PC Drivers & Software

[tazling]

I recently upgraded my driver (hoping to fix a very intermittent crash in a SteamVR game).

To my dismay, upgrading the driver appears to have made matters worse:  my win10 machine now routinely fails to wake from sleep, and my game is crashing every 10-12 minutes instead of "every now and then randomly".  

Is the most recent (Feb 2021 I believe) driver known to be problematic, and if so, is there a stable version I can roll back to safely?

I'm pretty new to Windows, and not expert at GPU hacking, so I'm trying to be very careful.  I have heard that Trixx can cause problems, and since I don't need any of its features I'm happy to uninstall it if that's the collective wisdom.  I don't overclock.  It's a really vanilla installation.  It used to be rock solid (other than the very intermittent SteamVR game crash) and now it's seriously impaired.

Hints, clues, etc. very welcome.

[RPX100]

Hey there.

The recommended driver is still 20.11.2, when you look on the driver page for RX 5700 XT.
Some community members still point towards 20.8.3 as most stable.
I personally can recommend to use whatever driver works for your card, as long as games can run on it.

There could be different reasons for your current crashes:
- installing driver updates without uninstalling the previous driver
- uninstalling the driver while Windows Update is enabled
- driver update may have overclocked your card (AMD default!)
-> AMD is overclocking your GPU via Radeon Software, when left on "automatic" performance tuning.
- maybe your GPU temps are too high? (check if they stay below 90°C hot spot)

How to fix your crashes:
1) start by using Display Driver Uninstaller (DDU) from guru3d.com
- make sure to use it in Windows Safe Mode and as Administrator, while your LAN cable is unplugged!
- it is important to do all this with no active internet connection (or Windows Updates disabled)!
- DDU will remove all current driver files and (known) registry entries for you

2) use CCleaner or WiseRegistryCleaner to clean leftover registry entries
- restart your system

3) run the Windows Command Prompt (CMD) as Admin
- enter: sfc /scannow
- press enter and let it finish (can take a minute or 2)
- restart your system

4) now install 20.11.2 as Administrator
- open up Radeon Software when install is done
- check if AMD applied any overclocking (which sadly is default) in your performance tuning area
- if your Software is set to 2100 MHz max (P3 power state) and 1.200 V, then you need to change that.
- you will have to enable manual control in order to see the actual values
- RX 5700 XT is more stable at 1900 MHz and depending on your make and model 1.05 - 1.1V

5) do not forget to plug in your LAN cable again ^^

Good luck.

[tazling]

Thanks for one of the most helpful and complete responses I've ever received on any forum!  bookmarked!  I'll try to verify the basic checkpoints you've suggested:

I am running 20.11.2.  My copy of Radeon SW seems to be also 20.11.2 (I don't find an independent version number for the app, so far).  When I get into the app...

In Graphics Tab, I choose the profile "Gaming."

In Performance/Tuning tab, I choose Automatic Control and Default settings.  (Not Overclock, or Undervolt)  And the numbers I see on this page are:

Memory clock:  1744MHz

GPU Clock: 6 MHz

Voltage:  725mV

If I select Manual (Tuning) then I get some Enable switches, all turned off.  I am not sure what it means if I turn them on, so better not...  Guess I need to find the manual on the "Radeon Software" app and learn what the knobs and dials do!

I seem to have solved the "can't wake up" problem by turning off ULPS in Win10, so now the remaining issue is the black screen crash in-game.  This could be an issue with the game config, as it's fairly heavily modded for graphics performance.

For now I'm going to leave the Radeon Software untouched and work on the game side of things, also plan to run the Heaven benchmark to thrash the GPU a bit and see how it performs.  If Heaven crashes it then I must have a config or driver problem.  If Heaven runs just fine but my game continues to crash, then I think I'd better start removing mods.

BTW I see that the Radeon software appears to support per-game profiles for tuning?  Is it possible that launching my game is imposing some pre-defined set of "optimisations" for that game which could override my preferences in the Tuning pane?

Sorry to be a bit clueless.  GPU hacking is not my thing, I just plug the card in and use it -- never get clever   So I'm on new and confusing ground here.  Many thanks again for a very clear recipe for a "reset, start over" install.  It's nice to know that there is a way to start clean, if I can't fix this any other way.

[RPX100 · Accepted solution]

@tazling hello again and cheers - I see you had a first look at some of the settings.
Thanks for pointing out a few things you have selected and/or not touched at all.

No worries about being "clueless", this whole problematic is different for each graphics card generation.
We all need to figure out this stuff again, whenever a new card is released. Things change all the time.
With AMD the trend seems to be: "it may work, or you have to make it work yourself (or with help)" ^^

---

Some info for you regarding Radeon Software, its features and the way it handles overclocking:
- Radeon Software itself is enabling the enduser (= yourself) to change graphic card behavior 
- the software enables you to change certain settings within the driver to fine tune your card
- these changes are mostly limited to "safe working conditions": so you can't damage your card with it
- so you can feel free to experiment with the settings and see what they do, without the need to worry.

