Topic: Difference in latency/performance of different PCIe-interfaces?

Hello everyone! Happy to become of a member of a forum that I've spent years and years reading.

I have this PC that I built about two years ago. It is built specifically for live performances and heavy processing using VST plugins. More specifically to be able to minimize the latency of drum and percussion samplers inside Ableton. It is a Ryzen 5950X-PC with a MADI FX HDSPe (rev 1).

I'd say I am pretty damn happy with what it accomplishes. Right now I am hooked up to a 12mic-D using coax MADI.

Using 96kHz as sample rate with a buffer size of 64 I have a Round trip latency 2.031 ms DA/AD (measured in RTL utility). I can also load the project with quite a few DSP-demanding plugins before crackles and pops start to appear.

Is there reason to believe that any other HSDPe interface would perform worse than this in my next build? Of course other system components could affect this.

Happy to post!

/Ian

2 (edited by ramses 2024-08-27 20:45:52)

Re: Difference in latency/performance of different PCIe-interfaces?

Hi Ian,

welcome to the RME user forum.

All RME products offer similar good performance, no matter if it is a USB or PCIe-based interfaces.
There are differences, but look yourself in the table below, tiny differences, not really significant.

In my blog, you can compare RTL of different RME recording interfaces.
https://www.tonstudio-forum.de/blog/ent … cts-en-de/

Please note: full digital cards have no latency values for A/D and D/A in the driver.
Then you need to add the converter latency of AD/DA converters that you connect via ADAT or MADI.

BR Ramses - UFX III, 12Mic, XTC, ADI-2 Pro FS R BE, RayDAT, X10SRi-F, E5-1680v4, Win10Pro22H2, Cub14

3 (edited by mukundasik 2024-08-27 21:42:46)

Re: Difference in latency/performance of different PCIe-interfaces?

Hello!!!
it's very interesting what you said about your build for live performance(vst server pc) !

would you tell us why you chose AMD Ryzen CPU instead of Intel ?
are there any pros on Ryzen side?

and

which motherboard brand is better for such a build?


I'm going to do the same.
and this topic is also important for me... which of RME PCIe cards will be better for this purpose.

Is RME AoX M\D would be better choice than HDSPe MADI ? (for live DSP\VST\PC server)

Hare Krishna )

4 (edited by ramses 2024-08-28 07:45:36)

Re: Difference in latency/performance of different PCIe-interfaces?

mukundasik wrote:

Is RME AoX M\D would be better choice than HDSPe MADI? (for live DSP\VST\PC server)

I would prefer the HDSPe MADI FX with resource optimizing driver. See my blog article with a review in the PDF file, here are the links:
https://www.tonstudio-forum.de/blog/Ent … Pro-FS-BE/
The full review in PDF format: https://www.tonstudio-forum.de/attachme … -v1-1-pdf/

The AoX M\D is optimized for plenty of channels but doesn't have a DSP for FX and therefore won't get Room EQ
if you are interested in those features. The HDSPe MADI FX is the only card with DSP for FX.

You can see in this sticky thread which HW version of HDSPe MADI FX supports Room EQ.
https://forum.rme-audio.de/viewtopic.ph … 31#p218231

If I may make a suggestion. The HDSPe MADI FX has, besides MADI and analog output, an AES port.
It would be ideal to connect a reference converter to the AES port for the monitoring through monitors and headphones.
ADI-2 Pro FS R BE or ADI-2/4 Pro SE have both support for AES, it's only a matter of features which one you prefer.

More information about the reference converters and their features here:
https://forum.rme-audio.de/viewtopic.ph … 06#p165706

My blog article contains additional information here:
https://www.tonstudio-forum.de/blog/Ent … ses-EN-DE/

This blog article has an interesting Excel calculation / diagram as attachment, here is the link:
https://www.tonstudio-forum.de/attachme … v004-xlsx/
It shows you the advantages of different reference levels (ADI-2 Pro FS R BE four, ADI-2/4 Pro SE even five(!)) in combination with Auto Ref Level. It keeps SNR and Dynamic high over a wise volume range:

Here how to integrate the reference converter into a setup:
https://www.tonstudio-forum.de/blog/Ent … our-Setup/

BR Ramses - UFX III, 12Mic, XTC, ADI-2 Pro FS R BE, RayDAT, X10SRi-F, E5-1680v4, Win10Pro22H2, Cub14

Re: Difference in latency/performance of different PCIe-interfaces?

ramses wrote:

Hi Ian,

welcome to the RME user forum.

