Re: AMD RX 480 is a defective product! Period.
Of course software can fix it. The power limits for all of the internal supply rails inside the card are controlled by firmware and the drivers on existing cards; there's no reason to think the 480 is any different.
One of my PCs has a Nvidia Kepler GPU, and it resisted my attempts to overclock it using tools like MSI Afterburner. I increased the overall TDP limit with Afterburner, but it was still scaling back its boost clock under heavy loads, with the power-limit (TDP) bit set.
Once I edited the card's firmware in Kepler BIOS editor or whatever it was called, I could see that I could adjust the power limits for each of the card's three power sources (each its own internal power rail)-- the PCIE slot, the 6-pin power connector, and the other 6-pin power connector.
By tweaking the individual power limits that controlled the draw from the 6-pin connectors and increasing the range of the overall TDP target (the max it can be set to in Afterburner) from 110% to 120%, I was then able to maintain about one step higher in boost clock under heavy loads than I had ever seen before under any amount of load (and never under heavy load; it would always throttle back about 2 steps from its maximum when under heavy load), and by making a slight tweak to the boost clock in Afterburner, I got it to go one more step up in boost clock under load, for a total of some 4 steps under heavy load.
All of that was done by altering the BIOS.
One thing I could not do is increase the voltage to allow even higher overclocks. The peak overvoltage on my card was all of 10mv-- trivially tiny, and because it was set in the driver, there was no way to use the BIOS tweaker to raise the limit. Not effectively, anyway; the BIOS has its own internal limit, but no matter what was done with it, the driver would prevent the overvoltage from exceeding 10mv.
BIOS would seem to be the proper place for AMD to fix the 480, but there is really no way for me to know whether it uses a system like my Kepler card or something quite different. Anything the BIOS can do can also be done by the driver, and the enforcement of power limits most certainly is done in BIOS in some (perhaps all) modern GPUs.
Now, it is possible that there is some engineering defect in the power control hardware on the card, causing it to disregard power limitations set by the driver or firmware, but that has not been established. Given that some media outlets have reported performance increases by undervolting the 480, it is quite possible that this is nothing more than a driver or firmware defect, and that the card will emerge from all of this as a solid product.
Or it could fall flat on its face.
We won't know until we see what happens. I'm inclined to believe it's the former; this just has the feel of a firmware bug, and while the driver fix could be a useful Band-Aid to prevent any more motherboards from burning up, it will probably take a firmware revision to fix it permanently. I don't really know, though; it's possible that all of this is functionality built into the driver too.
Biting the hand that feeds IT
