AMD acknowledges 390 series crashes, looking into it

[DEDRICK]

15.7 did weird shit to my R9 290 so I'm not surprised at all

[xAcid9]

i must be doing something wrong for not getting all these crashes, tdr and black screens.

[CTR640]

It has lower power consumption than a 970? On what planet?

Maybe on this planet:

http://linustechtips.com/main/topic/413841-nasa-discover-earth-20/

[winningsince1337]

15.7 did weird shit to my R9 290 so I'm not surprised at all

Such as?

[Humbug]

15.7 seems to work great on my 290, and my friend's 280x.

But then we haven't had any issues with previous drivers either...

[Humbug]

Meh, all drivers have issues. Next WHQL driver should be out next week for the windows 10 launch.

I know every driver will have issues for somebody.

But what makes you think they will have another WHQL driver for windows 10 launch? Catalyst 15.7 just came out and was quite a successful release with new features etc. I expect the next couple of releases to be beta, which is ok.

[zMeul]

I know every driver will have issues for somebody.

But what makes you think they will have another WHQL driver for windows 10 launch? Catalyst 15.7 just came out and was quite a successful release with new features etc. I expect the next couple of releases to be beta, which is ok.

the problem with W10 is that MS is forcing WHQL drivers over user-installed ones; there is a work-around, but not many people know about it .. they just assume it's "normal"

[patrickjp93]

You don't own a Gigabyte GTX 970 G1 Gaming so you really don't know what I'm talking about:

 

Read the box.

I know exactly what I'm talking about. In binning, the spec is the lowest common denominator for a set of chips which becomes a SKU. You got burnt and ended up on the low side of the bell curve for the set of chips they binned for that SKU. The box doesn't falsely advertise anything. Overclockers cannot expect anything more than what's the in the specs. That's the same lottery we all play, just with a higher starting point of possibility.

[Unhelpful]

You guys really this biased against AMD?....

I'm not. I'm just sad that the 7970 and 7950 are still mid-high end cards. They came out in the end of 2011 iirc. Not biased. Just sad.

[Khvarrioiren]

I'm not. I'm just sad that the 7970 and 7950 are still mid-high end cards. They came out in the end of 2011 iirc. Not biased. Just sad.

 

And why is it sad? 28nm is only going to get us so far.

[patrickjp93]

And why is it sad? 28nm is only going to get us so far.

And I predict 16nmFF will be where TSMC gets stuck for a while. If Intel's having trouble with high power, so will the other foundries. The very fact TSMC is using 20nm BEOL to improve yields (at the expense of less density) is proof things are not going to speedup any time soon. We'll get 14/16nmFF cards, and then we'll be on that node for a while. I doubt TSMC or Samsung will get 10nm working very soon, even if they use 14/16nm BEOL, and Samsung has said GloFo will not get its 10nm node licensed like 14nm was.

[Dabombinable]

I know exactly what I'm talking about. In binning, the spec is the lowest common denominator for a set of chips which becomes a SKU. You got burnt and ended up on the low side of the bell curve for the set of chips they binned for that SKU. The box doesn't falsely advertise anything. Overclockers cannot expect anything more than what's the in the specs. That's the same lottery we all play, just with a higher starting point of possibility.

Case and point, my GTX 970 G1 Gaming meets none of the target goals advertised by Gigabyte.

[patrickjp93]

Case and point, my GTX 970 G1 Gaming meets none of the target goals advertised by Gigabyte.

It hits the guaranteed speed specs. If it doesn't, you have an RMA case, and likely there was a handling or shipping mistake which caused damage to the unit as a whole. Other than that, you are owed nothing.

[Unhelpful]

And why is it sad? 28nm is only going to get us so far.

The state of the industry is what makes me sad. 3 and a half year old cards shouldn't still be competitive. AMD released 2 new high end cards in the last 2 "generations." As a result nvidia has also slowed to a chopped-up snail's pace.

[patrickjp93]

The state of the industry is what makes me sad. 3 and a half year old cards shouldn't still be competitive. AMD released 2 new high end cards in the last 2 "generations." As a result nvidia has also slowed to a chopped-up snail's pace.

Oh come on. It's not like rendering techniques and details have actually gotten so much better in the last 3 years as to necessitate 8x the power we have now. The only thing that necessitated newer cards has been higher memory usage and less optimal code that needs brute-force power to speed up.