All RME products offer similar good performance, no matter if it is a USB or PCIe-based interfaces.
There are differences, but look yourself in the table below, tiny differences, not really significant.

In my blog, you can compare RTL of different RME recording interfaces.
https://www.tonstudio-forum.de/blog/ent … cts-en-de/

Please note: full digital cards have no latency values for A/D and D/A in the driver.
Then you need to add the converter latency of AD/DA converters that you connect via ADAT or MADI.

Wow! your blog is a godsend! Love these types of reads.

The thing about latency for live performances is that it adds up in stage productions.

The latency of the electronic drums look approximately like this with one of the artists I work with.

3 ms: trigger response inside of eDrumin module (using multi-zone drum triggers)
3 ms: MIDI transmission over network (midi from module to iConnectivity mioXL also adds a tiny bit on top of this)
4 ms: output latency (core audio) of RME digiface dante inside ableton (128 buffer size/48 kHz SR. This is omitting the input latency since audio is not being processed).
0.25 - 1 ms: Dante connection (Dante to madi through ex.box
0.8 ms: RTL of Dlive monitor mixer.
1 ms: PSM1000 latency using analog inputs

The values above are all estimates. On top of this I'm sure a couple of ms get added due to AD/DA and digital format conversions.

Total latency guestimate: 13-15 ms

At these latency times the drummer will start to feel the latency when playing. Small reductions of latency at all possible stages can let the performer get out of the "uncanny valley" so to speak.

As I mentioned, there are more places than in the audio interface where latency can be cut. Midi transmission, for instance, is one of them. Directly connecting the MIDI-device via USB to the host computer is the fastest in my own testing.

There is another aspect of audio interface performance and that is the fact that better latency/performance can let me use a higher/safer buffer sizes in a CPU-heavy project before the drummer starts to feel that the triggered sounds are sluggish.

I think it's most likely that I'll get another MADI FX but was curious if any other interfaces have the same crazy performance.

Re: Difference in latency/performance of different PCIe-interfaces?

mukundasik wrote:

Hello!!!
it's very interesting what you said about your build for live performance(vst server pc) !

would you tell us why you chose AMD Ryzen CPU instead of Intel ?
are there any pros on Ryzen side?

and

which motherboard brand is better for such a build?


I'm going to do the same.
and this topic is also important for me... which of RME PCIe cards will be better for this purpose.

Is RME AoX M\D would be better choice than HDSPe MADI ? (for live DSP\VST\PC server)

At the time (2022) Ryzen 5950x had better performance than the i9s and on all processing tests I saw and runs cooler. Intel is once again on top since the 13th generation CPUs.

I use an Asus ProArt motherboard. Works well. I chose it because of the thunderbolt connections. smile

7 (edited by Kubrak 2024-09-03 20:44:14)

Re: Difference in latency/performance of different PCIe-interfaces?

@FullScale
Most of your latency is due to other causes than audiointeface. And the latency of audiointerface is mainly due to buffer size needed. You need either stronger computer (or better optimized the one you use), or less heavy Ableton projects (or better optimised).

You may save 3 ms on MIDI transmision and up to 2 ms using smaller audio buffer. Every 48 samples of buffer add 1 ms to latency.

Re: Difference in latency/performance of different PCIe-interfaces?

Kubrak wrote:

@FullScale
Most of your latency is due to other causes than audiointeface. And the latency of audiointerface is mainly due to buffer size needed. You need either stronger computer (or better optimized the one you use), or less heavy Ableton projects (or better optimised).

You may save 3 ms on MIDI transmision and up to 2 ms using smaller audio buffer. Every 48 samples of buffer add 1 ms to latency.

Yes! Well aware of that. smile

The thing is. There seems to be more to audio interface latency/performance than just the milliseconds. Better interfaces seem to be able to handle higher DSP loads at low buffer sizes as well. For instance, if I use a sub-par interface at a buffer size of 32 I'd be able to get the latency down to an acceptable level but I would not be able to get much out of the processing before the common issues of dropouts and crackles appear.

