ATI H2 2010 "New Generation Graphics Processors" [Xbitlabs]
 

Xbitlabs link

Basically saying that ATI (AMD) is planning new/updated graphics for H2 of 2010 (more than likely because of Nvidia's upcoming new cards). They are unsure of if it is going to be another round of new cards, or tweaks to current cards, or possibly a die shrink.

Anyone have any insider info on more details than the link provides?

Any thoughts on what is to come?

Enjoy.

I'm praying that Fermi isn't a big fat flunk. AMD needs to be prodded by competition or else, we consumers get shafted.

"H2" can be all the way into December 2010.

Anything before that would be idiotic on AMD's part since GF will only be able to produce their 28nm process in volume within Q4 2010. Any 28nm production in Q3 is what they consider "high risk" which isn't conducive to anything resembling high yields.

If AMD designs the chip on 40nm and chooses to forgo 28nm for the time being, they are more scared of GF100 than any of us think they are.

All that I have heard is that they are named after the Southern and Northern islands (don't know what those are). I believe the older rumours said they were LOTS of small SP's/cores, as one was named after something with 10,000 arms.

How about making ATI gpus that can actually fold on par with nVidia ones? AMD is missing out on market which is us folders. It would be great to have more options for diferent price ranges and ppd other then the ones we have to chose with with nVidia.

It's coming, it's coming. GPU3 is just around the corner.

IMO, even GPU3 won't hold a huge performance increase for ATI cards. It all comes down to the way ATI's memory controller over-taxes the framebuffer with constant reads and writes due to ATI not implementing nearly enough cache inside of their chips. Hopefully their next-gen chips will take GPGPU apps into a bit more consideration.

The problem isn't the GPUs, it's the fact that the GPU cores are not designed properly to take advantage of them. And I doubt that will change any time soon, since ATI's degree of involvement with the project has not changed.
Don't get your hopes up.
Actually, one of the problems with the current cores is that they don't make use of the local data stores in RV770 and RV870 AT ALL, which leads to a lot of repeat calculations being performed. Right now it doesn't matter how much cache the SPs have if it isn't being used altogether. And I'm not convinced that the size of the LDS will even be a bottleneck should it start to be used. The only way to make that assessment would be to actually test it out and see what the performance results are.

My understanding is that the ATI cores do not allow for proper error correction, leading to calculations having to be done over. This won't be changing with GPU3 it seems, as Zero has mentioned.

I am curious as to what ATI's approach to GPGPU will be, if any. ATI has usually been about delivering good FPS in games but not much more besides that. I wonder if they implement GPGPU heavily the die complexity and cost will go up....