PLEASE SEE UPDATE 22 AND 23, IMPORTANT
Flair changed to NEWS because this is now NEWS
PC Perspective has confirmed most of what was said in this thread, thanks to all who helped
It has come to my attention that some people are erroneously interpreting the PCI-E specification document, claiming absurd things like the motherboard being determining the maximum power draw. Claiming that >300W is possible from the slot.
The 300W figure is broken down as follows; 75W(slot) + 75W (6pin) + 150W (8pin)
The reference to >300W configurations considers the following;
8pin + 8pin + 6pin
It is patently untrue that you can draw more than ~75W within spec;
Someone kindly sent me the link to the electromechanical specifications for PCI-E
Page 44 onwards
Information in this thread is interpreted incorrectly https://www.reddit.com/r/Amd/comments/4qmlep/rx_480_powergate_problem_has_a_solution/
I believe this will be the last update.
THE PCI-E SPEC I LINKED IN THE PREVIOUS UPDATE SPECIFIES
12V @ 5.5A WITH MAX 8% TOLERANCE
3.3 @ 3.0A WITH MAX 9% TOLERANCE
12V MAX = 71.3W
3.3V MAX = 10.8W
According to Tom's Hardware:
RX480 GAMING (metro 4k) AVERAGE is 82W on 12V, 4W on 3.3V for a total of 86W
The PCI Express® Base Specification Revision 3.0 is the wrong spec.
The power usage etc. is defined in the PCI Express Card Electromechanical Specification. Please look at page 27 and 36 of the PCI Express® Base Specification Revision 3.0
From Page 27 "Document Organization"
The PCI Express Base Specification contains the technical details of the architecture, protocol, Link Layer, Physical Layer, and software interface. The PCI Express Base Specification is applicable to all variants of PCI Express.
The PCI Express Card Electromechanical Specification focuses on information necessary to implementing an evolutionary strategy with the PCI desktop/server mechanicals as well as electricals. The mechanical chapters of the specification contain a definition of evolutionary PCI Express card edge connectors while the electrical chapters cover auxiliary signals, power delivery, and the adapter interconnect electrical budget.
Page 36 states the Reference Documents and the PCI Express Card Electromechanical Specification
And from the PCI Express™ Card Electromechanical Specification Rev. 1.1 (2.0 is behind a paywall, but wikipedia has the same information):
A standard height x16 add-in card intended for server I/O applications must limit its power dissipation to 25 W. A standard height x16 add-in card intended for graphics applications must, at initial power-up, not exceed 25 W of power dissipation, until configured as a high power device, at which time it must not exceed 75 W of power dissipation. Refer to Chapter 6 of the PCI Express Base Specification, Revision 1.1 for information on the power configuration mechanism.
Considering the maximum tolerances allowed (+8% voltage), taken from PCI-E electromechanical specifications, the RX480 is drawing, ON AVERAGE - NOT PEAK, 14% more power than the MAXIMUM from the 12v rail (82W vs 72W) and 5% more than the MAXIMUM in TOTAL (86W vs 82W).
The PCI-SIG may not have to. My preliminary research indicates that licensees of a trademark, even non-exclusive licensees, have standing both to sue for trademark infringement and to request the ITC/CBP block the import of infringing (counterfeit) goods. See the answer to question 5 in this link. If that's true, NVidia can request that the US government block the import of all AMD products that violate the terms of the PCI Express licensing agreement.
Disclaimer: I'm not a trademark attorney, so this could be incorrect. I am not providing legal advice to a client. No attorney-client relationship is created by this communication.
Those peaks also violate the PCI Express spec, if they are not a measurement artifact. The card isn't allowed to change it's current draw faster than 0.1A per microsecond: that's the "maximum current slew rate" part of the spec image posted above. A 1A spike for 0.5 microsecond is 5A per microsecond: 50 times the max allowed current skew and seriously no bueno.
Update 25; PCPer confirms!
From PC Perspective:
The easy fix to this whole ordeal is for AMD to have used an 8-pin power connector on the RX 480. The PCI Express spec allows an 8-pin connection to draw 150 watts on its own, leaving the power from the PCI Express connection on the motherboard for overhead. This is how NVIDIA designed the GTX 970 (two 6-pin connections) and how AMD designed the R9 380 (one 8-pin connection): both cards have 150+ watt TDPs with power supply overhead available to them. The new GeForce GTX 1070 has an identical TDP to the RX 480 (150 watts) but uses an 8-pin power connection to relieve any concerns at stock performance or while overclocking.
I asked around our friends in the motherboard business for some feedback on this issue - is it something that users should be concerned about or are modern day motherboards built to handle this type of variance? One vendor told me directly that while spikes as high as 95 watts of power draw through the PCIE connection are tolerated without issue, sustained power draw at that kind of level would likely cause damage. The pins and connectors are the most likely failure points - he didn’t seem concerned about the traces on the board as they had enough copper in the power plane to withstand the current.
AMD probably didn’t want to include a second 6-pin or upgrade to an 8-pin connection because of the impression it would give for a mainstream gaming card. Having two power connectors or an 8-pin might tell uninformed buyers that the RX 480 isn’t power efficient enough, that it might not work with an underpowered system and power supply, etc. My hope is that AMD’s partners paid attention to this data and are over building their power delivery to alleviate any concerns. We will know very soon.