GPU Driver stability

I dug into the report a bit between meetings. Overall the methodology is sound, but I find the OS and driver selection odd. Win10 1803 was a bug-riddled mess, and they didn't use NVIDIA's official 1803 drivers for Quadro (R396), which were out at the time.

I think given the terrible state of Windows it's a feather in AMD's cap to only crash out 7% of the time. But I don't know if it's very comparable to NV given the driver choice. I'm also curious how the numbers would look under 1709; I bet everyone's stability would be much higher.

 

Most of the crashes happened on Quadros and FirePros, the mismatched driver is among the causes for that. Geforces vs Radeons have comparable results.

 

Or this 3rd party is simply incompetent. Either way seems just a PR piece for investors. Yawn.

 

Well, that seems to back up my personal long term experience. Granted the last AMD card I've used is the R9 290, but damn that card and drivers were rock solid over extended use.

I really wish I could say the same for the 1070. I like the card. I absolutely hate NV's drivers.

Regards,
SB

 

AMD Claims To Have Worlds Most Stable Driver – But Supplied All Graphics Cards Tested

Typical AMD marketing ...
Read more at http://www.legitreviews.com/amd-cla...phics-cards-tested_206705#2QvtKk5WEIHWogCP.99

 

February 11, 2020

 

Meh, AMD still fucked up their Ryzen Master driver or whatever was that.

1. It refuse to launch if you use Hyper V
2. If you manually hex edit the exe and disabled the checking, it runs just fine.
3. if you use old ryzen master (ver 1.2xxxx IIRC) it runs just fine with hyper V
4. AMD has acknowledged that issue since 2017/2018.

still not as hilarious as NVIDIA's fucked-up HDTV/monitor detection and failed to not use chroma subsampling tho. That bug is more than 10 years old and has been acknowledged.

 

The problems with Navi cards are much more pervasive this time, here are the list of issues AMD acknowledged so far in their latest issue report:

-A loss of display with working audio may be experienced on a limited number of displays when performing a mode change on Radeon RX 5000 series graphics products.

-Some Radeon RX 5700 series graphics users may intermittently experience a black screen while gaming or on desktop.

-Some games may exhibit stutter or appear to be downclocking on Radeon RX 5000 series graphics products.

-A black screen may occur when performing a mode change with a limited number of displays on Radeon RX 5700 series graphics products.

https://videocardz.com/driver/amd-radeon-adrenalin-2020-20-2-1

 

Do you still have the problem if you use this 1.5 year old driver Geforce driver 399.24?
https://www.nvidia.com/en-us/geforc...44-chroma-subsampling-with-recent-4xx-driver/

Edit: TBH I have never experienced these issues, currently using a Vizio HDTV and Dell monitor

 

According to Hardware Unboxed, retailers are reporting that the return rate of 5700 cards are 5 times higher than that of similar NVIDIA products (despite NVIDIA shifting higher volumes), many among the audience of Hardware Unboxed are also reporting returning their 5700 cards in favor of alternatives from the competition, all due to frequent driver problems.

 

I didn't realize there were problems like this going on. Sounds like drivers can't fix the black screen problem whatever the cause is. And it affects both Navi and Vega?

 

At least the official stance at the moment is that it affects only Radeon RX 5700 -series (I wouldn't be surprised if 5600 XT too though, since it's the same chip), not all Navi and definitely not Vega. Also the fact that the workaround (disabling hardware acceleration on background apps) seems to work, it does suggest it should be driver issue rather than hardware.

 

Problem definitely happens on Vega as well. I think a driver update is due tomorrow though not sure it will target the black screen issue.

 

Let me rephrase: Vega doesn't suffer from the same issue, even if the crash itself is similar in effect (black screen). Also it's not widespread issue with Vega, but rather individual cards with different exact symptoms (even when the end result is the same), as suggested by the thread you linked to too.

 

Pharma's video a few posts back seemed to indicate that Vega cards have black screen problems too. Which means this has been a problem for a long time. People seemed to nail it down to something related to HBM2 dynamic clocks.

AMD is no stranger to problems caused by memory clock changes.

 

Just because something crashes to a black screen doesn't mean it does it for the same reason as something else crashing to a black screen.

Even in that very thread there's clearly different causes for those crashes (and different ways they crash even when the end result is the same) and it has never been a widespread issue like with 5700's where they all crash the same way (with this specific issue).
1st: Different type of crash if MSI Afterburner is running, if it's not there's signal loss and lockup and requires him to replug the monitors to the video card or the computer won't even post - clearly not the same issue. Reading further replies from him, looks like a bad card.
2nd: Freezes for several seconds before going black - clearly not the same issue as above or Navi
3rd: Freezes, no mentions of black screens - clearly not the same issue as above two or Navi
4th: Has audio playing normally for about a minute - clearly not the same issue as above three or Navi

Hope I didn't miss any with specifics how the crashes happens.

Vega might have some of the crashes related to HBM clock switching, but Navi isn't using HBM and I haven't seen any reports suggesting memory clocks switching before crash on Navi to tie even remote link between the two

 

I think Swaaye is referring to the Adored video. I've linked the point in the video where he experiences the black screen with Vega.
I agree it might not be the same issue, but it closely approximates the issue on Navi. In this case locking HBM2 memory fixed the issue. If a fix does come out for the Navi black screen issue it will be interesting to see what the culprit is (if that info is provided).

 

Yes I was referring to that video. That's why I wrote it was a few posts up.

So 5700 is much more problematic then.