The 480: power consumption, PCI-E powerdraw

[Abwx]

I wonder if my black screen had anything to do with this over current into the PCI-E Slot. I was getting spikes to 1.275v and not even overclocking or raising the power target.

Unlikely since Hawai cards do not use a lot of power from the PCIe connector.

http://www.tomshardware.com/reviews/amd-radeon-r9-390x-r9-380-r7-370,4178-9.html

Looking at the profile of this graph the 12V on the MB seems to be dedicated to the RAM..

 

[IEC]

Unlikely since Hawai cards use not a lot of power from the PCIe connector.

*snip*

Rvenger is talking about his RX 480s he picked up from Microcenter.

Edit: Rvenger answered below.

 

[Rvenger]

Black screen on AMD cards is symptomatic for undervoltage conditions. I had a Sapphire Tri-X that initially worked fine but as it aged required a BIOS flash that gave +25mV to correct the black screen issue. Are you getting a black screen only on OS boot, or also in BIOS?

Overvoltages don't manifest themselves via black screens. Unless it's a permanent "black screen" like I got when I fried one of my cards via a voltage hard mod.

The 290x black screen the monitor went into power saving mode. This one did not. At the time I was getting a ton of spikes to 1.275v. I put in the second card and no black screens and no voltage spikes.

I put the first card in my back up rig and no voltage spikes and no black screen so far. Very odd behavior, I almost think its a bad driver.

 

[IEC]

The 290x black screen the monitor went into power saving mode. This one did not. At the time I was getting a ton of spikes to 1.275v. I put in the second card and no black screens and no voltage spikes.

I put the first card in my back up rig and no voltage spikes and no black screen so far. Very odd behavior, I almost think its a bad driver.

That's certainly bizarre. It seems unlikely to be the components themselves in either system due to working in separate rigs (unless one card is just marginal).

As you say, perhaps a software issue.

 

[redzo]

I wonder if my black screen had anything to do with this over current into the PCI-E Slot. I was getting spikes to 1.275v and not even overclocking or raising the power target.

You may be getting over-paranoid over some unsolved issue. My point is that it doesn't matter if it's the pci-e slot. It's the motherboard and that is it. There is no point in going further down the road since you are not going to repair the motherboard.
If there is one component to look up for in the long run, that is your PSU. Both your MB and gpu will age faster if the psu is sh*t.

 

[Abwx]

The 290x black screen the monitor went into power saving mode. This one did not. At the time I was getting a ton of spikes to 1.275v. I put in the second card and no black screens and no voltage spikes.

I put the first card in my back up rig and no voltage spikes and no black screen so far. Very odd behavior, I almost think its a bad driver.

The 480 max GPU voltage is 1.150V apparently, possibly that there s some protection triggered if you did set it at 1.275V but it s impossible to be sure of this currently.

 

[IEC]

The 480 max GPU voltage is 1.150V apparently, possibly that there s some protection triggered if you did set it at 1.275V but it s impossible to be sure of this currently.

Pretty sure Rvenger didn't manually up the voltage. Wattman doesn't allow you to set beyond 1.150V so he's indicating something outside of his control was causing the voltage spikes.

Which was seemingly resolved by putting it in a different system. Suggests something at the BIOS (boot-time calibration?) or driver level is out of whack, unless the card itself was marginal.

 

[96Firebird]

There's no need for ad-hominems.

I know right, when someone says another is biased that is considered ad hominem. Abwx, take advice from IEC here.

 

[antihelten]

But the GTX960 peaks at 250W on the PCIe connector, why doesnt the site said that according to his own words :

snip...

So where was the outcry at the time, and why smaller numbers are suddenly an issue...?..

You're making it sound like this was a general issue with the 960, when the truth is that Toms only observed it with a single after-market model (the ASUS 960 Strix).

And they did in fact call ASUS out on it:

This is because the otherwise very good Asus GTX 960 Strix leaves the motherboard connector to deal with unprecedented unfiltered power spikes all on its own:

The very frequent spikes beyond the motherboard slot’s supposed limit won’t cause immediate damage to the hardware, but there might well be long-term repercussions that are hard to judge now. The same goes for how the system might otherwise be impacted with problems such as “chirping” on-board sound when the mouse is moved. The Asus GTX 960 Strix should do a much better job smoothing these spikes out.

