Thanks for your reply!

It seems to be the nvlddmkm.sys that pushes it over every now and again but the subtle crackles do still occur even when LatencyMon still reports "Your system appears to be suitable for handling real-time audio and other tasks without droputs".
If I leave LatencyMon running for long enough Nvidia can reach ~1500 at times but when at the desktop it can be ~500.

Yes - This was only a troubleshooting step as it's mentioned in a few places that hyperthreading can have a negative effect and contribute to the kinds of issues that I'm experiencing. It's not disabled currently; It was just to cover all bases whilst gathering evidence.

Outstanding: Core Parking, Platform Power Management, PCIE ASPM, XHCI Hand-off and exclusive mode. Otherwise optimal settings already applied.

I believe it's currently connected to a port that runs through one of the two chipsets. I tried connecting it to one of the two ports wired to the CPU and it made no difference.

Please run LatencyMon on an IDLE system for around 5-10 minutes, best without any keyboard/mouse activity.
LatencyMon already creates a DAW-typical load.
If you booted the system fresh and just logged in, wait until CPU activity goes down (typically 1-2 minutes).

To observe CPU-, GPU, and network load, I installed 8Gadget Pack and some older "not-bloated" gadgets from orbmu2k
https://orbmu2k.de/category/sidebar-gadgets

It's very useful to see what's going on on your system. The Power Supply Gadget allows you to switch manually between the 3 power profiles: High Performance, Balanced, and Energy Saving. The CPU gadget shows you the utilization of every core (also temperature if you install the additional driver; see settings). Also the memory utilization. It's interesting to see how an application under load utilizes the cores. To see the correct CPU temperature you might have to adjust the field TjMax (°C).
I compared the temperature of each core with the values reported by "CPUID HWMonitor". See screenshot at the end.

What do you mean by "wired to the CPU"? The USB3 and USB2 ports are coming either from the chipset or from a 3rd-party USB3 controllers. If you look at the UFX III manual, ch. 40.3, then you can read that USB ports from Intel or AMD chipsets are compatible with USB 3. But not all 3rd-party USB 3 controllers.

If your mainboard should have a 3rd party USB3 controller on board, then those are in most cases from ASMEDIA, and these are not fully compliant with the USB3 transfer mode that is being used by the MADIface driver.

Therefore, you often see the recommendation in this forum to "try all USB ports". It is much easier to write it this way. Otherwise you have to consult the mainboard manual, and sometimes it is not easy to find this info.

Another recommendation is to update your AMD BIOS. In the past there were issues around USB that could be fixed by performing a BIOS upgrade.

https://www.dropbox.com/scl/fi/ro49clvf … m&dl=1

May gadgets on the right side of the desktop, see this screenshot, you have to click here (picture is too tall):
https://www.dropbox.com/scl/fi/av7wqfpn … q&dl=1

It depends on, what happens in your computer. It looks like graphics might be the cause... So, I would start with nVidia SW/drivers. Or drivers for USB. I do not have experience with Lasso, maybe Ramses could tell you more.

And what about to use Lasso on LatencyMonitor? And try if there are different results if LM is restricted just to cores with x3D and just to cores without.

----------------

Sidenote, AMD has issued CPU with extended L3 cache on the both chiplets. So, all cores have 3D L3 cache.

Sure, two 3D caches would solve just one of two problems. It would solve the problem if given thread reasonably benefits from bigger cache and is moved to chiplet without 3D cache. All cores would have it, would be even in this respect.

But sure, L3 is not shared between chiplets, so copying data to L3 on target chiplet would be still needed.

Ramses, does Process Lasso work also for AMD CPUs, or is it only Intel targeted?

I might look into Process Lasso. It would be nice if it was a simple swap to be able to disable "Game Mode" and just use Process Lasso in it's place. But it seems a bit tricky to get working consistently with some anticheats / game launchers.

Some interesting findings from today...

I haven't done much in terms of checking my recorded audio for clicks / pops etc, however I do believe I may have made a small discovery in my pursuit of lower latency. Honestly I can't believe these two individual items didn't even occur to me prior to now.

