Discussion in 'Videocards - AMD Radeon Drivers Section' started by mtrai, Dec 17, 2019.
set your monitor not to sleep ? PC I should say
you know summut I was waiting for someone to say that same here
Closing Radeon Settings: Desktop Overlay and Radeon Settings: Host service along with the Radeon Software: Host application should fix it. Works for me.
This driver seems to cap my Vram at 500MHz, not matter what i try. On the pro side, my card runs really cool now. Gonna reinstall it with DDU.
Switch to 120hz so lets hope blackscreen will occur again (btw my monitor is connected via Displayport the Radeon Settings App shows: "2.7Gbps x4" so it should be fine (it worked before 19.12.2 ..)
otherwise we know:
It's a 144hz bug OR
No luck yet as it was really rare before - and it may still occur at some time
Do you have your card set to an overclock or auto overclock when it happens?
Hm nope , i undervolted the card with the same specs as before, but after it happened first i left it on stock settings and it happened again
Sometimes the window decorations of firefox get black and reappearing when hovering with the mouse cursor over the area
(it's like black rectangles, but doesn't affect the website - just the title bar / tab list / menu button ..)
i did a clean reinstall of windows 10 (1909) and installed the 19.12.3 driver and got the same problems as before - the system is clean - nothing installed except windows, gpu driver, steam and discord and firefox
Memory problem ? Just a thought
I'll try 19.11.3 next as i didn't had any problems before upgrading to 19.12.2 / 19.12.3
what's really annoying is that there is no way to 'provoke' the error - so always wait time / using the pc to see if it happens x.x
Dont suppose u have another pc to try gpu in ? (advantages of having a pc shop)
Haha nah i just got my pc
But tbh i didn't had any problems until i installed 19.12.2 + the problems were just on desktop use like browsing or video playback / chatting on steam or discord etc
games seem to run fine (or i haven't played long enough yet to get the error there too)
Also another thought (shot in dark) maybe the driver itself invoked an issue that was there before have you tried reseating ram works wonders somtimes or if you have more than one strip try one at a time
Same Issue here with Google Chrome, 19.12.2 did not have this Problem for me.
Edit: Blackscreen when i open a Task on chrome, if i start steam or other clients.
GPU OC is back to standard clock speed, after Blackscreen i can load OC profile and everything is fine.
I've installed 19.11.3 and will see if that changes something
but i'll be out till tonight now - so i'll be able to test after gym tonight ^^
I've run HCI Memtest a few days ago without any problems but it might be a good idea to check all connections + rattle the memory sticks
It's really rare for me, so maybe the bug was there for you in 19.12.2 too but you haven't experienced it
But it's really good to know that others have the same issue
Maybe i did not noticed.
But the other problems like Blackscreen and reset did happen only on 19.12.3.
Hardware acceleration isn't entirely fixed so I'd experiment with that first sadly this also applies to other software so it's not just browsers but CEF/Chromium in general and possibly also non-Chromium such as Firefox.
Easy enough to test if you have a good way to trigger the problem which should then confirm it.
Alternatively it might just be the overlay finding it's way into stuff and messing up things that way, you'd have to use the task manager and shut down the two Overlay related Radeon services and Radeon Settings itself.
I renamed them already but the first one is the desktop overlay and the second one I think is just Radeon Settings Service or something, start up Radeon Settings and the two helpers should no longer pop up and then testing should also quickly confirm if it worked or not.
There's...quite a few things in Adrenaline 2020 including the first fixes for it in 19.12.3 here that applies to Navi and earlier which just needs fixes or additional work so if the added features aren't much of a deal and the performance impact is minimal a cleanup and revert to 19.12.1 until it's resolved isn't a bad idea instead of either cutting functionality for workarounds or whatever is needed until a full fix is provided in a later driver update.
EDIT: Problematic behavior can also happen with full-screen media playback with examples from what I've read so far going to Media Player Classic and also the Mad VR software for playback and enhancements just causing a hang or other issues.
There's some game specific stability problems as well and the Overdrive situation and Wattman is about as bad as it's ever been with testing showing that fan settings frequently ignore set values or are fluctuating heavily and voltage and clock states can be a mess often spiking at random or getting ignored so that's going to seriously impact existing "stable" saved configurations which for all could still be stable but the driver will no longer be thus well it crashes or shuts down due to hitting too high or too low values.
(Auto mode hitting way too high for a overclock setting should be fixed at least but I don't know if this still applies to auto undervolting but it probably does as it's possible to get different values from it instead of the general conservative -50 it usually sets.)
EDIT: Or it can also work reasonably fine which yeah that's kinda how it goes, it does need further fixes but it's not 100% everything's all broken on every system so well let's see what the January drivers will be doing.
So much headache. Simply underclock and fan minimum speeds are enough for efficiency. Touching gpu is not so friendly like cpu.
Yeah but even those adjustments are no longer reliable now so you have to take that into account when using Wattman or you might quickly find that the values are not very functional.
One user even set a fully flat 50% fan "curve" to test it which even if it dips and drops since it's more like immediate stepping points instead of a gradual curvature the actual reported fan graph and RPM was varying from low to high which shouldn't happen at all though some variance can occur of course but any more makes it very hard to set custom values without a lot of trial and error and testing to account for errors and what's actually used, ignored or over or under shoots the targeted settings.
Looks like MSI Aferburner can force more specific settings at least, possibly other tools for custom power play registry settings and such but it's something to keep in mind and also test if using Wattman for anything to make sure there's no issues and if there is some variance also account for that with a second profile until it's fixed taking these into account for the new values to be used in the meantime.
EDIT: Though I am very curious how it works and why but it looks like Vega, Polaris and Navi are affected and various third party models of each and at least some of the stock ones for Navi as I believe one of the cards was a 50th AE model so that's the standard blower design and PCB and hardware with no changes.
(How it breaks? Well it might behave or it might not, worth taking into account at least.)
EDIT: Need to find more info on the power fluctuation and transient highs and spikes too, so far it looks to be around (Or more.) than with Vega thus for a fraction of a second instead of pulling ~300W it might be hitting 600W so no wonder even with a lower power draw those dual cables just will cause the card to crash since that's above what these can carry.
Also why certain PSU's even if well within the specs for what the card will draw with plenty to spare for the rest of the system won't be stable as they can't handle that sort of spiking without triggering a surge protection or shutdown to avoid damage.
Need more info still on how that's going though but it's a thing.
EDIT: Which also leads into that PCI Express power save setting, wonder if there's a chance of the power draw here being a issue with some of the black screen problems especially the non-recoverable ones.
Might explain though it's nothing more than a guess if there's a chance that's a big thing for any motherboard that's not a quality X470 or the newest PCI-E 4.0 certified and compatible (Not merely "ready" like the issues with some X470's) and a higher amount and better quality traces and lanes for this.
Could be, could be nothing though.
(PCI-Express has to be both forward and backwards compatible though even if there are advantages beyond merely the speed for the newer revisions too.)