[Dabombinable]

It hits the guaranteed speed specs. If it doesn't, you have an RMA case, and likely there was a handling or shipping mistake which caused damage to the unit as a whole. Other than that, you are owed nothing.

62.1 Asic, binned supposedly for 'the best power efficiency and lowest power consumption' (you don't get that with such a low Asic score)-that's not meeting the guarantee especially when you consider that the 970 can not be overclocked at all (aka its the worst graphics card I've ever owned in regards to overclocking-even when compared to my old Diamond Stealth cards). And read my previous post in regards to the RMA.

[itachipirate]

So much AMD hate in the comments here

 

Like Nvidia never had Driver problems. HAHA

[Unhelpful]

Oh come on. It's not like rendering techniques and details have actually gotten so much better in the last 3 years as to necessitate 8x the power we have now. The only thing that necessitated newer cards has been higher memory usage and less optimal code that needs brute-force power to speed up.

Top-end GPUs still struggle to run newer games at max 1080p 60fps without dips.

[patrickjp93]

62.1 Asic, binned supposedly for 'the best power efficiency and lowest power consumption' (you don't get that with such a low Asic score)-that's not meeting the guarantee especially when you consider that the 970 can not be overclocked at all (aka its the worst graphics card I've ever owned in regards to overclocking-even when compared to my old Diamond Stealth cards). And read my previous post in regards to the RMA.

But unless you can pull an objective study against the cards in the lower tiers, then you haven't a leg to stand on. All you are guaranteed is the speed in the spec. It very well could be a one-off miss, but it could also be that one card that does the clocks that they can sell at that SKU and doesn't do any better, and that card has to exist. If it didn't, the whole SKU would be clocked higher, even if marginally so.

[patrickjp93]

Top-end GPUs still struggle to run newer games at max 1080p 60fps without dips.

And that has everything to do with poor coding. I disassemble and decompile code all the time. It doesn't help Microsoft's compiler doesn't optimize that well, but it comes down to code that ships broken and buggy in the first place. Nvidia has already stated its drivers fix code that is actually broken when it ships. It has nothing to do with games squeezing out every possible drop of performance. GPUs are far more powerful than game programmers today make them out to be.

[Unhelpful]

And that has everything to do with poor coding. I disassemble and decompile code all the time. It doesn't help Microsoft's compiler doesn't optimize that well, but it comes down to code that ships broken and buggy in the first place. Nvidia has already stated its drivers fix code that is actually broken when it ships. It has nothing to do with games squeezing out every possible drop of performance. GPUs are far more powerful than game programmers today make them out to be.

While ALL OF THAT is true, we are still at the slowest point in GPU development since the 8800 gt rebrands, and it isn't showing any sign of getting better.

[Khvarrioiren]

And I predict 16nmFF will be where TSMC gets stuck for a while. If Intel's having trouble with high power, so will the other foundries. The very fact TSMC is using 20nm BEOL to improve yields (at the expense of less density) is proof things are not going to speedup any time soon. We'll get 14/16nmFF cards, and then we'll be on that node for a while. I doubt TSMC or Samsung will get 10nm working very soon, even if they use 14/16nm BEOL, and Samsung has said GloFo will not get its 10nm node licensed like 14nm was.

 

Indeed. Tri-gate designs are a bitch and a half to scale down anyway.

[Doughnutnator]

Must be a driver issue.

 

Happened like a month ago on Nvidia drivers, not everybody got that issues though (and must be the same with AMD, just a few people get this)

Nice to see that AMD acknowledges the issue, not like Nvidia that just releases a freaking hotfix or keep releasing drivers without even mentioning the issue.

 

Do you guys thing that this is related with both AMD and Nvidia getting their drivers ready for Windows 10? It's so weird to see this happening right before the new OS launch. Just a theory, maybe it's something else.

[Techhog]

But unless you can pull an objective study against the cards in the lower tiers, then you haven't a leg to stand on. All you are guaranteed is the speed in the spec. It very well could be a one-off miss, but it could also be that one card that does the clocks that they can sell at that SKU and doesn't do any better, and that card has to exist. If it didn't, the whole SKU would be clocked higher, even if marginally so.

To be fair, something like that should fall into the standard Windforce X3 SKU (which is literally exactly the same, except that it is lower clocked and doesn't have GPU gauntlet or a backplate).

[Cainn]

Hey AMD users, don't feel bad. Nvidia has some serious driver issues right now also. It's fucking garbage.