HWiNFO64
I had this running in the background constantly for the sole purpose of having my CPU / GPU temperatures in the tray area. I have stopped this from running in the background.

FanControl
This is another piece of software I have in the background constantly polling. I noticed in the settings that you can disable the "GPU" section of LibreHardwareMonitor. You can also disable the NVAPI integration. I don't do any fan curves for my GPU so I disabled those two items too.

LatencyMon results
So exactly the same system, balanced power plan, no processor min/max changes. NVIDIA performance mode normal. No other changes made to the system other than the above.

~30 minute run:

~15 minute run (much later in the evening):

In fact, the latency warning only occurred after I used the snipping tool to make these screenshots. Though I did disable the "prefer max performance" in the NVIDIA control panel as I hadn't seen much difference - So that could be part of that.

Interesting results nonetheless. I'd never make it past the ~5 or so minute mark prior to this let alone 30+ minutes. Every single time it would be the NVIDIA driver or the NTOSKRNL.

Edit: It appears I spoke too soon. Trying more runs and 6 minutes in I get hit with 1100 nvlddmkm.sys.

Edit Edit:It does seem to at least be more consistently able to have good periods at least.

Well to be completely honest, I expected to be able to do both. I didn't think that was an unrealistic expectation going into this rabbit hole. It just so happens that the processor that I have really needs to be used with the Balanced power plan for the scheduling to work properly which is unfortunate.

All that is immediately clear to me is that everywhere I have read has indicated that specific power plans need to be used for X3D processors to work effectively; especially those with dual CCDs. I won't pretend to understand the ins and outs of any of it.

--

Anyway - After some fiddling today I did discover that the most important combination on-top of what I already had was to disable C-States, set the NVIDIA power mode to high performance and also set the Windows power plan to high performance or a modified balanced performance with min/max processor 100%.

No other combinations of things made a substantial difference - Including temporarily disabling the frequency CCD, setting high priority / affinity for audidg, etc.

High Performance Power Plan, NVIDIA High Performance, CCD1+CCD2, C-STATES AUTO

Balanced (min/max processor 100%), NVIDIA High Performance, CCD1+CCD2, C-STATES Disabled

The latter image reflected similar results for the High Performance power plan.

--

Ultimately my issues with latency are a combination or my processor taking a nap or winding the clocks down and the same with the GPU. I still get massive spikes from the NVIDIA driver when not actively sat idle on my desktop but they crop up less frequently during idle LatencyMon tests and the highest process latencies are better generally speaking.

I just need to do some in-depth reading into power plans and requirements for my CPU to know of any potential issues this could introduce with the X3D stuff. But leaving this information here in case it's of any use to anyone in future.

Not tested for crackles yet - Job for another day.

> With Process Lasso Pro, you could try to get rid of AMD Game Mode and to run with all cores active. Then to place audio-related processes to the CCD/cores with the larger 3D/V-Cache.

Unfortunately I don’t think it’s quite as straight forward as that with my current setup. In terms of keeping the audio related services on the V-Cache cores, sure. I could definitely have a setup for games and a more optimal setup for DAW related stuff.

But there are for example some games that are whitelisted by AMD to stop the processor from actively stopping CCD1 etc and some anti-cheats supposedly don’t play nicely with hard setting the affinities of the games using taskmgr or process lasso. AMD also doesn’t expose the database that they use for the driver either from what I can tell.

There doesn’t seem to be a universal one-size-fits-all optimal setup for my use case / current hardware.

I’m considering moving to a symmetrical or single CCD CPU in the near future. When I bought the 9950X3D originally I wasn’t half as invested into audio/audio stability as I am now.

It’s also entirely possible that the crackling I’m hearing are more mouth clicks than digital artefacts / crackling. That being said obviously scrubbing media in Windows 11 is definitely a huge issue.

Need to continue looking into it as it could well be that latency, be it higher than it could be, isn’t the source of the artefacts that actually hearing.

What do you mean by this statement? That the crackling might be physical clicking of the mouse, and not the sound from loudspeakers?

