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

[Arty]

Pitcairn first ever GPU (factory clock) to break the GHz barrier?, 7950 OC potential is "substantial", translated from here

AMD tells us more easily able to mount frequency Pitcairn since TSMC has gradually refined its manufacturing process to reduce leakage currents compared to what was found on the first batch of GPU Tahiti. AMD tells us about it if it has a wide margin for overclocking it is not by chance but due to the thermal envelope of 250W, which has limited the scope for the reference frequencies. We would therefore not surprised to see AMD offer an evolution of the Radeon HD 7970 with higher frequencies in the next few months, taking advantage of improvements related to the production to complicate the life of Kepler! Finally, note that AMD indicates that the margin for overclocking the Radeon HD 7950 will also be substantial, but still keeps its specifications confidential.

 

[DarthShader]

I'm still trying to find out why Asus charging more though.

Voltage Tweak sticker.

 

[Arty]

1000/1375 Windforce 3X

 

[Alexko]

Nice cooling system. By the way, two SKUs are back in stock on Newegg. So the supply stream may be tenuous, but at least it looks steady.

 

[ECH]

Voltage Tweak sticker.

That's what I was thinking as well, the bios and the tweaking program.
---------------------
Edit:
It would be nice to see how the 7970 performs using a AMD FX process before and after the Win7 Scheduler Patch. Even though it suggest at best 2% improvement for heavy threaded apps.

 

[mczak]

Pitcairn first ever GPU (factory clock) to break the GHz barrier?, 7950 OC potential is "substantial", translated from here

Of course 7950 OC potential will be substantial (the non-top parts always overclock much better since their lowered shipping frequency is mostly artificial, unless the voltage is lowered, but I don't know if with "OC potential" they are talking about voltage tweaking too).
Pitcairn breaking Ghz barrier sounds about right too (well not sure if it would "break" the barrier or just "hit" it for the reference part). The lower end chips have same frequency limits but typically more power/thermal headroom (Barts was also clocked higher than Cayman, if only so slightly), even without process improvements. At least if Pitcairn is something like "2/3" of Tahiti it should have no problem hitting 1Ghz while staying well below 200W TDP (I can only assume the TDP target is similar to HD6870, which was 151W, supposedly PowerTune should make it easier to hit that).

 

[ECH]

Here is a list of games providing vram usage at 1080 and at Eyefinity
Overclock3D

 

[Jawed]

Ooh, count the pads!

http://www.hardware.fr/news/12088/ces-powercolor-hd-7970-pcb-custom.html

 

[fellix]

http://www.hardware.fr/news/12088/ces-powercolor-hd-7970-pcb-custom.html

2120

p.s.: ~1700 pads for Cypress, in comparison.

 

[air_ii]

I was just wondering. All the reviews I've seen (save for some compute stuff) concluded that PCIE3 made no difference with HD7970. But, wouldn't it play a role in CrossFire with limited number of lanes?

I mean with both Sandy Bridge and Ivy Bridget you only have 16 lanes (so 8 per card in dual card config), but Ivy Bridge should provide twice the bandwidth whereas 8 lanes of Sandy Bridge could be a limiting factor.

Or are 8 PCIE2 lanes sufficient per card for CrossFire?

 

[sir doris]

I was just wondering. All the reviews I've seen (save for some compute stuff) concluded that PCIE3 made no difference with HD7970. But, wouldn't it play a role in CrossFire with limited number of lanes?

I mean with both Sandy Bridge and Ivy Bridget you only have 16 lanes (so 8 per card in dual card config), but Ivy Bridge should provide twice the bandwidth whereas 8 lanes of Sandy Bridge could be a limiting factor.

Or are 8 PCIE2 lanes sufficient per card for CrossFire?

Take a look at Tomshardware 3 part sli/ crossfire investigation from the summer, in dual card setups i think Sandy bridge was just as quick if not quicker than x58 and 1156/55 setups with the nf200 chip.

 

[Kaotik]

Take a look at Tomshardware 3 part sli/ crossfire investigation from the summer, in dual card setups i think Sandy bridge was just as quick if not quicker than x58 and 1156/55 setups with the nf200 chip.

Other cards results are irrelevant, they can stress the PCIE bus differently.

 

[simbus82]

I think the "driver" can say a lot of things!

http://www.hwupgrade.it/forum/showpost.php?p=36720480&postcount=1509

Unigine with this driver (8.921.2 dedicated to 7900 series, compiled 4-01-2012)

http://support.amd.com/us/kbarticles/Pages/radeon-7900-series.aspx

+19% versus Catalyst 8.94 compiled 3-01-2012

The work that AMD is doing to the driver is... IMPRESSIVE.

 

[air_ii]

Let's hope we'll see such improvements in games as well.

 

[Arnold Beckenbauer]

http://twitter.com/#!/ID_AA_Carmack/status/157512179749371907

Are you going to add support in Rage for the 7970's hardware PRT Mega texturing anytime soon?

7970 hardware virtual textures are limited to 32k, so it requires layout changes to avoid tex crossing. Doom will support.

 

[Jawed]

Did AMD release the PRT demo that was shown to journalists?

 

[Silent_Buddha]

I missed the days when a new video card released was accompanied by public releases of demo's made specifically for those cards.

Regards,
SB

 

[Alexko]

http://twitter.com/#!/ID_AA_Carmack/status/157512179749371907

Exactly what does he mean about the 32k thing?

And by the way, when is Doom 4 due?

 

[fellix]

Exactly what does he mean about the 32k thing?

32K*32K maximum size for a single texture array probably. Rage uses 128K*128K.

 

[Alexko]

32K*32K maximum size for a single texture array probably. Rage uses 128K*128K.

Thanks. I thought PRT was inspired by Rage, but if that were the case, I imagine AMD would have made it compatible. So perhaps they came up with the idea independently? Or 128k × 128k was somehow too costly to do?