AMD Cayman info (or rumor)

[Sentential]

The issue with a "simple" BIOS change is that you have to take all the cards back through Quality Assurance/Control. And if they've already produced tens of thousands of cards, that can take a lot of time.

In fact, Dave Baumann of AMD posted about this over at Beyond3d


Reading between the lines a bit, he's saying that changing clocks too late in the game can cause potentially months of schedule alteration.

Now if the 2-3 week delay is where it stands, it seems to rule out changing clocks and BIOS'es as well.




The second issue, which you touched upon in your first paragraph, is that AMD would have a hard time on the same process.

Yes, the same process is tough... but AMD has done it before. Remember RV670->RV770? That was a tad bit more than 40% more performance on the same process, don'tcha say?

And here's the kicker... if Cayman is really as new an architecture as the newest leaks/rumors are suggesting, then 40% isn't unheard of on the same process - RV770 had major changes but wasn't nearly as new an architecture.

The second thing to point out is... where is AMD's goal in reaching *just* 480 performance? Keep in mind Cypress came out last September. They've had over a year to come out with a replacement. They saw Fermi come out late, but they knew (Nvidia said it themselves) that GF100 is supposed to have 512 chips. They also knew that the Fermi series clocks really well.

Why would AMD expect that a year AFTER Fermi was supposed to be launched, that Fermi couldn't have a 512 SP card with more clocks? Of course they can expect it, they know Nvidia has enough resources that they can have extra teams devote to fixing whatever was broken for a refresh before a card is even released.

So that goes into the other point... what logical reasoning would AMD have to settle with getting a card out that only matches what their previous competitor was supposed to have a year ago? It makes no business sense, especially to release the 68xx's at a price point which forces your higher end cards to perform even better than the 480.



Yes, executing != knowing. Both sides have seen this happen many times.

But AMD got Barts out on time, on the same 40nm process. Heck, AMD got the entire Evergreen family out on time, on the same 40nm process. Whether they reached their performance targets is another matter, but I don't think AMD is likely to suddenly stumble when they had no problem getting things out month after month after month

As for drama... well I hope people like popcorn

Couple points Id like to add to this, yes what you say about the RV770 against the RV670 is quite true but by the standards of the day both were small chips. The last "real" big chips from ATi we have are the R520/R580 and the R600. Both had notorious leakage issues, both had mixed yields. When you compare to nVidia you have the G80, GT200 and now GF100/GF110. People like to make comparisons about how wasted and bloated nVidia's die space has been as it has been running about twice that of ATi's. But they have several years of eventually pulling off large monolithic chips at reasonable profit margins; ATi does not.

ATi made a tactical decision several years back that they would not continue large monolithic chips ever again; I applaud that effort however in order for that tact to work well they need to have functional drivers.While they as you put it have had a year to finalize Caymen they have had over 10 to have acceptable drivers that I can immediately think of starting with the 9700PRO. Promises have been made and broken and while I like what I hear about Caymen the talks of them having a superior product as a whole simply has not happened.

This has been their achilies heel in the mobile segment as well which is why they continue to not make inroads on switchable graphics as their implementation blows. So now they are faced with a non broken Fermi, and now the same people who laughed and said this was the worst chip of all time now all of a sudden people think its the best damn thing since sliced bread. Yet I and others reminded people that they (nvidia) have a tendency to do poorly on first gen parts and exceedingly well on revision parts were laughed at and were told nVidia is at their deathbed. NV30, NV40, GT200, and now GF100 are good examples of this. All were laughed at and were later vindicated as ATi continued to flounder at making an acceptable software package.

So again here we are with a similar situation; nVidia has a working chip, ATi has had an excellent chip but once again are the drivers going to pull through? My guess is no. Unless Caymen is significantly better GF110 it will fail; why? Because ATi has yet to address the fundamental problem with their products; their drivers. Until they address that in a serious way I will continue to *not* take them seriously.

In the end; powerhogging aside; which nvidia is certianly guilty of; what exactly does ATi do that is so much better than nVidia that would justify the littany of problems that purchasing said card brings?

Heat can be controlled.
Power can be controlled.
In most games they are neck in neck, in others ATi gets thrashed, in no game does nVidia fall behind as ATi does.
I am not a programmer and therefor I cannot control driver bugs.

Therefor I take the lesser of two evils which has been for as long as I can remember; nVidia. Also while I am thinking about it;

Now, if you place the 570 there, you say its meant to counter the 6970...

But that would say the 6970 is ~ GTX 480 performance, which means its only ~16-20% faster than the 5870.

Then where does the 6950 go? The 6870 is already nipping on the tails of the 5870 (lets say 10%), so the 6950 certainly would have to be slotted higher, especially if AMD wants to make any sense of return on a cut down Cayman chip.

In that case, lets say 6950 is in a worst case scenario only 15-20% faster than the 6870 - reasonable given that Cayman is at least Cypress size.

However, that would mean the 6950 is only 5-10% faster than the 5870, and thus is only 10-15% slower than the 6970 which is at 480 performance.

Which again, is way way way too close for how AMD has ever slotted their cards - even the 5770 and 5850 had very sizeable performance differences.

A great chart for performance summary this is here from Xbitlabs

I heard this exact same argument back in the x800XT days about how DX9c didnt meant anything; turned out they were wrong, dead wrong; the bottom line is this: ATi as it stands now will continue to fall behind as time goes on like it has in the past because often they are not ahead of the technology curve. They are either behind or just at it and so anyone who reads this must understand the moment real DX11 games arrive that use tessalation; all the numbers you listed are going to get on. ATi's use of tesselation with the current crop of 6800 series is at best half as good as nVidia and thats being generous. Also Id like to say that this isnt just an ATi issue; I remember a couple nvidiots claiming how ty the 9700PRO was as it was beaten soundly by the FX5800 in DX8 games but got reemed when it was actually forced to use DX9; dont be fooled.

If Caymen gains significant grounds on tesselation then I take what I said back; but if its more of the same you'd be a fool to buy it.