wrenching minute when I clicked "firmware update" and my PC bluescreened in the middle of the process. The update worked however (phew!).

Anyway. If you have a USB soundcard and playing your midi keyboard sounds just fine to you, don't ever try playing a HW synth

Note to PC users: the card requires a real Thunderbolt interface and won't work over usb-c. From what I can tell, Thunderbolt add-on cards are manufacturer-specific and require motherboard support (has to have a JTB header connector). I have a MSI B660M motherboard so I bought the MSI Thunderbolt 4x add-on card.

I will answer any questions if anyone is curious about some aspect I did not cover.

[rasmusklump]

I have hardware synths, organs and a piano, but nothing feels wrong or delayed when I play a softwaresynth with my USB soundcard. And the bottleneck certainly isn't USB 2.0 but the efficiency of the driver

[Terlisimo]

And the bottleneck certainly isn't USB 2.0 but the efficiency of the driver

Yeah, I suppose some USB cards are better than others. According to this, not many have <5ms latency:

https://gearspace.com/board/showpost.ph ... count=5092

[Funky40]

good read, thanks !

I run a Quantum2626 too. On a apple silicon mac.
it was the only affordable solution i found, vs. small(-est) latencys.

i run to 99% realtime-play patches in GigPerformer.
Mostly keyboard- VST-instrument based. Sometimes Bass guitar roundrip.

the whole topic is a little deeper as reflected in the OP, assoon you *really* want to do the realtimeplay-thing,
since there are most often other -software based- contributors to the overall latency !
Means as a take out: you *really* just want to have the fastest AI possible. period

i personally mostoften do not go lower than 128samples. Sometimes 64 samples.
But it became rarer and rarer.
What i figured out is, that my old M1mini ( with just had 8GB of Ram) had a b uffer/latency sweetspot area @ 64-128 samples. I would guess thats still same with my 16GB RAM M2pro.

once i played Pianoteq in Studio1 with 32 samples buffer.
It was so fast, it nearly felt unnatural, haha.
it was IMO faster than most Workstation Hardware would be.

alos have i played bass thru my M1mini.
and when i once played it thru my modular, thru a ( much hyped + expensive) Synthtech E520 FX,
i found the module had more latency than playing roundtrip thru my Mac.
Made then some tests and checks, to be shure to use similar algos ( vs. FFT based for example).
....sold the E520 afterwards.

Realisticly speaking , i would say, i´d deal usually with something like 12ms latency !
thats noticeable, but acceptable for me.
At 6ms, things feel really good !.....lower, and it feels like hardware or even better..
at around 3ms it feels stupidly fast ( for Piano playing) . Faster than hardware, imho.

to notes:
the AMP-Sim plugins i have and used (PA) would add some latency too. ( mostly around 4ms, in my case)
so for Bass and guitar, i´d suggest to leave out Amp-sims. Take a good HW tube-pre.
Most limiters add latency too ! .....often higher than you would expect.
Take the TDL limiter 6 GE. This has a low latency mode.

you quickly have 6ms or so -unwantedly and unnecessarly- added just for some basic plugins.
THATS why it is so important to chose the fastest AIs possible.

important note:
personally, i have to this day crackles with my Q2626.
I never could get rid of completly.
But i live in an old appartment house, with no real Ground, compared to modern mains power standards.
I stumbled over a post, where some guy mentioned crackling too, and he was refering to his old house with no real ground as the reason. So, could be, its just that.

i also run a NI komplete Audio6, which has waaaay more latency ( at same settings).
But it does NOT have the crackles.


i do quite complex patches in my host of choice (GigPerformer) with LOTSA realtime FX manipulation !!
i eat up one core with my patches, even on my M2pro-Mini, and would say from what i experienced, that a slower AIs might be less prone to crackles.
can´t afford a Lynx or top RME, so, i never will be able to compare with these.

[Funky40]

i´d like to have numbers vs. the MOTU M4 !

since i´d like to replace my NI komlete audio6 !
would like to avoid to buy another Q2626, and keep it more affordable.

if the Q2626 could be stacked, it would be no question, though.
yeah, there it lacks.


and: thanks for the link @Terlisimo ! i searched for such the other day

[kritikon]

I have hardware synths, organs and a piano, but nothing feels wrong or delayed when I play a softwaresynth with my USB soundcard. And the bottleneck certainly isn't USB 2.0 but the efficiency of the driver

Likewise. I have latency way higher than 32 samples and it feels fine. Similarly I have quite a few hw synths and never noticed any obvious differences for a long time. IIRC I have 256 samples, might be more but it works out at around 7mS and that's no issue at all for me. And at 192kHz...kinda asking for problems IMO. OP must have the hearing of a bat if he needs unreal-world latency and 192.

