New rumor about ATI Southern Islands

[tajoh111]

Oh really!?!?!?
Overclocking both the core and memory by 9% brings a 7-8% increase in performance, a few cases ~5% but more cases than that +10%.
Core clock increase of 9% brings usually ~4-5% performance increases or roughly half the clockspeed increase.

While GTX480 with ~18.5% increase to the core/shader clock usually brings a 5-8% increase in performance, with a few cases being ~10%. So at best roughly half the increase to the clocks but in most cases about 1/3.

http://www.overclock3d.net/reviews/g...tx480_review/8 (used the average FPS numbers)
http://www.firingsquad.com/hardware/...ng/default.asp

http://www.guru3d.com/article/kfa2-g...rchy-review/19

I don't know man.. in this review a 8.5% OC provides for the most part a 10% or greater speed boost and at the bare minimum a 5% in only a couple games.

And if you look at hwbot, every single, single card 3dmark record is owned by a gtx 480(I guess not 2001 which requires optimization rather than GPU power)

3dmark 06, 3dmark 05, 3dmark 03 and vantage. (the gtx 480 always being first and the top place a 5870 has achieved)

--------480" "5870" "Slowest gtx 480 beating the top 5870
3dmark05 1340clk(58738) 1350clk(48619) 88th place 701mhz
3dmark06 1360clk(42921) 1390clk(36040) 84th place 800mhz
3dmark03 1340clk(157095) 1400clk(115853) 93rd place 750mhz
vantage 1450clk(39281) 1505clk(31004) 47th place 1100mhz

To give you an idea of the lead, vantage is the best benchmark for the 5870 and it doesn't enter the ranking until 47th(the other benchmarks the 5870 shows up at 80th+ place) place and the card is overclocked to 1500+ mhz which is an exceptional overclock for the 5870. It is getting beat by the gtx 480 at 1100mhz.

If a 18.5 percent improvement only gave a 5-8% increase, I don't think this would be possible, especially when in most benchmarks, a gtx 480 with clocks of less than 1 ghz are beating 5870 at 1400mhz.

Ask an bencher and they will tell you that a gtx 480 is way better than a 5870 for single card benchmarks, not really for clocks but scaling with clocks.