I have replaced my graphic card recently, had an nVidia GTX 750 Ti and got a second hand AMD RX 460. It was slightly defective so I got a new RX 560D shortly afterwards.

These test runs confirm what I have read previously from various sources that nVidia's driver is problematic for audio production. I am thrilled with the current card's performance and now I can run more plugins without clicks or crackles at lower latencies than before so I thought I could share my findings with you.

Here are the results:

1: nVidia GTX 750 Ti





2. AMD RX 460





3. AMD RX 560D

I wonder why?

As you can see from the data nVidia's driver is more aggressive and affects other processes negatively.

what does aggressive mean with regards to Graphics cards?

What is DPC Latency?

thanks, I'm just curious what aggressive means. I never heard that term in this context

Aggressive would translate to high execution, as you can see from the pics above.

Thanks for the video, but I already did what he recommends, used DDU, disabled all the unnecessary things like GeForce experience, telemetry and other nVidia stuff I don't use as I am not a gamer.

In fact, I do not recommend installing newer nVidia drivers because, more often than not, the older ones gave better DPC latency results like the driver I was using from April 2017 (381.89 WHQL).

Looks like the RX560 is a winner for a new audio PC.
Thanks for the test.

For anyone interested Microcenter has 8 core Threadripper at 3.8 ghz for $379 and - another $30 off if you buy a compatible motherboard together.

AMD Ryzen Threadripper 1900X 3.8 GHz 8 Core TR4 Boxed Processor YD190XA8AEWOF - Micro Center

Wrong thread?

thanks

Best graphics cards for audio are the lower end pro cards, since they draw less power and subsequently do not cause as much fan noise (which can be annoying) and they are also optimised for stability.

A friend of mine installed a WX 3100 Radeon Pro and his PC setup is very stable. He used to run a GT1060 on that same system. The GT1060 was of course much better for gaming since the WX3100 isn't optimised for gaming but rather stuff like Blender, 3D modeling, CAD, etc.

His audio performance improved after he replaced his GT1060 and installed the WX 3100.

For some reason Pro cards seem to fare better for audio/video production due to better stability.

But gaming cards can still be used. Basically the fewer and less power drawing components the more stable the temps are and the better the stability of the system overall.

An audio PC should ideally not be connected to the internet or use WiFi either...but that's a tall order especially for authorizing VSTs, apps etc.

These figures, even from the Nvidia card, are nothing to worry about.
Lower is better, but all of this is far from problematic.
You do want to let LatencyMon run for an hour at least though, to catch incidental spikes. Both AMD and Nvidia drivers like to burst to 600µs (default non tuned), which still only *may* be problematic when using very low buffers and when recording audio at the moment that spike appears.

Here is a 20 minute run, DPC is at 60 µs.

Thank you for the report!

For anyone interested, for Nvidia better to install only the driver
Install only the driver for Nvidia graphics cards - gHacks Tech News



For AMD GPU better do a custom GPU driver installation and install only the GPU driver and the HDMI Audio and nothing else.


After reboot go to "Device Manager" and disable the "AMD High Definition Audio Device".


To further tweak get Autoruns for Windows - Windows Sysinternals | Microsoft Docs
go to the TAB "Scheduled Tasks" and disable "AMD Updater".


Go to TAB "Services" and disable "AMD External Events Utility"



If ever want to update the GPU driver, use DDU and allow it to boot into safe mode to remove the installed driver.

Some AMD models are easier to clean like the Sapphire models with Quick Connect


Even more the XFX models with hard swap fans

@ Pictus

Unfortunately those proven "less is more" or trimming strategies doesn't really solve the mentioned Nvidia issue.. Maybe there can be also other variations of that or you'll get some small improvement (well the removal of unused components certainly doesn't do any harm), but IME if some setup is affected by those 300-400+ us spikes, removal of those components doesn't help you, because none of those is the culprit, driver itself spending excessive times in DPC queue at certain setups.
I've experienced this at 3 different configurations, really tested everything imaginable at all fronts, but only cards, which are safe bet are AMD Radeons and I also never experienced that with Quadro.. mentioned that couple of times before.

Weird thing is, it's also not possible to isolate it to particular Nvidia chips, for example I've found GTX 750 as affected by that at friends rig (it took 20s, until you've got near 500us readout from the driver).. On the other hand, the same Gigabyte card is installed at our main studio DAW with 3 display setup at different Supermicro board and there aren't such issues with that (sometimes peaking to 150us in longer measurement).
It's also independent on OS.. (Win 7 or 10 it doesn't matter).
Paradoxically only time, when DPC values were lower at those setups, were when I ran some foreground OpenGL application.. doesn't matter what, for example rotating 3D logo at Nvidia driver settings was enough to calm down DPC times under 100us. Left it running for 10 min. or so - nothing, closed the GL app.. back at spikes.

Michal

Maybe Nvidia got some gremlins...

No offense but, the only way you can or should make claims like this is by performing one of the DAW latency benchmark tests and post before and after results. Even then, it's easy to manipulate the results, thus providing questionable results for other users due to various other factors.

There is no debating the high execution numbers of Nvidia's processes, and I have also used both Nvidia and Radeon cards, and there zero difference of resultant plugin counts for either card. Or extremely marginal.

You can also get micro-improvements by changing out things like RAM, adjusting PCI card slots, BIOS tweaks. IME the only way to increase plugin counts if your computer is already functioning well is by increasing FSB/CPU speeds (when CPU power is required).

A big misunderstanding I see all the time is that people think DPC latency is the same as buffer latency or that it effects latency in any way, which in my 20+ years building my own systems it does not.

Realistically, IME, those numbers don't affect plugin counts if your setup is already in a 'green-light' or a 'good place'. So for others, I would not go and run out to buy an AMD card and expect better DAW performance.

