The ATI Radeon 5XXX Thread

[***Deimos***]

No, it's not.

And SLI / CF is like running the framebuffers ( GPU Memory ) in RAID 1 ( Mirroring ), not RAID 0 ( Striping ).
Meaning, the effective bandwidth in dual-GPU cards is the same as a single-GPU card using the same core.

The 4870X2 has 115Gbps not 230Gbps memory bandwidth.

4870x2 = 5870
5870 is like two 4870 cores.
(if we don't agree on this, then we need to take it up with AMD presentation slides).

4870x2 is 750Mhz, and CF CPU overhead and inefficiency
5870 is 850Mhz + IPC efficiency improvements.

How else do you explain 4870x2 being faster by up to 45%?? [EDIT: typo... its 42%] Magic?
BTW: 230GB/s vs 153GB/s = 49% more - coincidence?

The only other explanation is that 5870 uses different code path than R7, and is so poorly hashed together. But, if that was the case, 5870 wouldn't be faster than 4870x2 in other games (where 850Mhz 1600 SP is more influencial than 153GB bandwidth).

[EDIT2:
I also noticed GDDR5 can't be overclocked past 1300Mhz (8%).. even though volt-mod allows 1000Mhz+ (20%+) on the GPU, the performance improves only slightly corresponding to slight memory speed bump - I'd personally wait until they have cards with Samsung 6Gb/s or even 7Gb/s chips or the inevitable 5890 that will be released to spoil GT300 launch]