Forcing The R9 Fury's Performance State For Better AMDGPU Performance?

[bridgman]

Sapphire has been very bad in the last 10 years, it's a wonder people still buy them.

Not sure this is a Sapphire problem at this point - don't we have users with Sapphire Fury's reporting better performance ?

[darkbasic]

Correct. At first we thought it might be, but on investigation we found that it was not.

Sorry for asking so directly, but I would like to know if it's possible to send you (where you means someone from AMD) the faulty card to check what's wrong. I can't image why it should be so difficult to spot the problem when the solution is so easy: sending the faulty card.

[bridgman]

Sorry for asking so directly, but I would like to know if it's possible to send you (where you means someone from AMD) the faulty card to check what's wrong. I can't image why it should be so difficult to spot the problem when the solution is so easy: sending the faulty card.

It's a reasonable question. If we can't repro the problem internally then we will need to get access to a system where it does show, although AFAIK it's generally easier to just remote-access into the user's system than mail cards around.

There's no indication of it being a "faulty card" at this point, just a configuration the driver isn't handling properly today.

[Passso]

Sorry for asking so directly, but I would like to know if it's possible to send you (where you means someone from AMD) the faulty card to check what's wrong. I can't image why it should be so difficult to spot the problem when the solution is so easy: sending the faulty card.

Better invite AMD devs at home to investigate so you could share some beers

[BenPope]

Not sure this is a Sapphire problem at this point - don't we have users with Sapphire Fury's reporting better performance ?

I think mine is a Sapphire Tri-X OC of some kind. I can provide further details later, if that's not specify enough. Should we provide bios versions? From memory mine ends in a bunch of zeros and an 8.

[bridgman]

I think mine is a Sapphire Tri-X OC of some kind. I can provide further details later, if that's not specify enough. Should we provide bios versions? From memory mine ends in a bunch of zeros and an 8.

I don't know yet, let's find out what HW we have in house first and try running Michael's most recent tests. The compute and graphics teams have different cards so we should be able to do some swapping around.

[andrei_me]

Correct. At first we thought it might be, but on investigation we found that it was not.

Oh, sorry then for the confusion...

About the compute and graphics team, I remember one test where Fury ran top charts and asked why, and you said that compute performance wasn't bottlenecked as graphic is in Michael's card, maybe with your team swapping card you can figure it out

[magika]

No, wrong. But probably you're so convinced by your opinion that nothing, not even facts, would change it. Ain't that right?

Convinced by my opinion? While I don't have AMD GPU's anymore, I follow Michael's benchmarks, and they are pretty damn convincing.

[pal666]

Convinced by my opinion? While I don't have AMD GPU's anymore, I follow Michael's benchmarks, and they are pretty damn convincing.

moron, their competition does not have linux driver at all

[trifud]

Michael, somebody mentioned that the problem could be related to Ubuntu. Did you try with another OS?