[bmanic]

For really low latency I always recommend RME soundcards, if you can afford them (even a basic Babyface or 2nd hand). You can easily run them at 32 or 64 sample buffers and their AD/DA conversion is also very low latency.

You can sort of "feel" the difference between 3ms and 6ms but it's not a huge leap. Plenty of hardware stuff have way worse latency (AKAI MPC Live/Live2/Mini/Force is around 15ms).

If you need absolutely "zero" latency then Analogue is the only way you can get that, provided that the envelopes are fast enough to open (most are though). All current digital systems have comparable latency (from 4 to 10ms for most).

[Tj Shredder]

I doubt that USB is ever the bottleneck. It could be drivers, OS settings and such.
Anybody remembers that 6 channel audio interface from Emagic? That was USB 1!
I always preferred Firewire 400 over USB 2, but we are now on USB 3.x and that smokes these older protokolls by magnitudes… I still use Firewire interfaces, for example my trusty Quatafire. With my last Macbook Air, I needed to go to a vector size of 256 or even 512. Now on my M2 MB Pro, 32 isn‘t an issue. That is firewire 400…

[Jim Roseberry]

RME has been known for rock-solid low latency performance... for decades.
That said, when it comes to lowest possible round-trip latency, Presonus Quantum is currently one of the best choices available.
ie: At 96k using a 32-sample ASIO buffer size, round-trip latency is 1ms (measured).

Thunderbolt yields PCIe level performance with an external peripheral.
While RME USB units are very good, they can't achieve round-trip latency equal to the best PCIe/Thunderbolt units.
Better USB audio interfaces can achieve sub 5ms round-trip latency.
Few can achieve sub 3ms.

USB-3 offers more bandwidth (vs USB-2), but thus far I haven't measured a single USB-3 audio interface that offers lower round-trip latency than the best USB-2 units.
The RME Fireface UFX III claims sub 3ms round-trip latency. (haven't yet measured it).
It will not go as low as Presonus Quantum or Antelope Orion Studio Synergy core at a given buffer size (due to the limits of USB).

All the connectivity mention above has FAR more bandwidth that what most users need.
Bandwidth in and of itself doesn't make a peripheral faster.
ie: If you have a car with a top speed of 100mph, it doesn't matter if you put it on a two-lane highway... or a 16 lane freeway (more bandwidth), the car can't go faster than 100mph.

Lowest possible round-trip latency may or may not matter to an end-user.
If you're trying to push performance limits, right now... that means using either PCIe or Thunderbolt.

[Funky40]

Lowest possible round-trip latency may or may not matter to an end-user.

the thing i´m in, IS realtime performances based on patching in a plugin Host.
usually, just one way => out / sometimes roundrip with Bass.

a total latency of around 6ms feels good !
I´d claim: nobody needs really less.
4-5 mS is in my opinion where hardware FX units for example would come to sit ( just a guess).
So, at 4mS latency are we reaching a level, where we can stop any discussions.


just, as i have posted: if you begin to play ITB based, is it with instruments, or FX alone,
are you quite rapidly dealing with pieces of Software that would also introduce some latency on their own.
* There* happens the cause, where minimal latency AIs become to be "a topic" fresh again.


and btw: dealing with modern FX, doing GREAT patches, for example for live performances, as contrary to production work, nobody would ever setup a buffer size of 32 samples. To me, thats totally unrealistic ! ...and even more so at 96khz.

at 48khz, 64-128 samples is a realistic number, 256 samples at 96khz, 512samples at 176khz.
Thats then the small numbers i would set up, vs. these given sample rates.
The manufacturers claims are paper specs. Its just not how real live happens vs. realtime-play performance tasks, inclouding FX work.

Realtime-play FX mangling, will begin to crackle quite rapidly at those posted small buffer settings. Depends on the FX. But to the bigger part, for good, and nicely jammable FX patches, these (manufacturer) numbers are totally unrealistic.

even my (small) numbers posted, are often enough unrealistic !...depends on the FX patches. at some point, the data bus speed of the computer becomes a topic too, btw.


the whole latency topic vs. realtime-play (FX based) performing musicians, is way more complex, than anywhere told.
the short one: watch also your computers overall databus speed !
it "can" become another factor !
i for example would run with my patches there into a bottleneck with my old M1mac.
which requiers to rise buffer settings,

which rises the question:
if we do NOT deal with smallest buffer settings, is *then* a USB AI still a smidge slower ?
lets say, a buffer of 128 samples/48khz and up

[Terlisimo]

if we do NOT deal with smallest buffer settings, is *then* a USB AI still a smidge slower ?
lets say, a buffer of 128 samples/48khz and up

Looking at that latency report pdf, I would say yes. The fastest USB AI tested at 128 samples is RME Fireface in USB mode (supports both FW and USB).