I'm not actually sure in all honesty! I haven't used Intel for years and years. I'm not sure whether they are so dependent on a specific power plan for the scheduler?

I'm also not sure what happens with the new processors but I have heard of the 9950X3D2, yes. Presumably with the dual V-Cache CCDs there's no need for parking in the way that there is for my current asymmetrical CCD layout. It's certainly an option worth considering.

So I'm monitoring through my headphones as opposed to speakers, but what I mean to say is that the artefacts that I'm hearing in my spoken voice recordings could be due to organic noises created by my mouth / jaw etc as opposed to something that's making it's way in through EMI, buffer underruns etc. They sound very similar to audio hitches and supposedly it's easier to tell within the waveform.

I'm going to do some more tests today with humming etc so it hits the mic with the same kind of resonance as speaking etc but it could potentially be what's going on here.

--

Would be the preferred outcome, of course... and mean that I've unfairly put it down to an issue with the RME unit / my hardware chain.

The issue with scrubbing content on Windows still exists even with all of the optimisations made to latency so I'm guessing it's just as good as it gets when it comes to Windows.

OK, a microphone may capture other sounds than desired ones.... But that is another story... Those are not clicks in audio playback.

Up to now, I have expected you have problems with clicks in playback that does not contain clicks...

Mouse vs. Mouth... Quite possibly, yes. Is there an uploaded audio sample somewhere?

And, does not get your mic mixed in? At least at the start. What about to try it with loudspeakers or with headphones without mic?

I have Win10, which may be different, but even using WDM is generally without problems. For sure no extra sounds at very start. I doubt it would be worse on Win11.

Never ever had such issues with a UFX, UFX+, or UFX III. ASIO and WDM were always super stable with 32, 64, or 128 samples buffer size (at single speed). Higher than that I never needed.

It was ~2 years ago that I tried Windows 11 by using a Rufus-tweaked installation image, so that the installation worked on my non-compliant HW (in terms of non-compliant CPU, no TPM, Secure Boot disabled).
It worked satisfactorily at that time, but I can't comment on the latest version(s) of Windows 11.

Such audio drops that I heard in your audio example sound to me exceptional, and should not happen.
Maybe some hiccups because of some upgrades? Did you ever try a vanilla Windows 11 installation?
Its always good to have a spare SSD for such tests, in case its needed.
Thats the best way to find out whether the current installation is hosed.

Scrubbing in a player is the worst possible demo as the clicks on repositioning can be fully normal and coming from the player itself.

Did you ever had the Settings dialog open to check what happens when you start songs and hear the crackle? I would not be astonished if a sample rate change happens in such a moment.. That can be seen on the UFX III display as well, maybe even better (real-time).

Also you could play the test tones in Sound control panel, tab Playback, Advanced - Test button. They should sound clean, from start to finish - as long as there is no sample rate change and the buffer size is not critical

Which buffer size is set in the Settings dialog? WDM devices sometimes need 128 or 256 explicitely to work correctly. Depends on the use case - most people use ASIO all the time and only partially something with WDM.

Let me tell you what I hear (and this is not RME related, my Apple Dongle and Realtek mobo soundchip do the same):

- small pop on start, then noisy test tones (most probably low bit resolution in the original file). When changing from left channel to right channel the end on left has a pop (mostly, sometimes not. The system player isn't the greatest invention...). Same when the right channel ends (here always as the played file ends). This is a typical sound for DC in the audio. Any way - there is zero (!) crackling, which is defined as multiple clicks in a short time (krrr sound).

See above. Same, including the pop when repositioning - fully normal and expected.

I don't hear any crackles in that recording. If you hear crackles instead of single pops, then they seem to happen during playback, on your setup.

I don't hear them.

Correct. I do not use Spotify but don't think this is an issue at all. My players (like AIMP) include soft-fade options when scrubbing, which sounds perfetly fine then. For things like this only the player is responsible, not the interface.

What I meant is that there is an upper limit for WDM. Try 64 and 128 to see if it changes something.

