SOME NEW INFO on the latest round of AMD graphics processors has suspiciously reared its head on a German hardware website, purporting to show the full spec list for DAAMIT's range of next-gen mid-range products, based on RV770.
The 4870, 4850 and 4850 GDDR 3 are all built on DAAMIT's 55nm processor, have 480 stream processors, 32 TMUs and 16 ROPs, according to the data sheet.
The former will come with high-end GDDR5 memory for those that simply have to have the latest in incremental numbering systems, and the real differentiators will be the chip and memory speeds: 850/1050MHz for the 4870 and 650/850MHz for the 4850.
The Germans reckon that we're on for a May release, with the top end card costing $349 and the midrange $269, with the lower of the three yet to be priced up.
You can have a look at the chart for yourself here, and you might be interested in a little English-language discussion here, which includes the stunning revelation that 'Mai' might just be German for 'May'. Really? µ
you might be interested in a little English-language discussion here, which includes the stunning revelation that 'Mai' might just be German for 'May'. Really?

LOL
Since it would take a total re-design of the GPU for ATi to use clock domains as Nvidia does, I'm gonna go ahead and say this is FUD.

See the discussion on this very subject @ beyond3d for reference - http://forum.beyond3d.com/showpost.php?p=1151416&postcount=1289. Maybe you guys should just get your news from there? 
Hai
According to the website, it is not the memory that runs at 1050 or 850 megahurtz, but them shaderz.. Speichertakt would be memory frequency, which read 1935 MHz for the 4870 GDDR5, 1728 MHz for the 4850 GDDR5 and 1143 MHz for the 4870 GDDR3.

Frankly, I've seen a lot of different memory configurations spun over the forums and they all seem a bit unrealistic. 1935 seems a lot higher than the 1100 current GDDR4 runs, while 1143 seems way too high for GDDR3 - current cards keep their GDDR3 mem closer to 900.
Yeah, I thought that memory frequency looked oddly low. 1050MHz using GDDR5?!? GDDR3 can go that high, why would you need GDDR5. That doesn't make any sense. The 1935 MHz (3870 MHz effective) makes more sense. I guess the article should be updated but theinquirer.net doesn't really do that kind of thing.
GDDR5 can hit 5 GHz...
Those 3870 days sure came and went pretty fast.
Do I believe Fudzilla or these Germans.

512bits or 256bits?

Or is it all FUD.
when will the 48x0 AGP be out?
Wasn't R700 supposed to be a modular system with multiple simple GPUs and shared memory? I was expecting RV710 to have 1 R700 unit, RV730 - 2 units, RV770 - 4 units, and R770 - 6 or 8 units. As Intel is making abundantly clear with Atom, smaller, more easily produced chips seem to be the best way to go in terms of cost and yields. The only difference between the different performance levels would be board layout, bus width, and memory controller. I also think that the significant increase in compatibility by ending single board crossfire would outweigh any latency issues created by having shared memory. Oh well, I guess standard policy is: "Talk big because they'll dream bigger. Release whatever is easiest since nobody has a long enough attention span to remember the promises. After all the delays, they'll be so thirsty that even the most rancid piss'll be sucked up in no time. And most importantly, report major bugs just moments after your biggest clients have installed your fancy new chips on every node of their most complex supercomputers."