Says 3.923ms output latency at 128 samples. Does not state at which sample rate, so let's assume 48KHz. 128 samples at that rate is 2.66ms, so the USB adds at least 1.3ms on the output. And this is with a top-of-the-line $2k+ interface. Most other USB AIs listed have at least 3-4ms higher latency. Best AIs in terms of latency are all PCIe and have <1ms latency.

It does seem that PCIe/TB > FW > USB. PCIe/TB latency is probably measured in nanoseconds, and I know FW has direct memory access while USB is a packet based protocol so it will always be slower.

One more data point... I found that, in addition to having almost zero inherent latency, the Quantum 2626 tolerates lower buffer sizes a lot better than my previous one. For example, an fx chain that required at least 256 samples on USB works just fine with the Quantum at 128 on the same hardware. I can't tell whether this is because of USB vs. TB or just better drivers.

You can sort of "feel" the difference between 3ms and 6ms but it's not a huge leap.

Agreed, that about sums up my impression.

To me, anything above 12ms becomes unnerving. 20ms is literally unplayable. The 8ms I had with my previous AI sounded okay and it IS okay, our brains can adjust to that. If you asked me then how I felt about latency, I'd say it's pretty much not noticeable. But as I've said, then I tried a HW synth and it felt just... better than okay Like it was easier to play, especially something with fast staccato notes.

So how much important the interface latency really is? Well, if your speakers are 10ft away then that's 9ms of additional latency just because of speed of sound (~1.1ft/ms). One could shave that off simply by using headphones instead. Another factor is the kind of instrument you're playing. If you're playing an electronic drum kit, you will notice latency much more than with some warm long-attack synth pad. The initial preset on that HW synth I tried was a pure saw with zero attack so that probably contributed to my impression.

Is it worth it to spend hundreds (or thousands) of dollars to eke out 3-4ms lower latency? Debatable.

To me, that HW synth gave me an itch which I had to scratch. The point of my original post was that if you're hung up on latency for any reason, the Quantum 2626 is a real bargain at $600. Faster than anything cheaper, and cheaper than anything faster

[Terlisimo]

So, at 4mS latency are we reaching a level, where we can stop any discussions.

Heh. Yeah, 3-4ms would be the natural latency of a real piano. Going below that would probably sound unnatural.

Although now that I think about it, I *will* try using headphones with a 16-sample buffer and some cheap synth just to hear what that is like.

[Funky40]

if we do NOT deal with smallest buffer settings, is *then* a USB AI still a smidge slower ?
lets say, a buffer of 128 samples/48khz and up

Looking at that latency report pdf, I would say yes. The fastest USB AI tested at 128 samples is RME Fireface in USB mode (supports both FW and USB).

Says 3.923ms output latency at 128 samples. Does not state at which sample rate, so let's assume 48KHz. 128 samples at that rate is 2.66ms, so the USB adds at least 1.3ms on the output. And this is with a top-of-the-line $2k+ interface. Most other USB AIs listed have at least 3-4ms higher latency.

Thank you @Terlisimo to picking up on this !

hard to estimate how much more latency a Motu M4 would have, vs. the Q2626.
I´d place it -just by guessworking- at 3ms.

2ms would be quasi negiligible (for me), 4ms absolutely perceivable, but 3ms ? ...Hard decissions, haha

Guess, the pricetag vs. NOT having a digital in-out will determine my final decission.
plus formfactor.......
BUT: i´d expected more such AIs to hit on the market.
why no small brother to the Q2626 ?....its overdue, imho.

[Funky40]

So, at 4mS latency are we reaching a level, where we can stop any discussions.

Heh. Yeah, 3-4ms would be the natural latency of a real piano. Going below that would probably sound unnatural.

in the past they sayed that normal folks would not hear differences below 7ms.
and iirc, thats also the numbers they gave/estimated for most digital Pianos / workstations.

my super fast patch with just 32 samples in studio1, probably at 96khz, felt indeed shocking at first.
Cause it felt so unnaturally fast.
Try it! its definitly a funny experience.
( yes, i often wear HPs. might have played a role too )

[creart]

Hi I'm thinking about getting a Mac Studio and Quantum2626..

Will I be able to run Pianoteq (or other standalone software) for instance without having to run it from a DAW??
thanks in advance?