PC Graphics

[MartinSd]

So not the first one to get this issue, there have been countless topics about this, and AMD has been a bit silent on this lineup. Would we be seeing any driver updates this month? 

To summarize the issue:

During gaming, all monitors go black, the GPU fans sometimes swing at full speed or remain at the same fan level, all lights are still on, the audio cuts off 4-5 seconds after that happens, then the lights turn off. I can't do anything except hard reset the PC

I've mostly noticed this happening under heavy load (GPU usage 80-100%) especially with a video playing in the background. This is not mandatory as the black screen crash could happen randomly within 5-10 minutes of gaming

Tried all possible solutions so far:

Manually updating drivers
Windows reinstall
CMOS reset
GPU reseat
Separate Power cable + reseat
XMP disable
Bios reset
AMD Drivers without Adrenalin
Windows Update

[mackbolan777]

Full specs?

[MartinSd]

Of course,

R7 3700x
HyperX 32GB DDR4 3200 (4x8)
B450 Tomahawk Max
Seasonic Focus Plus 80+ Gold 750w 
ASRock Phantom Gaming D RX 6800 OC

All on stock settings, no custom OCs or stuff


The PSU is fairly new as well, and I previously used an RX580

E.g event viewer crash log: https://i.imgur.com/NIywIm2.png
It's the generic system shutdown unexpectedly one unfortunately, not sure how helpful that would be

[mackbolan777]

You say "monitors", how many, and what size? Aside from that, you might want to explore the Adrenaline software and manually tweak the card some. Under "Tuning", choose "Auto", "Undervolt GPU", write that number down. Notice the stock GPU clock speeds of min/max. Switch to manual OC and enter the number from the undervolt that the software set. Make sure both the minimum/maximum clocks are set to minimum clock 100Mhz below the maximum one. Unlock the power limit and set that to max, it will only use that voltage if needed. Unlock the fan control and turn off "zero RPM", set a slightly more aggressive fan curve. We want to try to keep the card in the 70-80c range on the highest side of the operational temp if possible. Lower is better. 

I have a Gigabyte RX 6800 and my settings are: min clock 2400/max clock 2500, Voltage 1.00mv, VRAM 2150, Fan curve is custom so that at 50c they are at 50%, 60c/70%, and so on to 100% at 80c. Mine runs 70-75c full load but only one screen at 1080p/185Hz on DP. My settings are reference only and may not work for you. 3rd party card makers tend to toss a bit too much voltage to get things done when it's not required which causes heat, lots of it. It's a shame these "OC" cards don't come ready to use universally and we need to refine what they made by delving into settings that should be "ok". I looked at yours on AsRock's site and the difference between clocks is too wide in an effort to reduce temps vs. dropping some extra juice and getting much more bang for the buck.

I'd also check out what that "volmgr" error is in the log a few seconds before the error 41 which was caused by the shutdown. The error if caught will be a few minutes to seconds before that code 41.

[MartinSd]

Three monitors, all 24" two of them are at 60Hz, one of them is 144Hz

I did try with only one monitor and a Displayport/HDMI plugged in (The 144Hz)

Now what I haven't tried yet is a heavy underclock + undervolt

[MartinSd]

Also not sure but I think the PCI-e cables might be problematic for this GPU, not an expert on PSUs but here's what is being used at the moment (2x 6+2), as that is what my PSU came equipped with

The +2 is drawing power from the 6x perhaps?
The PSU connection part is also missing two pins (as seen in the picture)

https://i.imgur.com/gbMOtbE.jpg

[mackbolan777]

That's how mine is powered. A single/double 6+2 connector off a PCIe output on the PSU. Just make sure it's marked as a PCIe or VGA output on the PSU and not "CPU" or "Motherboard". In any case, if the output was wrong, the problem would happen right away. 

