AMD: R9xx Speculation

[PSU-failure]

Well Neliz, ATI own slides stated it was >225W (8pin+6pin). While it might not be close to 300W, its not also far from GTX580 244W does it? To keep the same perf/watt of Cypress, they really have to perform great then!

Are you trying to compare 240 watt TDP and 244 watt max gaming power draw?

Remember GTX580 is in fact a 350 watt TDP monster, if using the same TDP definition as the one they used for the GTX200 boards...

 

[NathansFortune]

ditto me on this query....what gives?

28nm delayed to mid 2012.

 

[Picao84]

those slides are outdated and this value isn't the kind of value, which should be compared to nVidia's "typical gaming power" or how they call it

Why do I get the impression that ATi is pulling a Fermi in "Yields are bad at 1920 SPs, so we make it 1536 SPs with lower TDP, YEY!"

Are you trying to compare 240 watt TDP and 244 watt max gaming power draw?

Remember GTX580 is in fact a 350 watt TDP monster, if using the same TDP definition as the one they used for the GTX200 boards...

Where does that change the fact that Cayman is supposed to need 6pin+8pin, versus Cypress 2x6pin?

Not false, Fuad conveniently leaves out the low TDP at which this is achieved

Not anymore: http://www.fudzilla.com/graphics/item/21156-cayman-6970-has-less-than-300w-tdp

He might have revealed also whats that switch for. At this point, with this kind of info, I believe hes speaking the truth.

 

[Mize]

if those French prices are accurate and you couple that with estimates on the die size how can you conclude anything other than Cayman XT beats the 580 handily?

Certainly if it's slower AMD have ample room to undercut the 570/580 prices...?

 

[iMacmatician]

28nm delayed to mid 2012.

You gotta be kidding me…

 

[A1xLLcqAgt0qc2RyMz0y]

Who said, that HD6970 isn't able to outperform GTX570?

The 6970 just lost 20% of its ALUs from 1920 down to 1536. Thats a 20% haircut. Most of the speculation on this forum has a 1920 ALU 6970 slightly beating a GTX580. The GTX570 is about 15% under the performance of the GTX580. So that 20% haircut looks to put the 6970 around the performance of the GTX570.

 

[Mize]

28nm delayed to mid 2012.

I must've missed that. Ouch.

 

[NathansFortune]

http://www.pcper.com/article.php?aid=1045

Aricle says early 2012, but I have heard mid 2012 is a more realistic target, meaning the first cards with 28nm we will get will be Q3/4 2012.

 

[Mize]

the 28 nm delay affects both companies equally

 

[NathansFortune]

It doesn't since ATi haven't had 40nm issues like Nvidia. It also gives Nvidia more time to get a decent 28nm design going, something they won't have done since their 40nm designs needed more sprucing. It gives gives them a lot of help where they would previously had to rush designs falling further behind ATi's more power efficient chips.

 

[chavvdarrr]

28nm delayed to mid 2012.

Source? TSMC only or GF also?

 

[A1xLLcqAgt0qc2RyMz0y]

It doesn't since ATi haven't had 40nm issues like Nvidia.

What about the 1920 designed to 1536 working ALUs on the 6970. Doesn't that indicate yield issues?

 

[NathansFortune]

At the moment it's just whispers from Taiwan, but they are getting louder. Both companies seem to be having trouble with 28nm, hence the talk of delays.

It should be official in January sometime and schedules will be adjusted accordingly.

 

[NathansFortune]

What about the 1920 designed to 1536 working ALUs on the 6970. Doesn't that indicate yield issues?

Until we hear from somone other than Fuad I won't put too much into the 1536 claim...

 

[mczak]

Hmm some of the information doesn't quite fit still.
If that's only 24 simds, then it should be a quite small chip, as apart from the VLIW-4 change (which makes the simds smaller), I've yet to see rumoured changes which would increase chip size significantly. Ok the second graphics engine but I doubt that's really that large, and the ROP changes don't look very big imho.
But clock is also similar to HD5870, so why would it have higher (max) power draw?
Also, with only 24 simds, it seems that there must be significant changes elsewhere, otherwise I fail to see how it could be more than slightly faster than HD5870 (that is, the performance increase must come not from the VLIW-4 change and hence more simds, but from somewhere else - maybe write coalescing in the rop helps but I doubt that would be enough).

 

[Mize]

Like I said, if it's 1536 and slower than a 570 or even 580 then the French pricing has to be bullshit - you can't have both. No company is going to launch a slower card at a higher price when they have ample room to price lower.

 

[Kaotik]

What about the 1920 designed to 1536 working ALUs on the 6970. Doesn't that indicate yield issues?

First you need to remember that 1535 is just a rumor.
Then you have to remember that 1920 is just a rumor too.

We have no idea wether it was "designed to 1920" or not, or will it have 1536 in the end or not (though it seems likely atm due Neliz's post.
Heck, it could also be just 6850 having 1536, at least Neliz never said which model he was talking about on the "not false" part

 

[ECH]

Or Fuad was talking about the 6950 instead of the 6970 .

 

[mczak]

http://www.pcper.com/article.php?aid=1045

Aricle says early 2012, but I have heard mid 2012 is a more realistic target, meaning the first cards with 28nm we will get will be Q3/4 2012.

Wow. 3 years without any shrink on the same process node. Essentially 3 generation of graphic chips.
That's not so good. I think intel is on track to release 22nm processors in q4 2011, a full year ahead of gf / tsmc 28nm?

 

[neliz]

Not anymore: http://www.fudzilla.com/graphics/item/21156-cayman-6970-has-less-than-300w-tdp

He might have revealed also whats that switch for. At this point, with this kind of info, I believe hes speaking the truth.

No he didn't I have the same slides and even then he can't even read them properly.
He'll probably write about the Dual BIOS switch later, but his articles would make a lot more sense if he just copied what was on the slide and not add his own interpretation to it.

Like the part with the slide, he talks about 20W + or - on the performance, he completely FAILS there.
The -20 to 20 slider is a %, you can adjust the power of your card in the overdrive tab by 20% up or down.
the 250W is the maximum board consumption, but the non-OC power consumption is 190/200W.
Cayman can actually clock almost as flexible as any modern CPU, so it can run at 650 MHz or 800 MHz during benchmarking if you enable Powertune