And the general outcry you are seeing right now is probably because a single after-market model having issues with the PCIe spec is a hell of a lot different from the reference model having issues with the PCIe spec.

 

[Seba]

But the GTX960 peaks at 250W on the PCIe connector, why doesnt the site said that according to his own words :

So where was the outcry at the time, and why smaller numbers are suddenly an issue...?..

Only Asus GTX 960 Strix did that. Other GTX 960 cards tested by them did not consistently went above 75W from PCI-E slot.

And the RX 480 looks worse than Asus GTX 960 Strix:

 

[IEC]

Edit: Meh.

 

[William Gaatjes]

A question ?

How high is the chance that AIB's will equip their RX 480 cards with an 8 pin connector that feeds the six phases and use the pcie slot to feed the other two "low" power smps ?

Is that feasible ?

Most cards do not startup if you do not connect the pcie power connector yes ? Or give a warning and clock down ?

 

[PhonakV30]

Possible Wattman automatically sets clock on 1150mhz ? with most stable low voltage like 1.00 v ? with this there will be no issue.

Edit : one thing : perhaps ASIC is really related to low voltage and not possible to reduce Voltage on High ASIC.only Clock?

 

[Abwx]

I know right, when someone says another is biased that is considered ad hominem. Abwx, take advice from IEC here.

No ad hominem, it s just that this thread is one among many usual ones full of people complaining about a product they wont buy anyway...

 

[MrTeal]

A question ?

How high is the chance that AIB's will equip their RX 480 cards with an 8 pin connector that feeds the six phases and use the pcie slot to feed the other two "low" power smps ?

Is that feasible ?

Most cards do not startup if you do not connect the pcie power connector yes ? Or give a warning and clock down ?

I would say it's almost a certainty that at least some will. If we see 8, 6+6 or the rumored 8+6 they would almost certainly draw core power from the PCIe plugs and use the slot for the fan and minor rails as most video cards do.

Keep in mind, outside of the disappointing efficiency numbers, this situation isn't an issue with Polaris 10. It's strictly an issue of how AMD decided to distribute power on the board and their decision to saddle the board with a 6-pin connector.

It's really kind of sad and silly; had they shifted more rails to the PCIe power plug this would be a non issue. Even with a 6 pin connector, the thing is physically capable of supplying hundreds of watts. They'd probably end up being a little over spec on the 75W draw for the PCIe 6 pin, but under spec on the PCIe slot. The difference is that the PCIe slot spec is actually pretty close to the connector rating, while the Minifit Jr power plug can handle several times more current than the 6 pin PCIe spec.

This whole situation with this and the overpowered reference cooler just screams of last minute clock adjustments to get performance up.

 

[antihelten]

Keep in mind, outside of the disappointing efficiency numbers, this situation isn't an issue with Polaris 10. It's strictly an issue of how AMD decided to distribute power on the board and their decision to saddle the board with a 6-pin connector.

...

This whole situation with this and the overpowered reference cooler just screams of last minute clock adjustments to get performance up.

This is the weird thing though, AMD obviously didn't design RX 480 the way they did (with a single 6-pin) knowing that it would use more than 150W, so something must have gone wrong with the final power usage vs their design goals.

It is possible the last minute clock adjustments is the answer, but I think this is somewhat unlikely. AMD obviously sent of the cards to get the tested and approved for the PCIe spec, and this approval would obviously only count for the cards running the same specs as those they were tested under*. So if AMD originally got the PCie approval with the RX 480 running at say 1100MHz, then the approval wouldn't count for any versions running at 1266 MHz. As such I think they must have had it tested running at 1266 MHz, and then between the tests and the final release something went wrong (BIOS, drivers who knows). I could for example be that later Drivers had significantly better utilization (done to improve performance obviously), which would also have an effect on power usage.

*This is my assumption anyway. I know that AMD had them tested by an external company (Raja said as much), but I don't know what the exact test methodology was.

 

[guskline]

AMD, correct this.

 

[Rvenger]

AMD, correct this.