My settings are an overclock/undervolt not underclock by any means. The OC might stabilize the card, even using lower numbers than me as long as the min/max clocks are within 100Mhz of each other. Lowering the voltage some reduces overall temps under load as these as most AMD cards come running more juice than needed to cover differences in quality. That's why I go into the Radeon tuning and use the "automatic" overclocking area to find the minimum voltage that it calculates as a stable undervolt. Some cards may not be able to run at 1.00mv and need more power, others less all the way to .900mv. That's why setting the "power limit" to max is important. It allows for the card to add some extra voltage as needed and not full time. 

Moving the GPU clocks closer eliminate voltage spikes that happens as the card ramps up and down while gaming or anything really. Think of how the card going from 500Mhz to 2430Mhz as an example would cause the card to suddenly increase voltage, which can even cause the PSU to trip the protection circuit causing a reboot. These cards are powerful so I think in the future, the drivers will set those clocks closer after enough complain about it, lol. So setting the 2 clocks within 100 Mhz like 2330/2430 or something is less strain.

As far as refresh rate goes, the card should pick the lowest of all 3. If not, try limiting the 144 to the lower refresh rate. It's always best to use matching refresh rate monitors or the same brand/model in multiple setups since the GTG (response time) is different as well. 

[MartinSd]

Good suggestions, I applied the undervolt+overclock max values and brought up the min-frequency slider higher, closer to the max one

It still black screened unfortunately :/

[mackbolan777]

Using one monitor or the mismatched 3?

[MartinSd]

Tried both, using just one monitor, all 3 at the same time, all with the same refresh rate (60Hz)

Still no luck :/

[mackbolan777]

This may sound strange but try running 16GB RAM. Your board is a "daisy chain" topology vs. "T" topology and that means, 4 sticks might be a problem with a 16GB VRAM card. Just run slot 2 and 4, see what happens. (2nd slot from CPU and last slot). One monitor at first if it works.

[MartinSd]

Crashed as soon as I entered a game

2/4 slots occupied at dual channel, the other 16 were removed for this test


What I did find, prior to that, is that keeping the voltage the same, bumping the power limit and heavily decreasing (by 200-250MHz) the clock speeds actually made the GPU kind of stable, well at least it lasted for 10-15 minutes before I ramped the clock speeds to stock to check if it'll crash again. This did put the GPU to draw approx 140-150w of power, compared to 220 where it starts being unstable, from what I observed

This kinda points me to the direction that the PSU might not be enough, and It's a questionable model as well, because it seems like it had previous problems with Vega 56/64 cards 

[1] Units experiencing tripping issues with high transient power draw GPUs like AMD Vega and high-end nVidia Ampere. Includes Seasonic Focus based units made prior to Jan 2018, source.
(quote source: https://linustechtips.com/topic/1116640-psucultists-psu-tier-list/)


Although I bought the PSU mid-2019

[Coffeecat]

https://community.amd.com/t5/graphics/rx-580-nitro-black-screen-game-crash-pc-restart/td-p/422029

[mackbolan777]

@MartinSd 

That's very possible, which was why I had you stabilize the voltage so the card doesn't spike as much. If you return the PSU, I'd recommend the Corsair RM 750X, EVGA G3 750 (has some transient 3.3v issues but competes), or other RMx Corsair PSU. Always had luck with the Corsair RM. I'm switching to an RMx 750 because this RM 1000 is from 2013, I replaced the fan once, used to smoke like a fiend and that really messes PSU's up. Quit smoking 6 years ago but the unit has coil hum on boot now as of this past week, so to be safe I'm replacing it. 10 year warranty is nice too. Seasonic makes a good unit but Tom's Hardware has a review that shows some "cons" that turned me off. 

I was going to say PSU from the start but you said all was fairly new. So under 10 years on a Gold PSU, a little unusual for a failure.

[MartinSd]

I thought so as well, I'm just speculating the PSU at this moment

What seems the odd case here  is that the Seasonic PSU-s are overbuilt with some protection in mind that doesn't allow huge wattage spikes, so the PSU actually does what it should, it's just too aggressive and triggers a failsafe

If the new RX lineup is built the same way as the Vega 56/64 cards, then I'd totally understand this (more info: https://linustechtips.com/topic/997831-seasonic-issues-warning-for-potential-videocard-compatibility...

