AMD's Linux Driver Will Likely See A Power Change For The Radeon RX 480 Too

[devius]

This test of undervolting an RX 480 on Windows actually improved the performance by up to 4%.

I'm surprised no one here seems to have noticed this. It seems like AMD sent out the first batch of cards with unnecessarily high voltage, maybe to make sure that they were stable no matter what, and now they simply tune them better, resulting in lower power consumption and better performance. End of drama.

[L_A_G]

Umm, no. You should do some more research (beyond reddit) before accusing me of being a fanboy and spreading FUD, when all I did was discount your FUD-spreading.

Really, telling people that a graphics card drawing a bit more than 75W isn't going to kill any motherboards because there's been plenty of graphics cards that have done the exact same thing without causing any problems is FUD or are you referring to me pointing out that plenty of other cards have done the same thing? Because this is how the Asus GTX 960 Strix draws power over the PCIe bus:

https://s31.postimg.org/it9b3hs23/960_Strix.png

[oooverclocker]

I'm not very knowledgeable on the subject, but many places I've seen talk about this mention that the PCI 2.0 and 3.0 specs allow for more watts to be pulled per slot (100w with PCI 2.0, and more with 3.0, right?)

Based on the PCI Express Card Electromechanical Specification Rev. 3.0 I can't confirm that. It sais the card is allowed to draw 5.5A max. from the 12V- pins where the voltage can vary +/- 8%. So in total it can reach 71.28W. There is also a 3.3V trace that can be strained by 3A with a voltage tolerance of 9% which means 10.791W max. So we have 82.071W. But the specification sais that the limit is exactly 75W which should be split between the 3.3V and the 12V traces. Which indeed doesn't make much sense as the traces are designed for at least 82.071W.

So as long as you're not overclocking the card an extreme amount, if it's in a PCI-E 2.0+ slot it shouldn't damage the motherboard, no? Even cheaper ones.

Exactly.

[bridgman]

Right... my understanding is that newer versions of PCIE use more pins to carry +12V, however the spec has not been updated to reflect that.

[eydee]

8-pin variants on the way: http://imgur.com/a/IpRiI

[bridgman]

BTW the primary power change for Polaris 10 is now in Linus's 4.7 tree:

https://lists.freedesktop.org/archives/dri-devel/2016-July/113203.html

https://lists.freedesktop.org/archives/dri-devel/2016-July/113221.html

https://git.kernel.org/cgit/linux/ke...dc9bb9c22b5537

This is the change that shifts a bit of power draw from the slot to the power connector.