0
The reference 980 TI, which I have along with an EVGA ACX 2.0 cooler, has 300W worth of power available to it on paper. It has the slot, a 6-pin and an 8-pin. The Gigabyte G1 and upcoming Asus Strix DCU III have 375W because they have two 8-pins. Moreover, the reference 980TI might well be capable of taking more than 300, because the 6-pin slot appears to have a trace for the 3rd yellow power wire (to elaborate: the two extra 8-pin wires are just grounds, whereas the 6-pin is supposed to be a 5-pin, but PSUs make them 6+2 so the 6-pin is an 8-pin, just with the grounds removed).
Despite this, the 980TI has a 250W TDP, and you're not allowed to raise that past 275W (+10%). This hampers OCs like crazy. My card is able to stay running, at stock voltage, at a max boost of about 1460mhz. The problem is that the card slams into the power limiter in games, 3dmark, etc, and settles into the low-mid 1300s. Increasing voltage even a smidge stabilizes 1500mhz or more (1500@1.255 is rock solid, can't drop voltage and haven't pushed clocks more to see the limit), but adding voltage bumps against the glass ceiling of the power limit even faster, and the card actually performs worse. I was dropping below 1300 in games and benchmarks, and hit an optimal point by lowering the clock increase and keeping voltage to its minimum. I suspect the card loses performance even with higher clock increases@stockV because it shoots up high, smacks into the power limit, then has to drop savagely to compensate, so a middling but more stably present OC did better.
3dmark Firestrike seems to underestimate clock differences vs games a bit, but bears this out.
980 Ti SC: 13802
980 Ti SC +100: 14225
980 Ti SC + 152@1.25: 14490
980 Ti SC + 120: 14593
There's almost that much variation between runs anyways if you just run Firestrike whenever, but each test was run after a fresh restart to eliminate that effect, and results held, and the power limit was visibly being hit in all circumstances.
Increasing the power limit with a modified bios yielded promising results. I used Sheyster's "max air" bios from here (warning on nvflash: RIP headphone users), which bottoms the voltage out at 1.255 and raises the power limit to 350W, or 425 if you want to get bold... not sure I want to do that to my PSU or a reference card. This certainly exceeds the spec on the PCIE connectors, but people seem to be having no problem, even with reference boards. I'm powering mine with a Seasonic G-750 and having no issues tripping the OCP or anything at 350W. Again, I suspect the reference board is wired to take considerably more current than it's supposed to on paper, perhaps because that's required by Maxwell's wild power draw oscillations.
The result of lifting the power limiter was instant. Instead of bottoming out, at best, at 1360, the card holds at 1501@1.255 without the slightest bit of variation, and the voltage stays perfectly steady according to GPUZ, so provided it's being read correctly, power is getting there okay. I'm not even hitting more than 90% of the tdp limit, so this card has headroom to spare. With good enough cooling I bet a stable 1550 or even 1600 would be possible.
I've gotten 3dmark up to 14940 (a disappointing rise, but again, I think it's underestimating), may push a little harder to hit that 15k, but either way, I'm quite happy, and although a liquid cooler is on the wishlist for a month or two down the road, the ACX 2.0 cooler and a slightly aggressive fan profile is keeping things below 75C.
So there's my rant about nothing in particular for the day. modded bioses are out there for the Titan X, 970 and 980 as well, and thus far people have been having great success it sounds like, with no reported fried cards as yet. It's definitely worth looking at for OCers, imo. Now excuse me while I go back to being bored.
Edited by Catamount, 04 July 2015 - 08:51 AM.
