AMD Cayman info (or rumor)

[zerazax]

You guys are reading WAAYYYYYY too much into PR slides

They have those funny graphs that start at 0.8 for performance slides, why would they put them exactly in a layout like that?

I mean, reading into pixels? If so, then why is the 6850 touching the 5770 when its obvious by now the 6850 is faster?

And for that matter, why is the 5970 so much higher than the 5870 when the 5870 should be quite big compared to the 5770?

Besides, they always make their dual GPU cards look much better than in actual performance... wouldn't want to make your own drivers look bad

Also, on the fliip side, look at the 6950... it's placed higher than the 5870 so if the 6970 is 17% faster than the 6950 (like 5870 is to 5850), and we say 6950 is even a conservative 10% faster than the 5870, the 6970 is going to be close to the 580 and probably be ~ the 5970 depending on the scenario. But then, I'd be looking way too much into this slide like everyone else, so that ought to prove to you how ridiculous it is to use this to guess actual performance

If 6970 is 225W and 6990 has to be 300W, this means an incredible amount of crippling has to be done to two 6970's. And since when you lower the frequency of a GPU its cost to you doesn't magically decrease, this would mean that AMD would be selling a card that, at stock speeds, wouldn't be much faster than a HD5970 (or even a single 6970) yet would be sold at something like $700. Everything is performance/watt when it comes to a TDP limit.

Yeah, I know, AMD will "beg us to overclock" the 6990 too, and by "overclocking" them to stock 6970 speeds we'd get a stupidly powerful graphics card; but the fact is that every review site will base its review on the stock speed, and will add "Overclocking" as merely another page among the 10 something page reviews. It will no doubt make the card look bad in performance and price.

So I don't think that would be very feasible. To me, with Antilles AMD has got two options:

1. Disregard 300W TDP and let the card overwhelmingly break single card performance and power usage records.

2. Build Antilles with 2x6870 instead of 2x6970. 2x6870 is already 12% faster than HD5970. 6870 is already 150W. Doubling it in a card will save one PCB worth of wattage, around 20W, which they can use to increase the clocks slightly to fill it to 300W and around a 15% performance difference from HD5970. The card would cost less than $500, but on the downside it wouldn't have the (possible) tessellation and DX11 tweaks the new Cayman architecture has.

It's pointless to argue this - the drivers already put Antilles as 2xCayman, so using Barts is a non-sequitur

Furthermore, the 5970 was already theoretically crippled - it was 2 x Cypress XT cores clocked at Cypress PRO. AMD can take it a step further and make it 2 x Cayman PRO cores @ Cayman PRO clocks, with Cayman PRO supposed to be < 225W (my guess is in the 175-200W envelope) which would fit with a X2 configuration.

That would actually leave room for AMD to keep the higher binned XT cores to reach higher clocks and thus make the 6970 stand out even further while the ones that couldnt reach such clocks get clocked lower / have units disabled and put in the PRO and X2 config