Topic: I've never had such a hard time doing anything, then I have trying to undervolt. (Read 706 times)

I went back to the painstaking method of using Polaris on Windows to just modify the BIOS and then flash in Linux, and test. Changing setting one by one.

The problem that I ran into was in Windows, I could dual mine Eth-Sia while using Anoraks base powersave Bios where he adjust the PowerTune wattages (not voltages). And in Windows I think the driver or some piece of software overrides those settings to up the Wattage when the card needs it. But in Linux, it does not, and sticks to the BIOS settings.

So, while it worked great in Windows, it was a disaster in Linux. So the whole test in Windows thing is useless (however, tuning the MemClock by checking for Memory Errors in Windows first, does save a few steps.)

Best advice I can give is to do the other cards first. For some reason XFX uses different voltage controllers.

Windows 10
WattTool Profiles

Make sure to have a monitor or headless HDMI dongle plugged in to at least one of the GPUs.

Wattman is junk.
Ignore it.
In fact, don't even INSTALL the worthless bloatware piece of garbage if you can avoid doing so.


 If you insist on checking the cards in Windows, use Afterburner or GPU-Z.

try claymore 9.2

config


-cclock 900
-mclock 1750
-cvddc 800
-mvdcc 800
-powlim -30

That could possibly be the best news I've ever heard. I have 5 Sapphires to do next and then one Powercolor. I have been dreading it as this XFX has taken me literally a week.

I don't get why on windows it runs 10W lower then on Linux (but only on certain cards, which is why it makes no sense).

It's not you, it's XFX. I have no problems with my msi or powercolors but XFX is hell

It has to be. Wattman always shows 930mV on a card that has incrememntal BIOS set. And usually on those BIOS's Wattman would show voltages ramping up to 1048mV, but now it shows 930 across the board, which was the settings I had used on a previous card.

So something is being cached, and there doesn't seem to be any way to reset it, whihc makes testing on Windows pointless (except for memory errors while adjust clock speeds.  I did find that I was getting millions of mem errors because my MemClock was set a tad too high). So I guess Windows does serve it's purpose.

It's also strange that in the BIOS the MemClock is set to 1875mhz/1000mV - But I have NEVER seen 1000mv in GPUz it's always 900mv, and Wattman ALWAYS shows 900mv when I switch to manual voltages, no matter what I do. So there is an extra level of setting going on that we aren't seeing.... So I have no idea what Linux is actually getting. So fun.

Pretty certain these are driver issues. I've got a 290x and a 295x2 in a mining rig. After reboots, I find that sometimes I am unable to get one or the other to the same overclock settings/consistency without modifying the power limit or core voltage further.

EG:
1100/1300 @ -45 vcore/+0 power will work for my 295x2 except for the rare times it boots and I have to do -55 +5

The same thing occurs on my 290x. I haven't been mining for long, this problem did not occur when I was mining only on my 290x. After I added in my 295x2, this started. That was also when I moved the hardware to its own dedicated rig and thusly changed everything... could be something in that hotpot of variables. Still though, I have a pretty high level of confidence it's the drivers.

I just put in a card, that has the incremental automatic mV bios, which WAS running at 1.025V, and now it's running at 900mV, so there MUST be some kind of cached setting for the mV for the GPU Clock. This makes absoluetly no sense.

I have used Wattman to reset to defualt, and even those the BIOS is set to auto - it still thinks the mV setting is 930mV, but the actual draw is 900mV.

 

Yes, that is correct. And I am adjusting it in the BIOS, the problem is to actually make sure the BIOS settings are working is to use Windows to monitor it. So I just forced a 930mV for all GPU Clocks in the BIOS. BUt when I run the card in Windows, it runs it at only 900mv (and runs it gosh darned stable!) even though Wattman recognizes it at 930mV. So where is it getting that 900mV setting from?? At the wall it's 160W total with computer. However, in Linux it runs god knows what and is at 180W at the wall.

Now if I use that exact same BIOS on an identical card, that new card shows 930mV in windows. And then if I revert the BIOS back to Anaroks defualt with auto-incremental voltages on the card that was running great at the 900mV, it continues to run at 900mv, UNTIL I put a new card in the MOBO and go back to the 900mV card, THEN it runs at 1.025mV

I am just going to have to chalk this up to impossible unless someone has dealt with this before, because I imagine trying to follow this is impossible...

My understanding is the only way to undervolt AMD cards in Linux is to do it through the Bios. In Windows I use the top slider in Afterburner after enabling voltage control in the settings to undervolt.

This is ridiculous. I have a bunch of identical XFX RX 570s. I am using Ubuntu 16 with Claymore 9.5.

I am using Windows 10 to test for mem errors and check voltages. I have Wattman and Trixx (got rid of Afterburner). I have GPUz to monitor voltages.

Wattman seems to have a mind of it's own and sometimes reads the BIOS voltages I set and other times just flat out ignores them. Using the default auto-incremental voltages sometimes it runs at 1.025v but then on other cards it will run at 900mv (which is great). However, if I swap cards and it decides to run at 1.025v and then I swap back to a card that was just previously running at 900v, suddenly that 900v card will now run at 1.025v. I have reset to default everywhere to try and force the BIOS settings. Even manually setting the card to 925v, the Wattman will show 925, but if I reset the card back to default, Wattman still shows 925 but the power usage is much higher.

I know this makes no sense to read, but how do you all manage undervolting in LINUX? I have a kill-a-watt and most of my cards will run at 120W, but it's seems totally random at whatever the last read on Windows is, affects how it runs on Linux.

I'm using a modifed Anarok Powersave BIOS and slightly lowering the MemClock as it wasn't running stable. He does lower the TDP/TDC settings, which seems to work well, but again, once windows gets involved everything is totally random. I don't get why sometimes it runs at 1.025v and others at 900mv when the settings aren't changing....

I'm just so confused....