AMD radeon 6000 series (big navi)

archae86
archae86
Joined: 6 Dec 05
Posts: 3160
Credit: 7255329009
RAC: 1453313

Shortly after my joyful posts

Shortly after my joyful posts on the successful introduction into service of my new 6800 card, I started seeing very troubling episodes in which power consumption, temperature, utilization, and progress would all drop gravely in an instant, and stay low for periods from a few seconds up to 8 minutes.

I've seen things a little like this back when I first started using AMD cards with 570s about two years ago and never was completely free of it, though eventually it was down to a minor nuisance level  More recently with four 5700 cards and contemporary drivers, it has almost completely disappeared.  And I've not yet seen it a single time in running my 6800 Xt card for some weeks.

So far I've tried turning off the Windows function to power down the monitor, and I've set a minimum clock speed on the AMD software.  Neither fixed it.

In the past, this issue was often at the worst soon after a reboot, and would sometimes get greatly better in a day or so.  Perhaps I just need to be patient.  The only other current thoughts I have are to try the (somewhat older) WHQL driver, or perhaps to see what happens at 1X.

Surely other AMD users have seen this sort of behavior.  Have you found anything that makes it better or worse?

If it stays as bad as in the first couple of hours, this is a deal breaker.

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6529
Credit: 9631888160
RAC: 2859581

I would try turning off all

I would try turning off all the customization except keeping the fans going strong. If that doesn't return to baseline performance then I would shutdown the gpu management software completely and run on baseline.

Tom M

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

archae86
archae86
Joined: 6 Dec 05
Posts: 3160
Credit: 7255329009
RAC: 1453313

so I'm getting 177 second

so I'm getting 177 second elapsed times running 1X, with none of the third party software having run since I switched the driver to the WHQL version with a full DDU-cleanup removal of the newer non-WHQL one.  So far no trouble, but I have touched no knobs and it has only been 45 minutes.

If that condition runs OK overnight, I'll try things one at a time, waiting for trouble to return.  If that setup does not run for a night, I think I either have a sick system or a sick card, or I've somehow misconfigured something even without doing any intentional GPU-specific configuring.

Clearly the system is sick in some way, as the time from typing the password into the Windows login to getting the desktop is right about ten minutes (yes, I got to watch this several times today).  I'd be a bit surprised if the same issue is causing that plus my problems running the 6800, but it is possible.

Keith Myers
Keith Myers
Joined: 11 Feb 11
Posts: 5002
Credit: 18873757732
RAC: 6178611

If it takes that long to

If it takes that long to paint the desktop, something is hanging up the bus.

So, yes likely a device is dying or a bad driver is installed.

 

solling2
solling2
Joined: 20 Nov 14
Posts: 219
Credit: 1577881279
RAC: 11411

archae86 schrieb:

archae86 wrote:

...

Clearly the system is sick in some way, as the time from typing the password into the Windows login to getting the desktop is right about ten minutes (yes, I got to watch this several times today).  I'd be a bit surprised if the same issue is causing that plus my problems running the 6800, but it is possible.

So checking for incomplete windows updates or corrupted system files or updating mainboard chipset driver might be helpful as well ?

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3110
Credit: 4995516814
RAC: 1131863

To ARCHEA86, Unless the

To ARCHEA86,

Unless the computer that you are having troubles with, such as long boot times, is also your go-to computer for daiyly use where you have files saved on the primary hard drive, wouldn't it be easier and faster to just wipe it clean and start over with a fresh install of your operating system and then a fresh install of BOINC?

You can then set up your system as you had it before the problems began and start over.

Otherwise you seem to be wasting everyone's time trying to help you even if they are not sure where to begin.

Just a thought from an impartial observer.

George

Proud member of the Old Farts Association

Tom M
Tom M
Joined: 2 Feb 06
Posts: 6529
Credit: 9631888160
RAC: 2859581

From a admin command prompt:

From a admin command prompt: sfc /scannow might help.

A Proud member of the O.F.A.  (Old Farts Association).  Be well, do good work, and keep in touch.® (Garrison Keillor)  I want some more patience. RIGHT NOW!

GWGeorge007
GWGeorge007
Joined: 8 Jan 18
Posts: 3110
Credit: 4995516814
RAC: 1131863

Tom M wrote: From a admin

Tom M wrote:

From a admin command prompt: sfc /scannow might help.

"sfc / scannow" ??  You've lost me with that.

George

Proud member of the Old Farts Association

archae86
archae86
Joined: 6 Dec 05
Posts: 3160
Credit: 7255329009
RAC: 1453313

George wrote:Otherwise you

George wrote:
Otherwise you seem to be wasting everyone's time trying to help you even if they are not sure where to begin.

I was not seeking help on my extremely slow Windows login matter, though I've seen a suggestion here I found elsewhere I intend to try and I am grateful for suggestions including yours.

I was seeking to help fellow participants by providing information on a new card, and giving extra conditions of use notification to help put my observations in context  Also, this machine in question is my daily driver.  A quick wipe is not in the cards for me.

Back to the topic of the 6800 card.  I having removed the most recent non-WHQL driver and put in the latest WHQL on, the system ran flawlessly overnight at 1X (with no monitoring applications running, and not having visited AMD Radeon Software).  Bumping up to 2X it ran flawlessly with extremely little variation for several more hours.

Then I went into AMD Radeon Software, and hit a few buttons.  The only thing I intentionally changed was to set a maximum GPU clock rate of 2000.  I immediately observed a reduction in clock rate, GPU voltage, and GPU power consumption.  The first few tasks ran to completion consistently, and just a little slower., so things seemed good.  However much greater variation appeared later in the first hour.   I don't know whether it was coincidence, or another button I clicked, or simply setting 2000, but the behavior has moved in the wrong direction.  I suspect the AMD Radeon software is interacting with the card to do things I don't like.

Adding insult to injury, I tweaked settings further in AMD Radeon Software by setting a manual fan curve which pulled down reported GPU and memory junction temperature well over 10 C to very ordinary levels. 

So far the first six task completions after my latest interventions are a closely-packed group around 4:40 elapsed time (still running 2X) which would be splendid if it kept up.  We'll see.

I recognize that I am probably giving too much running commentary.  With one day under my belt I'll summarize by saying that Gavin's prediction that the 6800 might well be the price/perfomance and power/performance king of the three original Big Navi cards (6800, 6800 XT, 6900 XT) when running Einstein GRP seems borne out by my experience added to his 6900 XT observations.

Gavin
Gavin
Joined: 21 Sep 10
Posts: 191
Credit: 40644337738
RAC: 0

archae86 wrote: I recognize

archae86 wrote:
I recognize that I am probably giving too much running commentary.

For those of us that are interested I don't think there is such a thing as too much running commentary. Keep up the good work! 

archae86 wrote:
With one day under my belt I'll summarize by saying that Gavin's prediction that the 6800 might well be the price/performance and power/performance king of the three original Big Navi cards (6800, 6800 XT, 6900 XT) when running Einstein GRP seems borne out by my experience added to his 6900 XT observations.

It's nice to know that sometimes I am correct!


The base model 6800 seems to be the sensible way to go if you are looking for a new AMD card for crunching, the 6800XT and 6900XT variants are only really bringing increased core clock speeds to the table, which are little help in the grand scheme at Einstein.
I've said it before (probably in this thread) that AMD have missed a trick by not differentiating the raw performance of cards within the 6xxx series enough. The 6900XT should have been given hbm2 memory at the very least to make it the premium choice over the cards below it.

If 6800 cards were available to buy I would be in the market to invest (providing the price is somewhere near palatable).

Gav.

 

  

 

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.