AMD: Southern Islands (7*** series) Speculation/ Rumour Thread

[Kaotik]

Interesting - a presumed leaked review apparently written by AMD's marketing staff.
I wonder, since it doesn't say, if they used 3 GB GTX 580 cards or intentionally crippled the figures for larger resolutions by using a 1.5 GB version.

Does it really?
At TPU, using same drivers, games etc, closest 2 matches I could find regarding clocks (while having same drivers & games)
GTX580 3GB @ 832/1050 is 5% faster than reference 1.5GB at 2560x1600
GTX580 1.5GB @ 816/1002 is 3% faster than reference 1.5GB at 2560x1600

So 3GB had 2% higher core and nearly 5% higher mem clock, and was only 2% faster at the highest resolution, compared to 1.5GB version.

You really think that 1.5GB cripples it?

(yes, there can be games where it does, but on average it doesn't, and the TPU benches include many of the games used in AMD benches)

 

[Mize]

Yeah, 25x16 rarely chokes 1.5 GB (have to have very high MSAA), but it's also not a terribly high resolution. Those slides have 76x16 which will definitely cause problems. Also, like most PR reviews, the settings are rather curious for each game - tailored for sure.

I have no doubt the 7970 will slaughter the GTX 580, but I do so detest slanted data.

 

[Picao84]

3GB vs 1.5GB GTX580 at 25x16
http://images.hardwarecanucks.com/image//skymtl/GPU/EVGA-GTX580-3GB/EVGA-GTX580-3GB-35.jpg

http://images.hardwarecanucks.com/image//skymtl/GPU/EVGA-GTX580-3GB/EVGA-GTX580-3GB-39.jpg

http://images.hardwarecanucks.com/image//skymtl/GPU/EVGA-GTX580-3GB/EVGA-GTX580-3GB-45.jpg

http://images.hardwarecanucks.com/image//skymtl/GPU/EVGA-GTX580-3GB/EVGA-GTX580-3GB-68.jpg

http://images.hardwarecanucks.com/image//skymtl/GPU/EVGA-GTX580-3GB/EVGA-GTX580-3GB-50.jpg

http://images.hardwarecanucks.com/image//skymtl/GPU/EVGA-GTX580-3GB/EVGA-GTX580-3GB-55.jpg

http://images.hardwarecanucks.com/image//skymtl/GPU/EVGA-GTX580-3GB/EVGA-GTX580-3GB-57.jpg

I think he was talking about the CrossFire/SLI comparison @ 7600x1600 with 8xAA where 1.5GB per GPU are definitely not enough

 

[DieH@rd]

Crossfire scaling is really impressive.

 

[Mize]

Call me jaded from my CFX set-up and so many "hot fixes," but I'm anxious to see how good the drivers will be.

 

[Love_In_Rio]

The problem is that Nvidia next gen is going to be 2xGTX580 performance...

By the way the 3dMark physics results aren´t strangely lows?.

 

[Mindfury]

this is such a boring silly season............ nothing will live up to the hype of R600.................

The problem is that Nvidia next gen is going to be 2xGTX580 performance...

Now it is happening...

 

[Mize]

The problem is that Nvidia next gen is going to be 2xGTX580 performance...

By the way the 3dMark physics results aren´t strangely lows?.

Doesn't 3dMark use PhysX for their "physics" bench? That would explain the low numbers.

 

[Love_In_Rio]

Doesn't 3dMark use PhysX for their "physics" bench? That would explain the low numbers.

I don't know but the result with 6970 is higher than with GTX 580 and also higher than with 7970.

 

[Kaotik]

Doesn't 3dMark use PhysX for their "physics" bench? That would explain the low numbers.

11 uses Bullet, not PhysX. Vantage used PhysX

 

[Mize]

That is odd. 580>6970 makes sense but 6970>7970 ?
Kaotik: thanks, my bad.

 

[neliz]

Doesn't 3dMark use PhysX for their "physics" bench? That would explain the low numbers.

Le Bullet monsieur, Le Bullet.

 

[Kaotik]

The problem is that Nvidia next gen is going to be 2xGTX580 performance...

Just like 480 was released in 2009 and Kepler (the next gen) still coming this year?

(and the nV chart is about perf/watt, not performance, and it's DP, not SP which is used in games

 

[Love_In_Rio]

Just like 480 was released in 2009 and Kepler (the next gen) still coming this year?

(and the nV chart is about perf/watt, not performance, and it's DP, not SP which is used in games

Yes, but a 3-6 months difference is not enough to make people no wait for a card supposely 50-75% faster...Above all taking into account this generation is going to last for a long time. Getting to 20nm is going to take 2-3 years...at least for non intel fabs.

 

[Kaotik]

Yes, but a 3-6 months difference is not enough to make people no wait for a card supposely 50-75% faster...Above all taking into account this generation is going to last for a long time. Getting to 20nm is going to take 2-3 years...at least for non intel fabs.

Kepler can be "supposedly" anything, but so far we got rumors about just all over the place.
Unless they want to repeat Fermi with too big die, I doubt we're seeing such performance figures.

 

[Mize]

Yes, but a 3-6 months difference is not enough to make people no wait for a card supposely 50-75% faster...Above all taking into account this generation is going to last for a long time. Getting to 20nm is going to take 2-3 years...at least for non intel fabs.

I totally disagree. If 7970 delivers these kinds of numbers, CFX scaling *and* solid drivers I think plenty of people won't wait to see if Kepler misses its planned launch. A solid launch of the 79XX will require rock-solid and uber-compatible drivers though.

 

[Man from Atlantis]

you guys shouldnt click a website that you dont even understand it's language lol

 

[Man from Atlantis]

 

[Man from Atlantis]

 

[Man from Atlantis]

http://www.abload.de/gallery.php?key=qpOM1lvz