Sunday, December 26th 2010
AMD Radeon HD 6950 Can be Unlocked to HD 6970
Looks like Santa brought an extra present for us hardware enthusiasts this year. Thanks to a less secure locking method AMD's new Radeon HD 6950 can be unlocked to a full blown HD 6970 with a few mouse clicks.
As detailed in our article, you can safely perform the flashing process from within Windows. In case something goes wrong it is easier than ever to recover the card thanks to AMD's new Dual-BIOS feature.
We tested the unlock on three HD 6950 cards: one AMD engineering sample, one HIS media sample and one ASUS retail card. All of them unlocked perfectly and run at HD 6970 speeds now. More success reports are compiled into a table at the end of the modding article.
As detailed in our article, you can safely perform the flashing process from within Windows. In case something goes wrong it is easier than ever to recover the card thanks to AMD's new Dual-BIOS feature.
We tested the unlock on three HD 6950 cards: one AMD engineering sample, one HIS media sample and one ASUS retail card. All of them unlocked perfectly and run at HD 6970 speeds now. More success reports are compiled into a table at the end of the modding article.
302 Comments on AMD Radeon HD 6950 Can be Unlocked to HD 6970
I can see what you're thinking, but AMD just killed off the 6970 and I'm sure they won't be happy about this
This is freak'n awesome:)
1. Cayman yields were great, and most dies qualified to have all shaders enabled.
2. Cayman yields were terrible, and units were disabled on both parts. 6950 BECAME 6970 (Or at least shader specs drifted downward; perhaps revisioned more than once) allowing greater allocation to 6970.
The mill, and reality, points to number 2.
Our Chinese/Taiwanese friends supposedly 'in-the-know' mentioned long before the 6900 series release that it had (according to Googlnese) been "melted down" and changes had been made reducing shader count (From 1920 to 1792 IIRC, but I may remember wrong). I don't know exactly what was meant by that, but there was likely something to it. This is on top of reporting from a different source Cayman was taped out at least twice, the last time a pretty short time before launch (late Q3). TechEYE also ran a post (from 'inside-Taiwan sources') on Nov 7th indicating yields were in single digits and AMD was in a scurry to get SOMETHING out to compete because enough fully-functioning dies were not going to make it for a Q4 launch. So there's the hear-say.
Then there's facts. Not only was Cayman delayed, reinforcing yield rumors, but when you start looking at performance for 6950/6970 relative to Barts things get ugly. I don't know how any other conclusion could be reached other than this compromise satiated their need for yield, performance above Barts, if only barely, and the realization these two products were the only two they would be able to harvest dies for. It's hard to believe 6950 was planned to be only 10% faster than 6870, and the two 6900-series products 10% apart. We all know the standard gap between products (~15%), and this ain't that on multiple levels. Also note not only is 6970 the first part in eons from ATi with an unorthodox clockspeed (They HAD to release something comparable to GTX570, yields be damned) but that this core was architected after they knew their direct competition (512sp Fermi, ~750mhz). When a mid-range GTX560 (upclocked, mind you, to unexpectedly compete with 6950) is coming that will destroy not only Cayman's intended MSRP, but put pressure on AMD's whole $200-400 stack when AMD typically prides themselves on efficiency, you know something had to have gone terribly wrong.
If anything, the flashing ability would likely indicate 6950 was not planned to have 1408 shaders from the get-go, or the dies WOULD be locked down from TSMC. What if 6950 was planned to have 1536sp? To me, and I may stand alone, that makes a TON more sense.
The plot thickens...
300 $ for 48 more shaders and 1 tess unit vs gtx 460 is quite not worth it in my opinion.The increase in shaders is the same with that of the 9800gtx vs 8800gt (~14%)wich u can easily compensate with overclock.
But 14% moree shaders doesn't mean the performance increase is linear in reality the shaders barely make a difference but the clocks affect performance more.
So now here is my problem do i buy 2 6950 or 2 6970 :banghead:
Now this changes the bang-for-buck, big time. If I've done my sums correctly, this now puts the 6950 as better bang for buck than a 6870 as well.
(I used this to calculate)
I think next year's 28nm shrink will be very 'soap opera'.
Can't wait!
There was glitch that ended:
At the Physics first test in 3dmark11 i ended up with black screen and the mouse cursor and freezed there.The pc didn't freezed completly so pressing space throw me back to desktop and i got an error from 3dm11 saying something unexpectly occured..
I raised the powertune glider to +20% and after this i run again two times the 3dmark11 bench and everything was fine..
From many bechies i saw that 6970 overclocked further gives very little performance improvement..With this unlocked card i hited the sweetest deal..
Does this increase/decrease overclocking potential at all?
. . . . cause it looks that way
Now the strange thing is that i have no way to change voltages in CCC or in Trixx...