Sorry, I missed that you were specifically talking about "scrubbing."

Regarding the pure playback in the first two of your last three demos: I also don't hear any distortion. That single click you hear when stopping a test tone randomly is actually expected. As for the scrubbing quality in your third demo, I think that really comes down to how the player itself handles those operations.

I tested this with MusicBee, which is a very good player with ASIO support. I tested scrubbing with ASIO and WASAPI (exclusive and shared) drivers. WASAPI uses the RME WDM driver, and I found no issues when scrubbing through the audio file.

I used a 128-sample buffer at single speed, but I also tested the smallest and highest buffer sizes in the RME driver. It makes no big difference; everything sounds smooth. I configured MusicBee to load the entire audio file into memory, which seems to make skipping through the file a little bit smoother, though it is a very subtle enhancement. To sum up, with that player, I do not see any issues at all regardless of whether I use ASIO or WASAPI (WDM).

The key mapping function in MusicBee is also very useful. I mapped 5-second skips to the left/right arrows, 20-second skips to Ctrl + arrows, and the spacebar for play/pause. It provides a clean sound and great functionality without needing the mouse—though even when I do click to a new position like in your demo, the playback stays smooth.

Thanks for all of your responses - I really appreciate it. I've read through all of them, but going to try to reply to the relevant parts to keep the post as concise as possible.

First of all apologies for areas where I perhaps haven't used the correct terminology for what it is that I'm hearing at specific times. What I do hear is a mixture of pops and crackles. I can't definitively say exactly when each appear as sometimes the audio will start with a crackle, other times it will be more of a pop. Sometimes they mix together.

I'm very, very confused as to how I can hear these items so clearly on my end but it doesn't seem possible for others to hear.

Ah, I see! Apologies for getting this mixed up. I'll have a play with it.

This issue for me isn't limited to a specific player. It happens everywhere; media player classic, Windows system alerts, Spotify, Games etc. I used the media player classic extreme scrubbing example to try to show exactly what kinds of sounds I'm hearing just simply using my OS.

Thank-you to both yourself and MC. I wasn't aware that this was a thing in all honesty. This has been a huge learning curve for me.

---

As a final test, if anyone is willing to briefly listen - I have prepared two wav files. Recorded with Reaper (ASIO) from ADAT1+2 (where Spotify sends audio to via Windows Sound Mixer).

https://drive.google.com/drive/folders/ … aP-L6vlYJW

Recording 1

• A little more and more representative of the frequency and intermittent nature of this.

• It's the more tame of the two examples; what you are looking for are subtle pops behind the beginning of the intro in a couple of places. More so near the start and the end of the recording.

Recording 2

• This is the kind of sound that I hear most often when simply using Windows, web browser, Spotify, games etc.

• Hopefully the one that's a lot easier for you to hear.

--

I do want to say thank-you again to all of you who have offered to help or simply taken the time to read this post. I knew absolutely nothing about thing such as DPC latency or how to even begin to optimise my PC for audio prior to making this so I do really appreciate all of your feedback.

This is so strange.

I haven't got a second pair of headphones to try currently (potentially another variable as to what's making it worse?). Perhaps the when the audio initially kicks in they could be adding to it a little within the headphones themselves?

I did however play these wav files from my iPhone and in the second one in particular I can hear an audible click between songs even without headphones. I guess it's more so the volume of the clicks and brief crackles that can/do appear in Windows 11 that's rather jarring compared to the Mac.

I also tried very briefly Windows 10 IoT Enterprise on the same PC/hardware and although again the pops at the start of system sounds were there, Spotify playback / scrubbing sounded more similar to how it appears on the Mac.

If this is just how things are when sounds start / stop, change between streams etc then I will put it down to that or potentially just "Windows things" or just how these things work. I'm very grateful for the input from everyone that has at the very least confidently ruled out any physical issues with the interface.

Perhaps these artefacts are just to be expected with starting/stopping audio streams. There's definitely still more work I could do in terms of lowering the DPC latency etc so perhaps I'll focus my attention more in that area.