AMD Radeon HD 6950 Can Be Unlocked To HD 6970

An anonymous reader writes "AMD's new Radeon HD 6950 can be unlocked to a HD 6970 via BIOS mod. Performance of the unlocked card is identical to the full blown HD 6970!"

It is still different HW

[BadAnalogyGuy]

A lot of manufacturers will do this, actually. Their first device will contain very high quality, standard HW that is somewhat overspec for what they intend, but due to driver support and ease of implementation they can get it out the door in a reasonable amount of time. Then for their successor device they will take the lessons learned, use cheaper parts, use better optimized software, and sell it as the "cheaper" version.

You are getting lousier HW, but arguably better SW, so the performance gap isn't as big as their marketing lit will let on. On paper, the expensive first gen device looks better, but when the rubber hits the anus it's pretty much a wash.

Overclocking guide

[nicholas22]

An overclocking guide can be found here [techpowerup.com]. You *might* get problems under extreme load, because the 6950 uses the 6-pin power connector, whereas the 6970 can draw more power, because it uses the 9-pin connector.

Re:It is still different HW

[SuricouRaven]

Or perhaps they turned it off because, while it works almost all the time, it'll fail one in ten million floating point operations at random, or is prone to fail at moderatly high temperatures or workloads. If you want to use the 'disabled' core, I suggest you run your own tests to determine if there is some minor fault. Slow the fans so it runs hot and calculate pi. If it can run for 24 hours and produce the right result, it's probably good.

9500/9700 all over again

[lostmongoose]

same thing was possible with the 9550 Pro -> 9700 and with the 9700 -> 9700 Pro both were done with BIOS flashing

It's about manufacturing yield...

[EmagGeek]

So if I am a graphics chip manufacturer, I know that the fewer unique designs I have, the cheaper it will be to manufacture my product line. If I make both chips and boards, the same economy of scale applies to both the chips themselves and the assembled boards.

If I can determine both my chip and board yield at in-circuit test, and configure each manufactured device to its maximum possible stable capability, then my manufacturing product yield is maximized.

This type of yield binning is nothing new.

Re:Overclocking guide

[Elledan]

That's not correct. The 6 and 8-pin PCIe connectors are identical. They have the same number of ground and 12V wires between the GPU and the PSU, the same wire gauge and can carry the same amount of power. The 8-pin connector exists because in the PCIe spec they had a sense wire for the 12V line specified on this connector, which would then allow the connector to carry more current as the PSU would be able to better regulate the voltage. In practice this is much more easily done at the PSU side, making the 8-pin connector useless and allows the 6-pin connector to carry the same 150 Watt as the 8-pin one.

Want to check this? Just use a 6-pin connector and short the remaining two pins on the GPU to ground to satisfy the GPU if it checks for a connection there and everything will work just peachy fine. If you check 8-pin PCIe connectors you'll see that this is all they do: short the two extra pins to ground.