I'll see if I can try with a different PSU of similar wattage and, update here again (soon hopefully)

[MartinSd]

Mini-update here

I got the card to a stable underclock (heavy but it works for now), at 1900Mhz with a slight undervolt as well, no crashes whatsoever for 2 days now. The card is hovering at about 150-180w power consumption when in game to keep transient spikes lower

The new PSU should be coming soon, and once I get to install it on the new system I'll test it again

[MrSleezeBag1964]

This sounds almost identical to what is happening with my ASrock Phantom 6800! My GPU fans will max out on some crashes too.  Very odd. I'm getting ready to RMA it, but I don't' know if ASrock's RMA team sounds too reliable... no telling how long it will take.

I haven't spent much time on this board, but I did see an issue where a guy fixed black screen issues by setting the pixel bit depth to 6 bits per pixel.  He was using an RX 480... BUT, I mention it because I noticed Radeon settings has my monitor at 10 Bits per pixel, however it is NOT 10 bit.  8 bit only.  I changed to 8 bit and will do some testing to see if it helps.

What games are you seeing crashes in?  For me, Destiny 2 wouldn't load to game play.  Crashes almost every time. 

[MartinSd]

Any game basically that could put some stress on the gpu (200+ watts)


I got the new  (Evga G3 750w) PSU and sad to say that it didn't resolve this issue. I was highly optimistic my older PSU was just faulty, but It seems like it was running just fine

[MrSleezeBag1964]

Ya, I did same thing. Swapped PSU through an RMA with Seasonic.  There's another post yesterday I saw here where a 6800XT user took his GPU to friends house and tested with different monitors and did not get black screen'd.  He concluded it was some type of driver issue with the monitors. 

I just bought another 6800 so we'll see what happens.   

[MartinSd]

Also tried swapping monitors and in fact cables too today, that ain't it

[MartinSd]

I did some additional observations:

- Installed the most recent driver that was released today just in case, did not work
- Benchmarks (Heaven) and Stress testing (Furmark/OCCT) work just well, stock settings, I'm seeing full gpu usage: no black screen crashes whatsoever

- Any actual game  - instantly crashes the moment I load in, and checking at sensor logs it does not even consume as much resources as OCCT/Heaven

[MrSleezeBag1964]

This is almost identical to what I've seen.  I've hammered it with OCCT.  Can't induce a crash.  Furmark - same thing.  Runs fine.  However, in both OCCT and Furmark z I noticed that GPU utilization is near 100%, but clock speeds are very low. 

However, if I load Destiny 2?  Crash.  Apex Legends works for a little while, then crashes.  Forza Horizon will crash too.

Doom Eternal is perfect.  Can't make it crash.

I forget, have you tried to disable Freesync?

[MrSleezeBag1964]

Wow, so I got my threads mixed up.  We have the identical Asrock Phantom GPU.  My Gigabyte RX 6800 should be installed soon.  I'm going to keep identical settings and see what happens.  

I talked with ASRock Support - you can RMA for 3 years.  BUT they have no Idea when they'll get stock in.  Communications with Asrock have been less than stellar.  You can call them up and they'll talk with you though.

You can also exchange an open GPU with Newegg - but if Item is out of stock, they refund you.

It seems that Gigabyte models have NOT been affected by tariffs.  I paid $680 for my Asrock in 2020.  It is now $840.  The Gigabyte Windforce 6800 is $680 on Newegg.  

[MartinSd]

 I did try with and without Freesync, actually with all extra options disabled 

https://i.imgur.com/ll6PUr6.png
https://i.imgur.com/Fh7Faak.png

RMA will definitely be the last resort, but I'm still trying

Lemme know how the other GPU works out, it could really be just a faulty line of GPUs

[mackbolan777]

@MrSleezeBag1964 , @MartinSd 

Personally, AsRock just is garbage and I have the unfortunate displeasure of owning one of their motherboards that was the bane of my existence for the first 4 months of ownership. I still cross my fingers each day that passes that this board still works. GPU's, I would never by an AsRock. They just seem to slap stuff together, add stupid features, try to make it look cool but in the end it's junk. My opinion, others rave about them. I RAGE about them.

