ATIs R870 surfaces

its quite hard to believe this:
http://www.nordichardware.com/news,7766.html

Probably a bit optimistic, but I wouldn't be surprised by a number thats close to it. The 3870 has 320, the 4870 is rumored to have 480.

http://www.tomshardware.com/news/a [...] ,5223.html

If they can double this, 960, then if you put two of these on a PCB, then you have nearly 2000 shaders on the R870. (even the article you posted said this would be for the x2.) The question is, can they double the number from 480 to ~1000? I'm not sure moving from 55nm to 40nm will give them the room they need for this. Not to mention the 4870 is supposed to do better then the 3870 because they increased the number of ROPs, etc. If they drastically increase the number of shaders, they won't have enough horsepower in the back to drive all of them.

It will be interesting to see. I'd rather keep paying attention to the 4870.

I read about it earlier from here http://www.pczilla.net/en/post/16.html where it also mentions a rv740 thats the first at 40nm

From Nordic Hardware "Although, there so much more to GDDR5 than just the higher frequencies, and even though the fastest chips of today operate at 4.5GHz, 5GHz is very much feasible. One of the new features is improved error detection mechanisms for less system margin and more stable systems. It also brings a new adaptive interface timing, which means there is no need for trace length matching, thus cheaper PCBs.

Last but not least is the data eye optimization, which includes such key features as data/address bit inversion and adjustable driver strengths, voltages and terminations. Best case scenario this will lead to faster interface tweaking, relaxed transmission lines and, again, cheaper PCBs. So in the end, even if the GDDR5 chips may be more expensive and require more traces, the PCB design will be simpler and overall cheaper. Add to that, that GDDR5 is many times faster than GDDR3 and can operate over a narrower bus and still reach the same actual bandwidth. " Thats why theres no 512 bus. I also had read up on the GDDR5 earlier, and it sounds good, Im hoping implementation is as good as the paper/showing

And itll be using GDDR5, at maybe 5Ghz. Maybe wont even need that 384 bus

doubling 480 wont give you 1000 or 2000

right , but its near. Dont get picky

Looks like the old ATI team became quite cautious with their memory bus architecture after the wasted 512 bit bus on the 2900 series.

Something tells me that they are not reaching for the high end crown but, again, for a solid high mid-range card.

Over 60% of the monies made is in that segment. I think the AIB partners will like it also

To be honest, i don't really care about ATI competing with Nvidia. I'm somewhat certain that NV will perform better.
The direction ATI is taking with the multi-chip approach is something i like though. If they manage to keep the chip size under control (maybe even shrink it with a refresh) and use GDDR5 to gain the benefit of better control over the bus traces on their PCBs, the x2 should be easy to manufacture. If they keep that up, i will start to hope for a x3 or so once the refresh looms on the horizon. And since that trend will continue, they better start to figure out how to have multiple chips share the same memory or at least parts of it (ie. texture memory).
I'm a little concerned regarding the power consumption and heat though. Then again, almost all new cards have been power hogs.

Theres talk of MCM for ATI, but then again, thatll cause more power usage. The G280 is too big to go X2, so it is what it is. Looking at the clocks on the ATI cards, theres lots of headroom, so thats good for ocing and of course power management.

As for the original rumour.

2000 on the RV888whatever...

Is that with Physics processors?

Considering all the 800SPU rumours about the RV770 just months ago, I wouldn't put too much faith in anything talking about over the horizon chips.

I read somewhere that GTX 280 with Crysis on 19x12 with AA was "no problem" for the card.

Thats the thing, I dont think we need that much power yet, software still needs to catch up to hardware. Even Crysis can be played with a GTS G92 at very high with some AA at 12x10 with like 25~ frames/s. Surprisingly, that runs smooth.

I hear rumors that the 4850 reached 990 MHz on air. If that is true then that's a hell of a good overclock, and also makes possible we can see some 4850/4870 XTX with even higher clocks later this year or on next update (which could be 40nm)

They already have.

The 4870 x2 already uses shared memory.

Actually it probably will, since GDDR5 allows this as outline in Qimonda's whitepaper.

So I wouldn't be surprised if we see just that in the HD4870X2 or R700.

HD4870X2 is R700