borandi
As with the 4890, power consumption is a big issue, especially when using three of four cards. Word is the 5870 is 170W, so the 90 variant will hopefully be 190W or less.
TDP of 5870 is 188W. 5850 is 170W. Considering 5850 has 10% of it's SIMD and TMUs disabled, and has a 10% less TDP, which may or may not actually be accurate in real power consumption, I'm just assuming something to do with 40nm leakage is requiring voltage to be similar. Not unlike 4770 which was capable of high clocks, but had a very high initial TDP for what it was.
Supporting this, it's been reported 1ghz on the stock cooler is very possible…like 4890…so it would seem 1ghz can be done at less than 225W. I'm assuming of course a 5890 would be clocked so it achieve over 3TF (938mhz), at least the percentage higher 4890 was (13.3%…964mhz), and 1ghz just looks pretty on the box.
That being said, consider these things:
1. The PCB power connectors on 5870 have the solder points and room for an 8-pin connector in place of one 6-pin. The 4890 has this as well, and was used in OC models that were clocked up to 1ghz.
2. 5870 uses 4-phase power (5850 3-phase) with room for one more. 4870 used 3, 4890 used 5.
Does indeed look like history will repeat, in a newly tiered fashion. If I had to guess, it will launch at 5870's price when 5870/5850 settle (~$200/300). There's also the possibility it will launch at a better price to beat it's closest GTX300 rival in price/performance upon their launch. That would be something to see…ATi spoiling an nvidia launch! :embarrassed:
Whatever be the case, I expect whatever the stock clock is on 5890, it will also be on the 32nm XT refresh…or darn close. Seems ATi has found themselves a tick-tock strategy of sorts!
1. Release moderately-clocked and harvested parts on initial process yield (4870/4850…5870/5850)
2. Release parts towards highest feasible clock on matured process yield (5890…or the refined 55nm rv790/4890). OC part with 8-pin connector an idea of 90 part on next-gen.
3. Start again on next process using **70/**90 clocks as baseline clocks.
Damn, guess I'll have to wait until 32nm to see if we have 800/1000/1200mhz parts and that actually happens. Bet it does. That's a pretty smart way to both get to market quickly and complete a product stack, IMHO.
