PC Graphics

[sinshiva]

Freshly installed 4/4/2020, Win10pro 1909

i7 4790 (non-K) (with cryorig cooler, runs ice cold)

Gigabyte B85m

GSkill 16gb DDR3 @ 1600 (due to mobo lacking support for the 1866 profile)

Corsair 850TX PSU

Gigabyte 5600xt with new bios, 4/1 driver

Windows power settings on High Performance, Fastboot disabled.  Bare minimum applications running while playing Control.  Control randomly crashes to black screen of death after variable amounts of time.  Recently, it happened while opening a one of those in-game documents for some reason.

Another oddity;  the mmo game Forsaken World is missing a random 2D UI element, the avatar in the top left corner of the screen.  this game uses some old af C&C engine or something to that effect.  this immediately happened after replacing a gtx960 with the 5600xt and has persisted through a clean install of both windows and the game.

Other games that have crashed on me; BF4 (dx11, generally stable, though.  initially tried mantle which worked the first time, then would crash on boot.  later learned about it being 'dis-mantled'  so disabled it in the profile config), BF5 (both dx12 and dx11, but dx11 is infinitely more stable.)  RE3, which is hilarious, since this latest driver supposedly specifically added support for that game.  Chromium based Edge crashed the driver while trying to register for this forum lol.

Had this system for years blah blah gigabyte gtx960 never crashed once, _ever_, in the entirety of its lifespan blah blah ad nauseum.  I've beaten this system to death with prime95.  All RAM downclocking suggestions will be ignored with prejudice.

random bf4 errors and crashes below.  generally i don't even get to see errors when other games black-sod.  when chromium edge crashed the driver, the whole screen was just frozen.

I'm not here expecting to find answers, mostly just to add another system to what i am sure is a long list AMD needs to figure out driver compatibility issues.  I'll happily try new stuff, though.

[sinshiva]

Here's a a video of a rarer crash.  Instead of the usual black screen of death, the screen is frozen and i have hdd activity lights.  still, just as a dead.

5600xt driver soft crash - control - YouTube 

Look out world, new entrance into the most boring of YT channels

[sinshiva]

Well, apparently i need to reinstall drivers again.  Now the Adrenalin whatever software crashes on launch 100% of the time.

report.wer, fwiw;

4/2/2020 radeon software - crashes on launch - Pastebin.com 

really, really good stuff amd.

[mstfbsrn980]

Your GPU is OC version. Perform 100Mhz underclock with WattMan and try again with BF. If the problem exist, definitely use your consumer rights. I suggest you try the game with DX11, not Mantle API. I guess the Mantle API is outdated and it may not have been updated with current GPUs.

[sinshiva]

So, i 'downgraded' to the 'stable' BIOS provided by GIgabyte.  WIsh i could say it made a difference.  Here's another new softcrash.  Also in Control.  In this particular case, i was evade-boosting across an area when the screen froze and had 3 consequtive popups.

[mstfbsrn980]

It is likely that there is a problem with the motherboard. When I made an excessive OC on my system, I got a similar Control game window. Perform Prime95 and OCCT CPU tests and make sure the system is stable.

Edit: With OCCT, it may be necessary to make 10 minutes of AVX and without AVX tests and see CPU+mobo status.

[mstfbsrn980]

17.9.1 does not have Navi support. Very old.

[sinshiva]

so, i uninstalled AMD driver/software suite and manually installed the driver via device manager from the C:/AMD folder.  Only did it before going to bed lastnight and didn't have time to test, so i'll report back my findings later.  Ideally, i wont report back for +24 hours

[sinshiva]

pure driver only didn't help :\ just crashed again in Control

attached are a small pile of crash dump files and things .wer files from the last couple of days

[edit1] found a minidump to attach [/edit1]

[edit2/] here's some watchdog dump from a crash yesterday; https://easyupload.io/hz9eoh 

[mstfbsrn980]

-- Use DDU and Windows Safe Mode. Clean the old driver installation. And turn off the PC.
-- Open the PC case.
-- Remove the graphics card.
-- Clean the motherboard PCI-Ex slot with a clean dry old toothbrush.
-- Replace the graphics card.
-- If your processor has a heating problem and you have thermal paste, make thermal paste replacement for the processor.
-- Run the system.

-- Install the lastest display driver.

-- and run OCCT CPU test for 10 minutes for AVX and without AVX.

-- and run OCCT PSU test for 10 minutes.

You are using a 5-6 year-old system. The motherboard may not be able to distribute the old stable voltage. Your PSU may not be providing required 12V stabilization. There may be transmission problems with your PCI-Ex port. 

An older system is likely to cause trouble. So not look for flaws in the most recently purchased hardware. And... if you get a similar problem, use your consumer rights. 

[sinshiva]

ran all your recommended tests + let the power/burn test run for 20 minutes.  also, this is with the fa0 oc bios from gigabyte because the 'stable' lower clocked bios didn't make a difference.  every test seems to indicate all the hardware here is stable as a barge.  including the gpu.

[mstfbsrn980]

Perform 100Mhz or 200Mhz underclock with WattMan. Your system is healthy and this system should work stably without any problem.
https://www.amd.com/en/support/kb/release-notes/rn-rad-win-20-2-2 
In addition, this driver is WHQL certified. You can try this driver.

Good luck...

[mstfbsrn980]

If I were you, I would follow this ways...

