Audio popping issue

[MiaShino]

Once again, many many thanks for working with this Mia! Call me masochist but I most enjoy this type of work.. perhaps why I once did it for living. ;P Same same for all others willing to entertain me.

You ask I deliver.

There was some popping I've never heard before towards the end at around 7:50. It did not get picked up by OBS - had no relation to FF14 BGM, sounded completely differently, and it continued even after I closed FF14. Only stopped when I stopped folding with GPU. Pretty sure that was completely unrelated and just due to GPU folding, recording with nvnec and playing ff14.

Very probably as you were putting massive unrealistic strain on your system. For science of course! Though I can offer you something for going through this. It seems you have thermal bottleneck. Something to consider with upgrading being so difficult right now. The CPU throttled quite heavily once it reached peak load. FX is a toasty boy

Anyway... Once again... I'm not an expert - but unless I did magically have overhead when folding proteins, I believe this proves beyond a shadow of doubt that the audio issues are unrelated to CPU workload.... and the audio issues themselves CAUSE the strain on people's CPUs.

Absolutely, that was never in doubt by this lala. A healthy program should never produce such distortions even when system is under full load with minor exceptions. Which it did not, this is good! Distortion only occurred when ambient was muted though to my ears did sound worse then your previous video. This was within expected parameters.

They are the origin, not the cause of CPU spikes and low FPS. Especially given the findings by that one poster here using accessibility settings suggesting that FF14 doesnt ever stop playing audio channels even if you mute them.

Many thanks to that person. Uncertain why I never mentioned this in any of my posts.. too many different things to focus on, I suppose. Heck, I even forgot to mention that Nvidia drivers have higher overhead then AMD drivers. It would appear it is merely sending a stop play sound to the sound driver. This is quite common when sound is not separated into different tracks and while shoddy is no cause for alarm by its self. It suggests that ambient is tied to other audio tracks and stop pause would mute those sounds as well. This was more insightful then I am possibly making it out to be.

I don't know enough about any of this to be able to argue anything... but I really doubt that floating point calculations when folding proteins is less intensive than floating point calculations for ambient right there.

You would be quite surprised as I have seen the lowest of the low, the dregs of horrible coding and it has left me scarred. Example, could write ten line program in Java that crippled your system with relative ease by simply calculating the time it would take for sound wave to travel ten feet. If I then attempted to account for every obstruction in my room, doppler, bounce, etc then calculate vector for that sound on return bounce. Yikes.

It's pretty stupid, that the calculations made for the audio are even made BY THE CPU, rather than the GPU. CPU FLOPS/ROPs are incredibly poor compared to GPUs.

FLOPS in vacuumed is horrible horrible metric and I really wish marketing teams would stop using it. A PS4 has significantly more theoretical FLOPS then the ancient ps3 yet would be absolutely slaughtered by Cell given certain tasks. It reminds me of how in 90's GPU companies and console manufacturers would use triangles as metric for rasterization performance that could never be reached without dedicated T&L unit. Additionally, consumer gpus have neutered FP32 and FP64 capabilities (thanks mostly Nvidia..) and are not as good as one may think at performing certain generic calculations. Audio emulation should be done on cpu, this is okay when it is done properly.

It doesnt seem like it's even about the FX Series weakness with those specific calulations. The way the audio is calulated just seems to be goddamn awful.

No, there is clear problem in how the audio engine is making api calls and one I am unsurprised would effect FX more significantly then other architectures. Moving forward Square must fix this as it will begin to cause issues on more advanced cpus and already has. I hope I have not suggested otherwise in posts! English not so good.

In finishing my giant walls of words and letters, I will come back to this thread perhaps when more coffee is in the system and give conjecture for what Embody team and Square can do should they stubbornly refuse to decouple Embody from their audio engine. Even if it has not appeared helpful nearly everything posted here by you and others has provided much insight. This was quite the fun!

Reminds me of the months spent "fixing" Dawn of war 2 Retribution which suffered from similar issue so friends could play.

Would appear at preliminary glance that the audio engine is spawning too many audio emitters or emitters and sound sources are being treated as spatial when they have no reason to be. This results in cascading number of api requests to audio driver (processed by the cpu) until the entire process begins to stall. Audio driver api begins outright dropping requests as the process can no longer keep up resulting in the wave form deformities mentioned in the article posted. This is done to maintain sanity and prevent application instability which would result in crash of application or sound driver.

You will see something similar occur if application would make too many draw calls, commonly spotted by "GPU driver has recovered from crash" message.

When "muting" sound channels additional requests are made in form of stop play each time a sound would play from that channel (which is does continually). This pushes you from close to maximum number of calls into maximum number of calls which is why muting results in immediate distortion of other sounds.

Square, please move each sound channel into individual audio tracks if you have not as this is not how any sound engine should behave.

Please limit the number of sound sources being processed for spatial audio.

Allow for dynamic adjustment of number of emitters or ability for player to select number of audio channels in sound config menu.

Dynamically scale down positional audio which ties into previous request. These spatial emitters must be unique sound track so they can be muted when obstructed by physical obstacles in game or viewed at different angles that would result in obstruction to and from player sound emitter.

I believe based upon previous encounters with this issue you have more then enough evidence to not only replicate but diagnose this issue on any range of hardware.

Lastly, try to avoid partnering with random tech startups with questionable snake oil applications in future. No no am not suggesting HRTF positional audio is snake oil as the concept of HRTF has been around for decades only that what Embody promised is shoddy at best facsimile.

Many apologies for the ramble.. I stated am not super good at this and hoped some of this makes the sense.