Tariffs are in play, that card is $569 MSRP back in November 2020. Still not too bad, I paid $877 for mine and the seller paid $730 with the sales tax in December 2020. Only reason I grabbed it vs. waiting is the stock is not projected to be any better by April nor the prices, since that stock is going to sell out as well. The problem is flash memory and DDR6 is scarce at the moment so no manufacturer for the Red or Green team can keep up with demand. Look at SSD prices, spiking in a similar way. 

[MartinSd]

tbh that's what I did as well, reading all the news that this stock problem could last for 4-5 months, I just opted for what's available and not complete weird, AsRock has been in the market for quite some time so I said " screw it, that's the purchase" 

Would've went with a Sapphire for sure; Not putting the blame though, I may have gotten the short end of the stick, if it ends up being a hardware and not a driver problem after all


Anyways, I'm still experimenting with clock stability, but I will be RMAing the card next wednesday

[MrSleezeBag1964]

I'm certainly regretting buying this model.  I figured something was suspicious when NOBODY had reviewed this model on YouTube... but, when I was able to check out, I didn't hesitate!  I almost got a reference 6800XT from AMD directly too.  Would have just sold the unopened Gigabyte on eBay.  Keep us posted on how the RMA goes, I'm going to opt for an exchange from Newegg.  

I'm certainly no electrical engineer... but my guess is that there is a bad component in these cards having to do with voltage regulation, or power delivery.  Igor's Lab recently published an article about how TONS of different components are in short supply globally - the boring chips on a GPU that nobody really cares about, but play critical roles. 

In fact - I was using a multimeter on this Asrock which made me think the PSU was bad.  Check this out:

https://imgur.com/a/5KU4GQH

12.5 Volts is really the most you should ever be able to measure on a PCIe 8-pin.  

[xlox]

there should be no difference between 3d benchmark and 3d game , if benchmark work , no reason (in my view) that there is a hardware problem ... so rma wouldn't change anything (in my view)

[MartinSd]

I'll be getting the GPU to a friend's rig to test it out today, I'll update with "results" and if this whole fiasco is just a problem with my PC

[xlox]

but if no hardware problem and testing same games , not sure you get more info ... have to test other demanding games maybe or maybe try to find something on the game's forums (which you certainly already did, but in case ..)

[MartinSd]

There's always some possibility that some other hardware is just conflicting with the GPU, hence why I'm gonna test it on a completely different system ^-^

[xlox]

don't forget DDU ! (...one thing I noticed last time i was on DDU webpage : it apparently only support 2004 version of windows actually...)  :)

[MartinSd]

I did try with both DDU and AMD's clean up utility :P

[MartinSd]

I think I'm  on to something here

I actually did a bunch of testing with a friend of mine who has quite a similar PC to mine, we managed to crash it a bunch of times on his rig too, but we got some pretty crucial info on how it can be reproduced (given that you have a similar malfunctioning card?)

Specs:

• B450 Tomahawk Max
• 32GB (4x8) HyperX Predator 3200Mhz (XMP profile)
• R7 3700x at stock
• ASRock Phantom Gaming D RX 6800
• EVGA G3 750w
• 1x M2 + 2x Sata SSDs (Windows installed on Sata SSD)

My friend's rig:

• B450 Gaming Plus
• 32GB (2x16) HyperX Predator 3200Mhz (XMP profile)
• R7 3700x at stock
• ASRock Phantom Gaming D RX 6800 - (my card)
• Corsair CX 750w
• 1x M2 + 2x Sata SSDs (Windows installed on M2 SSD)

What we actually found was:

Any modern game would cause this crash - "Apex Legends", "Black Desert Online", "Rust" for example

The Card will always crash at stock speeds provided by the Tuning view: https://i.imgur.com/rYpP9Fu.png, but what's interesting is that the Max Frequency advertised for the Phantom Gaming D should be 2190 MHz

