Quote Originally Posted by dungeon View Post
GCN hardware is newer architecture and use different driver, maybe that hardware architecture is just slower for these tests .

But when we look closely only Radeon 6870 performs strange yes strange , it perfoms very good but why (so) better then Radeon 6950 And if we exclude that particular card Radeon 6870, results are very scalable and GCN hardware performs better .
Because the HD6900 series cards use the VLIW4 architecture while all the rest of the HD5/6 and low end parts of 7/8 use VLIW5. The VLIW4 architecture is very similar to that of VLIW5, which is why it got supported so quickly, however it is nowhere near as optimized as only the HD6900 cards had it and as such they are of a much lower priority and the small public testing pool for them doesn't help things. Theres maybe a few hundred thousand(wild guess) HD6900 cards in existence, while theres actual millions of VLIW5 based GPUs out there. They are even in all APUs that are pre Kaveri and Kabini as those series added GCN based GPUs.

The fastest VLIW5 cards are the HD5870 and HD6870, both perform very closely, the HD5870 having an edge in double precision OpenCL and having had an Eyefinity 6 edition w/ 6x mini Display Port outs and double the vRam, while the HD6870 has better tessellation support, else they are actually almost identical in most benchmarks.