AMD Radeon 7000-Series 28nm (Southern Islands) | 7990 7970 7870 7770 | Discussion

[vid_ghost]

Southern Islands is a codename for the new AMD graphics architecture ( GCN - Graphics Core Next ) that is replacing VLIW4 used in the previous AMD HD6000 series of graphics cards. Southern Islands graphics processors will be manufactured with a 28nm process. The new architecture is expected to have significantly increased computational efficiency and higher double-precision FP performance-per-watt. Together with the promised enhanced programmability, Southern Islands should be more efficient for GPGPU tasks. Southern Islands based gaming graphics accelerators will be known as the AMD Radeon 7000 series.


_______________________________________

News:
2013 new cards incoming! http://www.fudzilla.com/home/item/30...rds-in-2h-2013
HD7870/7850 Review http://www.techspot.com/review/504-amd-radeon-7870-7850
HD7770 Review http://www.anandtech.com/show/5541/a...edition-review
HD7950 Faster then GTX580 http://www.fudzilla.com/home/item/25...md-hd-7950-day
AMD Roadmap http://www.xbitlabs.com/news/graphic...ary_March.html
Radeon launch schedule http://www.fudzilla.com/home/item/25...aunch-schedule
AMD Radeon HD7970 Full review http://www.anandtech.com/show/5261/a...hd-7970-review
Specs for lineup http://www.overclock.net/t/1190537/l...price-revealed

[Ospi]

Says late 2011 or early 2012 on that link. Anyway holding out for these.

[Pemalite]

I think we will see a larger increase than just 100%, especially with Tessellation and compute.
They did skip 32nm and went straight for 28nm after-all, so they have a fairly large transistor budget to play with.

Then again AMD had a larger die size than they would usually want this time around... So we could see a smaller die and a performance increase of around 100%.

[Ospi]

Ignorant question here. With a performance increase that large (potentially), will cpu's just bottleneck the system anyway?

[vid_ghost]

Yeah i dont think they will go "all out" on the new 28nm.. they will make a card they can make cash on over anything else because they know it will be fast enough... until the GTX680 hits they wont need to go all out.

[vid_ghost]

Originally Posted by Ospi

Ignorant question here. With a performance increase that large (potentially), will cpu's just bottleneck the system anyway?

No because by then we will have Bulldozer, second gen ( updated stepping with faster Mhz ) coming and LGA2011 Quad channel sandybridge as well as 22nm Intel Ivy bridge CPU's only months away. So it will be a very good match up.. CPU > Videocard

The real cards to look at will be the midrange 28nm cards that will be faster then GTX580 but cost around $200 bux.

[ChaosMaster]

I highly doubt this 100% upgrade. What makes you think that? If you look at the 4770, it was basically a 4830 that had been changed down from 55nm to 40nm. I reckon this will be about the same. I mean they just basically redid the architecture for the 6xxx series for Tessellation optimisation. There's no way there's going to be a complete overhall so quickly. Think somewhere along Intel's Tick Tock strategy. Performance optimisation and Efficency optimisation. They work on either of them, at different times, not together. I reckon a 50% gain will be seen. Form the 4770, we saw that from the same specs, the new fab and better efficency allows better clocks. And we can expect more SP's. So assuming it goes up to 2000SP's + 1Ghz clock rates at best, we'll see 50% increase in performance (30% extra from extra SP's and 20% from higher clocks and possibly higher VRAM bus [256bit for a while already] and clocks [doubt GDDR6 though]).

[vid_ghost]

Originally Posted by ChaosMaster

I highly doubt this 100% upgrade. What makes you think that? If you look at the 4770, it was basically a 4830 that had been changed down from 55nm to 40nm. I reckon this will be about the same. I mean they just basically redid the architecture for the 6xxx series for Tessellation optimisation. There's no way there's going to be a complete overhall so quickly. Think somewhere along Intel's Tick Tock strategy. Performance optimisation and Efficency optimisation. They work on either of them, at different times, not together. I reckon a 50% gain will be seen. Form the 4770, we saw that from the same specs, the new fab and better efficency allows better clocks. And we can expect more SP's. So assuming it goes up to 2000SP's + 1Ghz clock rates at best, we'll see 50% increase in performance (30% extra from extra SP's and 20% from higher clocks and possibly higher VRAM bus [256bit for a while already] and clocks [doubt GDDR6 though]).

The 4770 was a process shrink only just to make it cheaper to make and sell it had nothing to do with upgrading the architecture it was only a cost thing..

they may not be making a complete redesigning of the GPU but they will be making good use of the extra realestate with more pipelines and texture units... along with whatever else they had to cut out of the 6970 to make it fit onto the constrained 40nm process there's a reason the 5870 has more pipes then the 6970 and thats because they had to use the space elsewere. They wont need to do that this time around with the HD7970 and it will end up being the true next gen Videocard powerhouse with revised feature set and allot more then the 15% improvement we saw from 5870 to 6970.

One thing that makes CPU's and GPU's different is sometihng called feature creep... Everytime they release a new GPU they end up leaving out allot of features so to get the GPU out of the lab and into the store's its the reason 3DFX went bankruped, Feature creep stalled there products and killed them. Nvidia would skip features and add them into there next gen card that it was always already planing and working on. Think Direct X 10.1 and Direct X11 Nvidia didnt wait around to add those features they released what they had and worked on adding them onto there next card to come out 6 months later. Intel and AMD do not do that.. and realease new CPU's designs once ever few years sometimes longer.

Also 1.7GHz GDDR5 hynix memory already tap'd out a while back so we may even see faster GDDR5 memory!
but i think AMD will go for cheaper more available 1.7Ghz memory to maximise profits

[Concept CBF]

Why would they go for such insane numbers so fast when they can keep releasing products and making huge cash like apple does.
No software needs that sort of power lets be honest hardware these days is overpowered

[vid_ghost]

Its cheaper for AMD to make a 28nm GPU then a 40nm GPU thats why.

[Mr DOHC]

"THAN" FFS, "my wang is bigger THAN yours, THEN I will go to the shops"

Not that difficult! /rant

Hope it sees HD 6xxx drop their 2nd hand value, I may need a spare in a few months.

[Intensity]

i have heard about october this year. But time will tell.

[Autti]

6.8Ghz memory?

Those Hynix chips running at 1375Mhz are expensive enough, i doubt they even have a 1700Mhz chip to put on a card.

That and i doubt memory bandwidth will be a problem, perhaps make the bus wider which considering the shrink might be cheaper then using high end chips.

[walker_2003]

suits my next upgrade video card wise.

[2_stroke]

This is would be be sweet if it hits the stores bf3 time, now i've got some hope lol Its a wonder ati haven't got it to roll already with femi's delay in the 5xxx series and the spare time the had for r&d.