All they need to do is to cap the PCI-e power to a maximum of 66w and stuff the rest of the power to the 6 pin. The 6 pin will still be able to handle the extra 20w of draw no problem.

 

[Erenhardt]

nonissue.

Just plug any 75W PCIe card without additional power connectors to PCIe1.0 slot, which is supposed to supply 60watts

And blast from the past:
http://forums.anandtech.com/showthread.php?t=2411556

that was on the green side many times before non-issue?

 

[Erenhardt]

And if you are really concerned, get a pcie riser with molex power adapter. Fixed.

 

[William Gaatjes]

I would say it's almost a certainty that at least some will. If we see 8, 6+6 or the rumored 8+6 they would almost certainly draw core power from the PCIe plugs and use the slot for the fan and minor rails as most video cards do.

Keep in mind, outside of the disappointing efficiency numbers, this situation isn't an issue with Polaris 10. It's strictly an issue of how AMD decided to distribute power on the board and their decision to saddle the board with a 6-pin connector.

It's really kind of sad and silly; had they shifted more rails to the PCIe power plug this would be a non issue. Even with a 6 pin connector, the thing is physically capable of supplying hundreds of watts. They'd probably end up being a little over spec on the 75W draw for the PCIe 6 pin, but under spec on the PCIe slot. The difference is that the PCIe slot spec is actually pretty close to the connector rating, while the Minifit Jr power plug can handle several times more current than the 6 pin PCIe spec.

This whole situation with this and the overpowered reference cooler just screams of last minute clock adjustments to get performance up.

Or something else is going on.
Have you any idea how many chips use something similar as BTC (Boot Time Calibration) for polaris ?
I wonder if Nvidia and Intel use it as well at their 16nm nodes ?
I wonder if samsung uses it for their ARM SOC or the apple SOC ?

Does AMD use for the radeon the 14nm LPE or the 14nm LPP process ?
Do you know ?

What the stilt mentioned was very interesting with leakage and leakage fuses.
I really need to read up on it. It is interesting.

 

[Sonikku]

Do you guys think that the other brands will have this issue sorted out and include maybe an 8 pin connector in just a couple weeks or are they already being manufactured? I'm considering an XFX non blower variant in July provided some are available.

 

[Zstream]

Only Asus GTX 960 Strix did that. Other GTX 960 cards tested by them did not consistently went above 75W from PCI-E slot.

And the RX 480 looks worse than Asus GTX 960 Strix:

Higher spikes is definitely worse for electronic equipment. If you don't believe me, go as an electrical engineer. My friend is one and spikes, along with tin splinters causes frequent damage to electronics is far worse than steady voltages, as long as it's not too far beyond rated capability.

 

[bill3]

You're making it sound like this was a general issue with the 960, when the truth is that Toms only observed it with a single after-market model (the ASUS 960 Strix).

And they did in fact call ASUS out on it:



And the general outcry you are seeing right now is probably because a single after-market model having issues with the PCIe spec is a hell of a lot different from the reference model having issues with the PCIe spec.

This is disingenuous because only two aftermarket 960's were tested in this way.

One of them failed, doesn't seem like a very good ratio.

It's disingenuous in the same way as when AMD says "out of hundreds of reviews only a few had this 480 problem". Yes, because only a few tested power draw from PCIE slot....

Also, aftermarket cards are supposed to be safer, more over engineered, and better than reference cards. That an AIB 960 had this problem is more damning, not less. The AIB 480's are the ones people are looking to to correct the 480 problem!

Regardless, there was no hue and cry over the failing 960. that much we know for sure.

 

[bill3]

I just want to know one single simple thing that I cant really find answers for:

IS THIS (in theory) FIXABLE IN SOFTWARE? (Bios or Drivers)? What would that fix look like?

Part of me says yes, but part of me thinks the problem is a card that draws more than 150 watts, from two sources that max at 75 watts, is a card that draws more than 150 watts...

Somebody above said AND should just "rearrange" the power draw so that the excess is drawn from the 6 pin...but is that even possible in bios/driver? Or would it require a PCB revision? That's the million dollar question for me! Because I have a reference 480 sitting in a box...I could refuse shipment on it, but probably need to take it back to fedex no later than today if I choose that option...