The card did not crash on one pre-set condition - Underclocking to 2105Mhz or lower and putting the Power Tuning to +10% (to 250w) will work (sort of, we had a solid 30minutes of testing without problems where it usually crashes within 1-2minutes)

Voltage stays the same - 1025mV, then during in-game (Black Desert), we cranked the Frequency to 2300MHz (And we got the card to work at 2300Mhz!) without a single crash. The moment we restarted the PC, put the card on stock speeds or close to stock, the card crashed again

Then we restarted the PC again and put the game on Auto settings, launched Apex Legends, where it does not utilize so much power (~50 Gpu usage, 1400-1500MHz frequency, ~140w power consumption from the readings), and the system crashed again

Conclusion:

The system will most definitely crash on Auto settings (https://i.imgur.com/JTIff1D.png)

The system will crash if Manual has no underclock set up - https://i.imgur.com/4yfsQ0s.png

The system will crash if "Power Tuning" is not set to +10% (to 250w max) but is above 2000Mhz

The system will crash if the GPU is not underclocked first

The system will not crash if the card is just underclocked to 1900MHz and with -8% power limit)

This lead to two possible cases: Driver failure (I hope!) or Hardware failure

Offtopic:

We plugged his 5700XT and he tried overclocking a little bit, which rebooted his drivers, and got me thinking: What if my card can't initiate this Driver reboot sequence therefore it crashes the entire system and blacks out the screen until I press the power button to restart the PC?

[mackbolan777]

All I see is the common factors of the AsRock junk card, my opinion, and the fact that those games all use CPU intensive game engines. Apex and Rust both use engines written in C++. BDO is it's own thing but also very CPU intensive, but consoles seem to run it ok-ish.

Try running that RX 5700 XT with the 20.8.3 driver and see if that crashes the PC or the FPS is just poor. Use the AMD ctrl+shift+o to display the GPU/CPU usage, instead of the metrics overlay so you can see that the CPU is doing most of the work. That's provided it doesn't crash. 

Then you'll need to see if the Gigabyte card handles the game(s). If it does as I suspect it will, albeit not the best FPS due to game engines and possibly some future driver improvements, return that AsRock for RMA and sell it or get your money back.

[MartinSd]

Black Desert definitely uses the GPU a lot more, that's my base 'testing' ground tbh when it comes to trying to force a crash

[mackbolan777]

Missing the point. That game should be CPU intensive, causing a crash because the GPU isn't handling it well due to the coding used. As in FarCry5 or New Dawn, it uses more CPU than GPU, hence more VRAM and less than optimal frame rates with my former RX 5600 XT. This RX 6800 does better and can do Ultra with the fullest of textures but I still need to set textures to "2" to get over 100 FPS. The game's both use roughly 45-60% CPU and 30-40% GPU and 11GB VRAM over time. They start out at 5GB VRAM and wind up at ~11GB after playing for over an hour. Just storing textures, waiting for the CPU to decode them for the GPU to finally do the rest.

[Coffeecat]

guys its a driver issue read other peoples post before posting

[xlox]

@Coffeecatstop trolling thanks, you don't seem to read any thread , beginning by this one

[MartinSd]

Furthermore, I observed a regular pattern (they don't appear to be so random as I thought previously?)

• When in Rust: The GPU will hard crash/black screen very specifically if I alt-tab/switch to the 2nd monitor and play a YouTube video (both HW Acceleration enabled & disabled)
• When in Black Desert: It's random, it could happen in 5 minutes, it could happen in 30 minutes, the lower the clock, the more stable it is.. kind of
• When in Apex Legends: Crashes at random sometimes, with random clocks (not fully utilized)
• New: When in Battlefield V: I didn't manage to crash the computer/GPU. I tried with both DX11 and DX12, maxed out, with a 4K video playing on the second monitor even, runs fine (Single player test, I can do a Multiplayer test if needed)
  
  
  It'd be really great if an AMD representative tunes up if they need more information