per GPU

I did not Undervoltage as I don't pay for power on my workstation :-) But I can only mine 16h out of 24h :-(

don't confuse people.
I've got here 6 rx400 on 1KW doing just fine (the most intensive dual eth&sc takes up to 920W from the wall, ZEC is about 750W)
EDIT: undervolted to the very bottom ('bout 850mv), lowered core, 2K mem, bios mod

The problem can easily be related to the pcie lanes/resources availability, play with bios: disable devices, set lower generation for the slots, disable power savings, etc... Drivers setup in win can be an issue as well.

Thank you for the answer. I didn't know about that. Asked you in thread 3-4 times about that.

there r no extra cores to unlock on 290x/390x

1150/1900 obviously custom bios undervolted etc. Still pulls 890 at the wall, but the point of my post is as long as you know how to balance rails and tune everything right there is no reason you can't run a PSU at its max capacity.

mine is the powercolor first reference with 4 gig. Can I still do it?

Hey Claymore and gang.  I've got a funky one for you...

(tl:dr? Go to the bottom for update.)

So I just cobbled together this 4 card rig.  Previously I had all these parts running in a case with 2 7950 cards, one Sapphire and one XFX.  I put together an open-air rig and risers using the same hardware that was in the case (MB, Memory, proc), and added a pair of MSI 7950's, for a total of 4 cards.  I daisy chained a 500W to the 1000W power supply and have the 500W powering the MB, three 120mm fans to move air across the cards, a 1TB conventional HD, and 1 card.  The 1000W is running three cards.  I am not at my rig at the moment so I don't know which card is running off the 500W supply (this I admit might become important).  Fresh Win 7 install, 15.12 drivers, GPU-Z and HWMonitor, Teamviewer and nothing else.  After fighting with Windows and finally getting everything talking to each other I ran some spot tests for power draw and stability.  Everything looked good.  Started v11.0 miner and ran it for about an hour this morning and found that with my previous settings I used in v9.2 I was getting around 225 from the Sapphire, 220 to 230 from the MSI twins, and about 215 to 220 from the XFX.  I also discovered that one of the fans was vibrating about its shaft and generally being a pain in the ass.  I removed it and attempted to lube it but that wasn't its problem.  That being said, it still provided adequate cooling as indicated by GPU-Z telemetry.

About 15 minutes before I left to go to work I fired it all up and put it to work, walking out the door around 1pm.  According to the logs, temps across all the cards were steady around 62C to 65C (-tt set to 65) with fans ranging from 55% for the XFX to 40% (the baseline min) for the Sapphire.  About 25 minutes into the run the logs show the fans on the XFX ramping up to around 75%, but the temps were steady around 65C.  Seems the fan is starting to act up again.  Anyway, shortly after this I see the GPU temp fall rapidly and about 3 minutes later the watchdog found it unresponsive and commanded a restart. 

When it came back up it seemed to initialize fine but now it's running about 100 sols/s slower than the other three cards, at around 105 to 125 sols/s.  Looking at GPU-Z, the thermal and voltage readings are inline with a card at 100% duty cycle just like its brothers, but it's doing half the work.  Logs don't show anything extraordinarily funky except for one thing: when the watchdog ran a minute after the restart it showed:

"warning: solutions buf overflow, 1928276 > 24"

In the log prior to this was the speed check which showed the XFX running at 120 sols/s.

A few observations I hope might help.

I am not at the rig.  I am elsewhere monitoring it, so I apologize for not giving log snippets or screenshots.

All cards have stock BIOS.

All cards are normalized at 950/1350.  The XFX is based at 800/1250 but has been running at 950/1350 for months with zero issues.

All voltages are stock.

All cards are on risers (USB/powered).

The XFX card's riser is in the 16X slot and is driving the monitor.

The XFX is rock solid and happily hashing out 125 to 135 sols/s with no apparent failures. Telemetry suggests it's working as hard as the other cards.

I am not getting any rejected shares, neither prior to nor after the incident.

I can neither confirm nor deny that the XFX is the sole card connected to the 500W supply at the moment (aforementioned proximity to rig).

The 500W supply was not in the system when it was in the case running two cards. I removed it and installed a 1000W supply when I added the second card.

The 1000W supply is a single 12V rail supply.

The card hashes about 435 h/s using Claymore's Cryptonote 9.7 miner, while its brothers push out around 480 h/s. Oddly, the XFX is doing this using 37% GPU load, where the others are at 68%. O.o  Telemetry suggests this is true.

The card is deficient running v9.2 of the Zec miner.  It didn't have this problem with the same settings prior to transplant.  All cards are hashing around 180 sols/s except for the XFX (125 sols/s)

The tests show you are not the father.

So, whadda ya brains think? 

EDIT: Well, it looks like this card is lagging across the board, so it's certainly a hardware problem, and I'm suspicious that it's a power problem.  <s>To that end, how much power do these kind of cards pull from the motherboard?</s>

EDIT 2: Nevermind. 75W for PCIex16, 25W for PCIeX1.  I'm assuming that all cards are pulling 25 watts out of the board, thus 100W I didn't account for coming out of the 500W supply.  Hmmm...

 Keep in mind that the top-end 3xx (and 2xx) cards have 384 bit (280x/380x not sure on the 280/380 non-X) or even 512 bit (290/390 AND UP) wide memory access, which makes their somewhat slower memory clocks give significantly more memory bandwidth than the 256 bit wide memory on the RX 470/480.
 This isn't as critical on ZEC which doesn't appear to be as memory-hard as ETH, but it still matters.

 Core clock speeds on the 4xx also didn't increase all that much over the high-end 2xx/3xx series cards - AMD seems to have aimed more at dropping power consumption with VERY good success on these MAINSTREAM cards - while the core count on the RX 470 is the SAME as the R9 280x/380x (2048) and the RX 480 has FEWER cores (2304) than the R9 290/390 (2560) or the 290x/390x (2816).

 Expecting the RX 480 MAINSTREAM card to OUTperform the one generation old top-of-the-line R9 390X is just WRONG, given the actual specs on the cards - the clockrate improvement isn't enough to outweight the lower core count, and the memory bandwidth just isn't there at all comparatively.

 Keep in mind that AMD started near the MIDpoint on performance (and a much lower price point) on their new generation - until they release a potential RX 490 or higher, the older cards are going to remain the performance champs (at the cost of HUGE electric usage).

 This is actually some bad news for NVidia, as the midrange AMDs are arguing pretty well against higher-end NVidia cards at a much lower price point for MOST usage....

they didnt have faster ram it was just clocked higher it comes down to if u have elpida or hynix only diff besides that is the 390x stock has 8gb and fewer components cause the chips run more eff cause its a later spin of the same die just they fine tuned the process besides that they are = in speed clock 4 clock