AMD 7xxx (GFX cards)

[ExtremeDude2]

This was so cool I just had to post it

http://www.maximumpc.com/article/features/amd%E2%80%99s_radeon_hd_7970_43_billion_transistors_pure_performance

[Zee530]

I Want One

---

Something that can handle 4xIR with 9xSSAA perhaps

[NaturalViolence]

Quote:Something that can handle 4xIR with 9xSSAA perhaps

Not a chance in hell.

[lamedude]

These are launching already? Fortune Street had me glued to the Wii all week.

[NaturalViolence]

It's a soft launch for now. Only review/test models have been shipped. They are officially launching the product on January 9th.

[dannzen]

but a hell of a card...
ATI is still in the gpu branch ahead of nvidia
nvidias tdp just shows how inefficent those cards are ö_Ö
they are just feeding the cards with power to Compensate the architecture quality...
(if the production quality is bad... just put more power into the chip to get him work)

[DacoTaco]

(12-23-2011, 11:42 PM)dannzen Wrote: (if the production quality is bad... just put more power into the chip to get him work)

and thats how theyve been doing things since the GF8 sadly (but damn the GF8800 series are motherfucking bad-ASS- powerful ;D )

[ExtremeDude2]

(12-24-2011, 01:11 AM)DacoTaco Wrote: (but damn the GF8800 series were motherfucking bad-ASS- powerful ;D )

Fixed that for ya

[DacoTaco]

(12-24-2011, 02:06 AM)ExtremeDude2 Wrote:

(12-24-2011, 01:11 AM)DacoTaco Wrote: (but damn the GF8800 series were motherfucking bad-ASS- powerful ;D )

Fixed that for ya

even for today they are still powerfull. im still playing skyrim 30fps on HD. that makes me happy seeing how (at some parts badly) optimised skyrim is

[NaturalViolence]

Quote:nvidias tdp just shows how inefficent those cards are ö_Ö
they are just feeding the cards with power to Compensate the architecture quality...
(if the production quality is bad... just put more power into the chip to get him work)

That's because nvidia uses a superscalar compute oriented architecture with an emphasis on thread level parallelism. VLIW is better optimized in terms of power consumption per GFLOP/s but it only performs well with games running through d3d, which the drivers are very heavily optimized for. Nvidias architecture was better for everything else and allowed them to do reasonably well in both the gaming and professional market at the same time. Ultimately nvidias architecture actually proved more profitable because of this and now AMD is looking towards the future too, realizing that in order to survive they must have a product that is competitive in both markets. And that's essentially what GCN (graphics core next) is. GCN sacrifices some power/peak performance efficiency in exchange for more reliable performance, essentially they've now done exactly what fermi did. Of course the major difference is implementation, fermi relies more on TLP while GCN relies more on ILP. This means nvidia will have an easier time maintaining performance reliability with this type of architecture while AMD will continue to have better power efficiency. The fact that AMD released this new architecture at the same time that they made the switch to a new fabrication process (TSMC 28nm) helps hide the reduced power efficiency but they are being smart and they are going to continue using vliw4 for developing low/medium end desktop gpus, mobile gpus, and apus (something nvidia should have done, nvidia should have kept using G80-G99 design principals for designing low end gpus). I wouldn't say the production quality or architecture quality of fermi was bad. Nvidia chose to sacrifice some power efficiency to build a more compute friendly architecture because.....quite frankly.....they had to, and despite the naysayers it paid off for them. It was a trade off, nothing more, a tradeoff that AMD is now going to have to make to remain competitive.

I'm actually a little disappointing with the 7970s performance. It's only slightly faster than the GTX 580 in most tests despite the fact that fermi is still fabricated on a 40nm process while Tahiti is fabricated on a 28nm process. Kepler is a couple months away and will be fabricated on a 28nm process. Nvidia is promising a shift back towards focusing on power/TDP efficiency now that they've got compute performance out of the way and is claiming that Kepler will have three times the performance per watt of fermi. I doubt nvidia will be able to hold true to such ambitious claims, but if they do, AMD is going to be in some trouble. And if they don't hold true to this, well any improvement at all over fermi will put kepler above tahiti in performance.

Edit: God damnit I've done it again.....written another article style post that is.