Also as pointed out, the variety of actual parts used as opposed to the labeling of these cards warrants more detailed inspections than simply AMD or Nvidia.

I would guess that if you did see actual improvement in your DAW host, it is probably more because your PC was not previously in the 'green-light' area, or something else was wrong and you made multiple changes all at once and credited it to the graphics swap. That said, I have seen much stranger things and I'm happy if it's working out good now, for you, regardless.

It's great to see the results of your testing though, and I am running a Nvidia 1050ti right now, I would be interested in switching back to one of those cards in the future.

Those drivers really to appear to be pigs in comparison of execution counts.

Who knows Anyway it's bit annoying at least, because it's hard to predict, if some setup will or will not be affected by that and Nvidia never really fully addressed that or officially spoke about some remedy.

On the other hand, as mentioned by Leon (DAW PLUS).. the impact and severity of that really depends on particular workload and times of other drivers in the queue.
One aspect of that is, the lower is certainly better, because times can accumulate. When those spikes occurs and it gets combined with times of other drivers, there can be situation, when very short working ASIO buffer and short additional buffer (or offset) at audio interface driver level doesn't overcome that and you'll get dropouts.
On the other hand, when someone gets regular dropouts even at moderate buffer lengths (say >128s) and pushing his system utilization high (say hitting single core performance with demanding instruments or effect chains), then solving of that likely won't save him IME.

Michal

3. AMD RX 560D



Hi,

I´m thinking of buying the AMD RX 560 card. Just not sure I found the right card. Is this card the same as yours?

https://www.prisjakt.nu/produkt.php?p=4407895


If it isn´t maybe it doesnt matter, if they use the same drivers?


Right now I have the nvidia GTX 960 which gives the highest us in latencymon.




Thanks in advance!

I am using a Sapphire Pulse 560 (SKU 11267-19) with fewer stream processors and compute units than the regular one. Using a Crimson Edition WHQL driver as I am still on Win 8.1, there are newer Adrenaline driver versions for people using Win10.

Thanks for your quick answer.

Ok but the nvidia drivers are the culprit right? So switching to AMD drivers should do the trick?

What do you think about the card? (From last post)

I´m also thinking about this one:

https://www.prisjakt.nu/produkt.php?p=4407892

Which one do you think will work best?



Thanks again

Yes, you will get lower DPC latency with an AMD card. Regarding you question about which card to choose I would personally go for the dual cooler one because they tend to be quieter and better at cooling.

Thanks!

I´ll get it today. I can always return it where I will buy it if I´m not happy. Will be interesting to see the improvement.

Hopefully nvidia won´t fix their drivers right after I´ve change the card

I seriously doubt it.

Do you have any particular problem with nVidia, why are you changing your GPU?

No I don´t have a problem. I was just tempted when I did see your screenshots of the improvements after switching the GPU. Maybe I´m stupid to switch but here is the results with my GTX 960 card.

You think the AMD RX 560 card is worth it for me still, after seeing my results?

It depends. If you are working at low latencies and hearing crackles than it definitely is worth getting. But, if you use a higher ASIO latency buffer and do not notice any problems then it might not be worth it.

I noticed that now I can run more plugins until I overload the CPU, but my use case may be different than yours. And I also don't want to support nVidia anymore, which is another matter completely.

Your results are fine I guess, but if you are like me and want your system to perform as good as possible then go and get it.

I will post the results after getting the 560 GPU. In a few days.

Well, your results clearly show the nVidia driver having the highest DPC execution time, I hope getting a Radeon card will solve it for you. Have in mind RX 560 is around 30% slower than GTX 960 if you are playing games.

If you're a gamer and need high end graphics I can see some high end cards being an issue where it might dominate the computer bus but for most computer cards the video has little to no impact any more. Far less then it did in the early days where you have limited IRQ's.

The only thing you want to check and see is if your audio port whether it be USB, PCI, etc is on a different IRQ then the video or network card. Its best if all three are on different IRQ,s and even better if you audio device is on an actual IRQ and not one of the virtual IRQ's.

Getting them on their own IRQ's is the tough part. Sometimes in Bios you have the option to switch which IRQ the USB is on. Swapping cards can sometimes get them to jump IRQ's too. Some newer mother boards simply wind up being what they are and you're screwed if the video or network card are on the same IRQ as Audio.


You can set the video for best performance which frees up resources but to be honest I really haven't seen it add a dimes work of performance in newer operating systems, especially win 7 and above. The newer operating systems self adjust their performance based on your usage and even with things like services and resources optimized its not going to make that big a difference.

Win 10 has even removed many of the adjustments you used to have on older operating systems. I did an upgrade on my older dual core and certain plugins using open GL wont even produce graphics. The only thing I can do is upgrade the video card but for one plugin it's not that important.

Win 10 is very stable, far more stable then any other OS I've used. Haven't had a dropout is 6 months whereas win 7 I had one nearly every session using Sonar.
Typically dropouts occur when people aren't allowing enough buffering/latency with their interface settings. Computers cannot process audio without latency. It takes time to do that math. Most computers I've run that DPC Checker on shows an average of around 100us. Some interfaces report lower latency but whether they run reliably at those amounts is questionable. I typically set the latency high enough where I can run the maximum number of plugins without dropouts. From there I'm using direct zero latency monitoring when tracking so I could care less how great the latency actually is. It could be a full 2 seconds long and it wouldn't matter because I'm not listening to the processed sound when tracking. I don't use plugins when tracking either so it isn't even a problem.

If audio drops out and the computer is struggling to keep up, increase your buffers/audio latency till it goes away. Stop when it stops improving and mark it as your maximum setting.