- AMD is trying to compete with nVidia: price for performance is the key here
-> Radeon Software set to "automatic" will actually auto-enable overclock to 2100 MHz at 1.20 Volts
-> there is however no guarantee that your card can handle that (see: "silicon lottery" for more info)
-> if you leave everything on "automatic", then your card will be overclocked by default!
-> your card is will always trying to reach that 2100 MHz and draw as much power as needed
-> it will actually ran smoother on factory settings (around 1900-1925MHz) with less power draw and less noise
-> this is why I recommend to actually enable manual performance tuning
- the cooling solutions on these air cooled cards may struggle if you actually draw 200W of power
-> there are default Fan-curves stored on your graphics card BIOS (most of the time not 100% RPM)
-> increasing fan speed (RPM) will result in better cooling (and maybe better stability), but also more noise

Radeon Software profiles
- "gaming" / "e-sports" / "energy-saver" / "standard"
-> these profiles all relate to some performance and/or graphic settings
-> Radeon Boost, Anti Lag, Image Sharpening, Enhanced Sync, etc.
- If you do not know what they do and if you do not need them, I suggest to leave them turned off
-> recommended to stay on "standard" profile, as long as you do not need any of the features
-> I personally ran into black screen crashes with "gaming" profile enabled on my own RX 5700 XT.

Radeon Software Performance Game Profiles:
- The software enables you to create custom performance tuning game profiles
- to answer your question: these are only active if you create them manually (not active by default).
- these profiles can be used to select custom fan curves for each game
- you can also change max boost clock of your card to run certain games with less power draw

Benchmarks compared to playing games:
- benchmarks are "synthetic" and stress your GPU in different ways than a real game would.
- they are constructed/programmed to be the same on every machine (to enable comparable results)
- your graphics card might be able to complete benchmarks just fine, but may crash in games.
-> this does not (have to) mean that the crash is related to the game settings

--- 

Now to actually try to increase your cards stability, I suggest this:

1) switch your Radeon Software profile back to "standard"
-> this will disable all Software features like Radeon Anti Lag and Radeon Boost

2) enable manual performance tuning and also enable all those switches for "fine tuning"
-> You will see a graph for voltage and frequency relation
-> and beneath the graph, there are 3 pairs of input fields for power state 1, 2 and 3
- power state 1 is "idle" or low usage (2D low effort usage stuff)
- power state 2 is the "in between" work load (3D, with some energy saving features)
- power state 3 is the "max boost clock" under max load (100% usage)
-> refer to this list and find your make and model to see what value you should enter here.
-> on the top right of the provided list, you can see "boost clock: 1925 MHz"
-> this is the value that you should enter in the "P3" (power state 3) frequency input field
-> if you scroll down on that list, you find all other RX 5700 XT make and models for reference

3) now that you have the frequency reduced from 2100 MHz down to around 1900 MHz:
-> reduce the voltage (displayed in mV) from 1200 to 1100 mV (for the P3 state)
-> make sure to leave P1 and P2 voltage alone (P1 should not be lower than 750 mV)
-> this again is "silicon lottery" and your card may be able to handle 1000 mV, 1050 mV just fine
-> less voltage may at some point cause instability in itself, but it will also reduce generated heat.
- you can manually test the stable values for your card by reducing the voltage step by step (10 mV)
-> this requires to put the card under load (benchmarks/stresstests can work, but games are your use-case)
-> at some point the card will "hang" (blackscreen) and at this point you increase by 10 mV again and leave it.

4) Then navigate to the power limit (and enable control) and set that to max. (slider all the way to the right)
- This allows the GPU to take more juice as needed, if it needs more than the undervolt.
- Think of it as a "reserve".
- The GPU doesn't tap that power limiter until it's used up the max voltage set under "fine tuning"
- and then only takes what it needs, not all of it.

5) last step should be to optimize your fan curve
- increase it as far as it needs to: manual testing required or go for 100% RPM (if you don't mind more noise)
- adjust fan curve until your card stays below 90°C junction temp (under gaming loads)
- your cards core, VRM and memory temps should stay around 60-70°C

strongly recommend to use GPU-Z to monitor your graphic cards temps and frequency.
-> it has some nice "sensor" feature that will also show the values "over time"

These 5000 cards come out running far more voltage than needed.
Undervolting them will result in more stable clocks and better temps.

---

These steps might not be needed for all RX 5700 XT cards out there,
but I have encountered the 2100 MHz "auto-overclock" myself and can confirm,
that it is causing crashes on my own card as well.
I only get stable performance, when I use my card with the factory settings (1905 MHz in my case).

[colesdav]

See this: https://community.amd.com/t5/drivers-software/amd-driver-fix-and-install-using-ddu/m-p/426238#M13454...

I would install Adrenalin 2020 20.3.1. 

FRTC is back and that might help control your GPU  Frame Rate better for VR games. 

[taskforce131]

try this it will help 

part 1 

https://youtu.be/nHd0065duKE

part 2 

https://youtu.be/P6k2Q8kG98o

[RPX100]

@tazling any update on your issues?
Were you able to fix the problems?

[dontblink]

I've been having this problem for that month on my Sapphire Radeon 5700 XT Nitro SE. Tried everything and multiple drivers. The only thing that has stopped the black screens so far was effectively decreasing the max frequency to -5% I highly suggest trying at -20% and working your way up. I'll never buy AMD again. 

[tazling]

I am so sorry I dropped that ball.  Yes I did fix the problem, but now it is months later and can I remember what specific thing I did that fixed it?  of course not.  I did bookmark your answers though, in case I had subsequent problems!

[RPX100]

@tazling welcome back!

I am glad to hear that you seem to have a stable GPU at the moment. Congratz!
If you think that my answer solved your problem, then go ahead an mark it as solution.
This way the topic is marked as "solved" and other people can see what you tried.