Also, in the list I provided in the last post I didn't include what the audio interface latency would be if input audio latency was to be processed as well, which is something that is fairly common. So if I were to also process microphone/line signals going in to Ableton we would about double the latency.

I just want to make sure that whatever interface I get next is able to provide at least the same performance as my MADI FX HDSPe. Optimizing every part of the chain is what I'm looking for!

9 (edited by upton2 2024-11-26 09:50:42)

Re: Difference in latency/performance of different PCIe-interfaces?

FullScaleAV wrote:
Kubrak wrote:

@FullScale
Most of your latency is due to other causes than audiointeface. And the latency of audiointerface is mainly due to buffer size needed. You need either stronger computer (or better optimized the one you use), or less heavy Ableton projects (or better optimised). block blast

You may save 3 ms on MIDI transmision and up to 2 ms using smaller audio buffer. Every 48 samples of buffer add 1 ms to latency.

Yes! Well aware of that. smile

The thing is. There seems to be more to audio interface latency/performance than just the milliseconds. Better interfaces seem to be able to handle higher DSP loads at low buffer sizes as well. For instance, if I use a sub-par interface at a buffer size of 32 I'd be able to get the latency down to an acceptable level but I would not be able to get much out of the processing before the common issues of dropouts and crackles appear.

Also, in the list I provided in the last post I didn't include what the audio interface latency would be if input audio latency was to be processed as well, which is something that is fairly common. So if I were to also process microphone/line signals going in to Ableton we would about double the latency.

I just want to make sure that whatever interface I get next is able to provide at least the same performance as my MADI FX HDSPe. Optimizing every part of the chain is what I'm looking for!

How does the processing of input audio latency, such as microphone or line signals, impact overall latency in a digital audio workstation like Ableton, and what strategies can be employed to minimize this effect?

10 (edited by ramses 2024-11-26 17:30:33)

Re: Difference in latency/performance of different PCIe-interfaces?

Mic---A/D---UFX III---USB3---DAW---USB3---UFX III---D/A---Monitors
          ^--------------------------------------------------------^----------------- converter latency
                                   ^--------------------------^--------------------------- ASIO buffer size / driver quality


Mic---A/D---12Mic---MADI---UFX III---USB3---DAW---USB3---UFX III---D/A---Monitors
          ^----------------------------------------------------------------------------^----------------- converter latency
                                                            ^-----------------^------------------------------------ ASIO buffer size / driver quality
                                ^------------------------------------------------------------------------------ Transport via MADI per MADI device
                                                                                                                                     3 samples @single speed
                                                                                                                                     6 samples @double speed

To minimize RTL (round trip latency including AD, transport to/from PC, DA)
- get devices with good drivers
- highest impact on RTL/latency: (ASIO) buffer sizes from ~2.x -9x ms
- much lower impact on RTL/latency: converter latency nowadays a fraction of a milisecond.
- lowest impact: sample rate: the higher the sample rate the lower the converter latency, but higher compute load on the PC

in use cases where this matters, for pure recording use high buffer sizes
- choose devices with up to date converters which have

Regarding transport. Also transport over ADAT only takes a few samples, maybe 3 at single speed.

USB vs PCIe: depends on the PC whether driver quality is good and thus DPC latency low.
On a good machine there is no remarkable difference between using USB or PCIe/TB.
I ran CPU-Z load test and a playback of a DAW project, CPU at around 98% and no audio dropouts.

Regarding input/output latency / RTL of different RME solutions that I had see my post #2 in this thread.

The full RTL you have only when recording and monitoring something. If you play a MIDI keyboard then the A/D part and transport of audio to the computer is not applicable, then you have approximately "half" or RTL.

Overall … all RME products give you excellent values.
Important is that your computer's cores are fast enough and that the drivers are not blocking CPU cores for too long (low DPC latencies).

BR Ramses - UFX III, 12Mic, XTC, ADI-2 Pro FS R BE, RayDAT, X10SRi-F, E5-1680v4, Win10Pro22H2, Cub14