-- Download 20.2.2 (WHQL) and download DDU.
-- Use DDU and Windows Safe Mode. Clean the old driver installation.

-- Perform driver installation with Windows Normal Mode without the internet connection, then reboot.

-- Run Radeon Software, and go WattMan, and reset WattMan settings, and underclock to GPU max. core frequency by 100Mhz (no need other changes), and apply the changes.

--  Follow that way... Computer Management > Task Scheduler > Task Scheduler Library > Right click AMD tasks and disable all. And restart system. So with reboot, Radeon Software will not work with this way. And underclock will still be valid. (With GPUZ, you can check underclock.)

If there was a crash despite all these attempts, I would definitely use my consumer rights. I would never use the new GPU that caused the problems. I also recommend not using Mantle API.

[sinshiva]

actually, i think you were right the first time.  so, i noticed the gpu utilization had a funky ripple in the graph while the cpu stayed pegged perfectly flat at 100%.  everytime the gpu utilization dipped for a second, the cpu core clocks would drop to 800mhz and quickly get back to 3.6ghz.  so, i went into UEFI and locked my cpu clocks to 3.6ghz and disabled turbo.  this mobo is limited in what it can do, so i couldn't lock all 4 cores higher.  i was then able to set cpu vcore to mobo's minimum voltage of 0.9v.  ran prime95 smallffts for 30 minutes to verify stability.  then, i reran OCCT power test for 20 minutes and the gpu utilization stayed essentially perfectly flat as long as i didn't touch the computer.  also, the gpu temps got _much_ hotter with the perfectly consistent cpu core frequency.

i feel like i may be on to something here, but i wont be able to test game stability until later.  could the drivers be... sensitive... to severe changes in core clock, or whatever causes the changes in core clock?  maybe nV knows something about this state change that they account for to provide stability.  i'm reinstalling the radeon software suite now so i can up the fanspeed curves, those temps were a wee bit high for comfort.

[mstfbsrn980]

The advantages of NV and Intel over AMD are that they support high speeds with low vcore values. Unfortunately, this is still not possible for AMD CPUs and GPUs for now (Maybe RDNA2 will be better). I did not fully understand the question you asked. But if you think core clock fluctuation causes errors, increasing the vcore value may be fix the solution with WattMan (or mobo BIOS for CPU). (Note: Increasing GPU or CPU core voltage may cause overheating!)

Obviously you don't need any BIOS or WattMan changes. Just try the recommended drivers (or resetted BIOS). And if you have the problem, you have to use your consumer rights. Because the error messages you have seem to be hardware and GPU sourced.

[sinshiva]

sorry, i was talking about cpu core clock, not gpu core.  i feel like the gpu driver crashes might be related to cpu core clock fluxuations.  like, a dramatic change in the speed of the data causes the gpu driver to fall down.  so, i'm trying to lock the i7-4790 to 3.6ghz as tightly as possible to eliminate a variable

[sinshiva]

One last test before bed; occt with locked clocks/undervolt and with gpu fans set to 80%.  the sawtooth pattern still came back, but not nearly as severe as the first test.  the cpu core clocks are still dropping to 800mhz when it dips like that.  i suspect it wont be an issue for gaming loads?  fingers crossed.

[sinshiva]

New crash;  just started up MONSTER HUNTER: WORLD for the first time.  I was cycling through faces at the character creation screen.  I'm going to try and re-create this crash.  My computer actually recovered enough to produce a memory dump and rebooted itself.  I'll reply to this post if i can re-create the crash.  minidump is attached.  memory dump linked below;

https://easyupload.io/ukuo4p 

[sinshiva]

wasn't able to re-create the first crash.  second crash happened in the rewards menu at the end of the first mission.  tried to alt-tab out for a moment and the driver locked up to black screen.  had to hard reset the computer that time.

[mstfbsrn980]

You cannot fix this error by underclocking. Try the WHQL driver. If the result does not change, use your consumer rights. Retire the GPU.

[sinshiva]

Well, i'm done.  refunding the card.  24 hours prime95 stable.  numerous hours of stable OCCT testing.  The hardware is solid.  The drivers are just pre-alpha trash.  Sorry AMD, i tried.  There wont be a next time.

[sinshiva · Accepted solution]

So.  I was wrong.  I ended up keeping the 5600xt after my head cooled off.  I actually went as far as initiating and then cancelling the RMA.  Did some further testing.  And finally began to suspect the PSU.  the PSU was 11 years old.  I replaced it with a new seasonic prime ultra platinum 550w.  With that new PSU, things immediately stabilized.  I did have to bump cpu voltage a bit for stable 3.6ghz under full load.  But, the really amazing bit of news is that i am now able to run the GPU at the maximum OC of 1820 and 1750 mem clocks, while undervolting to 0.93v!  i feel like i might even be able to go a bit lower, but i am more than happy with where it's at.  Definitely will be beating on it for the next week, but i suspect i wont need to tweak anything much

Long story short, old PSUs and lower quality PSUs simply do not cut it with this GPU.  Take a long hard look at your PSU if you are encountering consistent instability, no matter how many switches you flip and dials you turn.  Absolutely huge difference in stability with this card.

[xirombinha]

Glad that you’ve figured out how to set it to work properly. But.. are you sure that the old psu was leading it to crash? What is the model of the old psu? I think it might be working fine right now not only because you’ve trade the psu.. as you set some undervolting it could’ve played a huge role into the system stablizing.

Keep us updated!