http://www.xtremesystems.org/forums/...postcount=2307
I really wish I could share the remaining slides, but I've been told not to. My source is worth too much to me to share these.
Maybe I can try and shed some light on things in another way. I guess that if my word isn't good enough, W1zzard's is. He said from an early point that there would be 160 shader clusters. RV670 has 64 shader cluster. Each RV670 shader cluster is 4+1 (5D). It makes little to no sense for AMD to change the cluster design to 2+1 (3D), which would then result in only 480 shaders. There has never been any talk about 96 clusters, other than speculations.
My Asian sources has claimed 800 SPs from the very start, but they are off a bit too often for me to believe them without doubt. Especially when western sources close to AMD report 480SPs. Later I find out that AMD has planted a number of rumors, including 480SPs to create a smoke curtain, and that it wanted to surprise both buyers and especially NVIDIA by launching a card with 800SPs.
GPU-Z is not proof of anything. It's just a database with information and if the database is outdated, the information displayed by the program will be wrong. Kudos to W1zzard for creating it though, it's a great tool. It's just not 100% reliable for cards that haven't been launched.
I've posted slides, Asian websites have posted different slides from AMD presentations that all show 800SPs, the only contradictory stories I see are the ones using GPU-Z as a source of information, which is just misleading.
-Natürlich kann es sein, dass hier der Vram Bug voll zum tragen kommt bei der 9800GTX -allerdings wohl eher nur in Crysis und weniger in den andren games, wo die 4870 dennoch 40-50% schneller rechnet 
