Its more of a refresh than a new GPU arch imo. I bet the actual R700 have been pushed back due to various factors (cutting cost from R&D is one).
32TMUs mean both 32 TAU and TFU (unless its even more beefier). RV670/R600 both had 32 TAU and 16 TFU. Comparing this with G92 and 64 TMUs, nVIDIA still has a much more raw texturing power. But im a little skeptical on the current specs since the ALU:TEX ratio is down to 3:1 from 4:1 which doesn't make much sense seeing as ATi has been pushing the higher ALU:TEX ratio pretty aggressively compared to nVIDIA for quite awhile now.
So a total of 480 ALUs in grouped into 96 vec5 shaders (well not really vec5) which is a nice boost but im not so sure just how much improvement we will see with the increase in shader units seeing as RV670 is already much stronger when it comes to shaders compared to its competition G92/G80 (unless its the utilizations of the shaders that cut alot of potential to its theoretically stronger shader performance). Maybe there could be more efficent shader AA along with this, which will make AMD/ATi to finally target high end market once again using a single GPU card.
GDDR5 is quite new, and the starting memory clock for those are 1.6GHz i believe. (3200MHz effective!) Im sure these will cost quite abit (since they are fairly new), and im abit skeptical on the usage of GDDR4. Seems like GDDR4 has been "skipped" since it doesn't really bring much onto the table when keeping cost in perspective.
For reference, a 8800 Ultra (full fledged G80) is roughly 30~50% faster than the full fledged RV670. Currently i think the RV770 is going to be faster or on par with the 9800GTX. (Maybe even faster when AA/AF is applied due to having much higher bandwidth). Even though this may not be the actual R700 that ATi was working on, i think its a good move from AMD seeing as its a really competitive product in all the big categories i.e performance/price/watt/features.
- better 'n r600/670]
Facebook
Twitter