AMD: R9xx Speculation

[ECH]

Hmm, I wonder if it's possible that AMD will finally have the 6970/6950 as 1st and 2nd in overall price/performance this round based on the info so far? ?

 

[AlphaWolf]

Hmm, I wonder if it's possible that AMD will finally have the 6970/6950 as 1st and 2nd in overall price/performance this round based on the info so far? ?

You expect the 6950 to outperform the 5970? I think that's a bit wishful.

 

[Gipsel]

Hmm, I wonder if it's possible that AMD will finally have the 6970/6950 as 1st and 2nd in overall price/performance this round based on the info so far? ?

You expect the 6950 to outperform the 5970? I think that's a bit wishful.

Actually, I do expect the 6950/6970 to be better than 5970 in that metric.

But I also doubt a bit it will be in the #1 and #2 spots of the performance/price category. This is normally the regime of the slower cards.

 

[Squilliam]

So if the 6990 is ~775Mhz, what are the chances we'll see the HD 6970 at 900Mhz? They listed the TDP for that card at < 300W IIRC which implies higher voltages and therefore higher clocks are possible as well, right?

 

[trinibwoy]

Quite low you mean right? If they can drop those specs within 300w it would be very impressive. But numbers are boring. The exciting stuff is gonna be the arch changes. Probably AMD's biggest since R600.

 

[Mize]

I hereby move that we table all discussion of 69xx until such time as AMD decides to release working hardware.

All those in favor?

 

[RecessionCone]

I hereby move that we table all discussion of 69xx until such time as AMD decides to release working hardware.

All those in favor?

What is the Internet for if not idle speculation?

 

[BRiT]

I hereby move that we table all discussion of 69xx until such time as AMD decides to release working hardware.

All those in favor?

Are you asking for this speculation thread to be locked?

 

[Harison]

I hereby move that we table all discussion of 69xx until such time as AMD decides to release working hardware.

All those in favor?

I hope you are kidding :smile: Given AMDs level of secrecy, its a miracle we got spec leak months before cards release, and you want to stop discussions until card actually launched and reviews appear?

 

[OgrEGT]

So they do it again - two XT chips at PRO clocks?

But this time I don't believe they can stay around 300W in furmark. Most likely they rate it the nvidia way - max gaming load or something. Because if it is ca. 150W @775mhz, then what will the single be at? 1GHZ?

I was speculating about HD69xx power draw earlier on 3dc.
http://www.forum-3dcenter.org/vbulletin/showpost.php?p=8350776&postcount=26

Cayman XT 230W
Cayman Pro 190W

When taking into account HD5970 (216W in Games) to Cypress XT Crossfire (291W in Games), the second Chip contributes ca. 37% of the power draw of one complete Cypress XT card.

Therefore, if using 2 Cayman Pro GPUs or downclocked Cayman XT GPUs around 190W each, and estimating 50% add. Power draw from the second GPU, this would result in ca 285W for Antilles in Games.

 

[Shtal]

Now this is more interesting!

What's up with ONLY 160GB's bandwidth??

 

[no-X]

This slide was posted already. And it was proved, that somebody edited the values, so at least some of the specs may be incorrect.

 

[Shtal]

This slide was posted already. And it was proved, that somebody edited the values, so at least some of the specs may be incorrect.

Thank you.

 

[Silent_Buddha]

So they do it again - two XT chips at PRO clocks?

But this time I don't believe they can stay around 300W in furmark. Most likely they rate it the nvidia way - max gaming load or something. Because if it is ca. 150W @775mhz, then what will the single be at? 1GHZ?

AMD since Evergreen has had hardware protection and monitoring to prevent power virus like activities like Furmark. 4870 had app detection similar to GTX 580 for Furmark after a certain Cat. version. So it should work like Evergreen in making sure a rogue application doesn't make a card perform beyond safe operating boundries, without the use of app detection. In other words, it may or may not be more, but if it is, it shouldn't be by much.

GTX 580 is a bit odd as Nvidia has also had hardware protection and monitoring in previous cards, yet something must have gone wrong with GTX 580 for them to explicitely put in app detection in order to appear to have working protection to the level that enthusiasts and reviewers expect.

Regards,
SB

 

[kresek]

From a Polish site, but the slides do speak for themselves:

http://www.frazpc.pl/artykuly/652601,Architektura-ukladow-Radeon-HD-6970-i-Radeon-HD-6950.html

 

[Alexko]

What is the Internet for if not idle speculation?

Porn?








Sorry, couldn't resist.

 

[SimBy]

From a Polish site, but the slides do speak for themselves:

http://www.frazpc.pl/artykuly/652601,Architektura-ukladow-Radeon-HD-6970-i-Radeon-HD-6950.html

One wonders what was changed at all on that supposedly faked slide.

EDIT: never mind I see now. SIMD count is not yet decided supposedly.

 

[DarthShader]

From a Polish site, but the slides do speak for themselves:

http://www.frazpc.pl/artykuly/652601,Architektura-ukladow-Radeon-HD-6970-i-Radeon-HD-6950.html

The slides mention the NDA is lifted today, so we should see it on different sites too. The slides are pretty self-explanatory though, some interesting things coming up!

But no 32-bit FMA? Only two times faster tesselation than 5870? So it won't even reach 1tri/clock in real situations?

They also mention PCzilla saying:
HD6970: 900mhz, 1920sp
HD6950: 800mhz, 1536sp

Also; both 1GB and 2GB versions for HD6950.

 

[I.S.T.]

I fail to see the point in their new EQAA(AKA CSAA for those of us with nVidia GPUs). I mean, they just did the MLAA thing. Why bother with this?

 

[no-X]

Only two times faster tesselation than 5870?

Up-to 3-times faster compared to Cypress... says the slide. 2-times better than on Barts seems to be fine.