AMD: R9xx Speculation

Are you thinking of atomics?

 

If it's that simple, I wonder why someone from ati isn't blogging about it then.
And is it driver caused?:
http://forums.amd.com/game/messageview.cfm?catid=260&threadid=134121

and can the mipmap slider make a difference, as if in testing on quality/performance instead of high quality?
http://forums.guru3d.com/showthread.php?t=105065

 

Yeah, sorry for not updating on this - the LOD values are incorrect and not the solution, adjusting the LOD 'till shimmering disappears makes the image for more blurred than nVidias image for example.
However, the value for nVidia to start shimmering is a LOT less, and that needs a bit more investigation (as in, which is sharper, nV shimmering, AMD shimmering)

 

With negative LOD, Nvidia always exhibited a stronger tendency towards shimmering, AFAIR, hence the LOD clamp switch in their drivers.

 

No. The comment said feature XY with datatype ZU is not used in games, that's why it won't be "usefull". With the OpenCL image extensions you get access to f.e. the "texture"-samplers, you can use them for all datatypes.
In the times of deferred engines you also have a higher variety of "textures", which are no textures but "2d"-arrays whos datatype is driven by the purpose.

Removing datatype-selection penalties (beyond memory-bandwidth) will make it less of a brainer just to fiddle around with the huge combinatorical space of possible formats vs. performance which will only apply circumstacial etc.

I don't want to impy ATI is planning f.e. a programmable blending extension for OpenCL, and it's was a speculation anyway, but I just wanted to hint at that advances for the big datatypes are probably not developed with games in mind. And earlier or later ROPs will also kneel before the unified programmable approach. ATI ROPs are already kind-of "programmable".

 

See MEM_RAT in the ISA:

This would be a 128-bit write to a 2D or 3D image buffer.

This would be an atomic.

 

UNIKA Radeon HD 6970 sold for 3999rmb ( around 605 $)

source: http://bbs.pczilla.net/thread-12652-1-1.html

 

That is 6850 or 6870 not 6970 (only 1 CrossFireX connector + Cayman has been longer in previous pics)

 

Is it me or that looks like smaller in length than the 5870?

 

The closer we get to launch the less speculation we have? What's going on?

 

Why can't I find the pic?

 

I think the moving launch date has spoiled some of the anticipation fun.

 

I think its that's the pics here:
http://translate.google.de/translat...163.com/digi/10/1202/16/6MTNAIP9001618J1.html

 

only one CF adaptor ?

 

HD6950 maybe?

EDIT - It IS the 6950. Look at the 6th photo in the article.

 

See the post above - the pictures are rather low-res, but it looks like it says "HD 6950" on the sticker ... maybe they'll deliberately ship it with only one CF connector to "protect" HD 6970 and keep quad-GPU-setups limited to the enthusiast market?

Maybe it's part of a different/new multi-GPU appraoch introduced with Cayman, too? If two Caymans on-a-board could communicate without the need of a CF-bridge, for example - wouldn't that basically render the hardware-support for a second CF-connect useless?

Or the pcitures are faked

 

Could be, but the pic is very low-res so its more like reading tea leaves... or my eyes are way to bad

 

Well, in one of the pictures I can clearly see "HD69" and on the other I see "50".

 

I think everyone's really fed up by now. Just get the damn thing out already! It was going to be a definite purchase for me, but now I'm wondering if I'm better spending/saving the money elsewhere. That's purely down to anticipation fatigue meaning I'm not sure I care enough any more.

 

I ran out of patience for sorting through bogus slides and rumor sites posting whatever they skimmed off of forum speculation in order to generate hits in the absence of having genuine insight.

As was already demonstrated, it is not really all that hard to fake a slide, but I am baffled as to why it seems much more popular to do as of late.

As far as speculative web site updates, half of their speculation is stuff that's been discussed to death on forums months prior, and much of whatever remains tends reveal more on how little the writers know of the subject than give us something to talk about.

I'm waiting for actual results, with confirmed data and confirmed context.
The SNR for this has sunk so low I can't continue on for fear that I'll go blind from all the eye-rolling.