AMD once again violating power specifications? (AMD RX-480)

[patrickjp93]

13 minutes ago, stconquest said:

...and you are also the one that believed this RX 480 would perform at the Fury level.  I like you, but give it a rest.  

It's what AMD promised so... And it can perform like a Nano in a couple DX 12 titles.

[stconquest]

1 minute ago, patrickjp93 said:

It's what AMD promised so... And it can perform like a Nano in a couple DX 12 titles.

What evs.

[79wjd]

4 hours ago, Briggsy said:

True, it's not foolproof, but at least it would give and time to release a prelaunch bios to lock down pcie power draw, if possible. If they gotta recall, that will be the second amd launch in a row that gets rma'd to death.

Keep in mind, Tomshardware could have (and very likely did) said something to AMD when the problem was discovered (before NDA broke). So you don't technically need NDA to end by launch in order to protect the buyer on day 1. 

 

2 hours ago, AlwaysFSX said:

Remember the MSI Krait having significantly worse performance with GPUs than literally every other motherboard?

What now? (Also which Krait Z97/Z170/970? -- I assume 970).

[Revinval]

The best thing about this issue is its only one if you buy the reference design! Of course its bad but even keeping the 6 pin and swapping some of the vrms would solve the problem (since most 6-8 pin power can easily push their minimum rating and then some).

[zMeul]

14 minutes ago, Revinval said:

The best thing about this issue is its only one if you buy the reference design! Of course its bad but even keeping the 6 pin and swapping some of the vrms would solve the problem (since most 6-8 pin power can easily push their minimum rating and then some).

"swapping MOSFETs" isn't the solution since the power delivery circuitry needs a complete overhaul to include a PEG power limiter - clearly RX480 lacks it and load balances the power draw between the 6pin PCIe and PEG

 

maybe this should be a wake-up call fro mobo manufacturers to include overpower protection for the PEGs

[Knaj]

Why can't companies just not lie...

We wouldn't have this bullshit going on then.

[Prysin]

5 hours ago, Majestic said:

Source?

 

EDIT: 

http://www.tomshardware.com/reviews/nvidia-geforce-gtx-980-970-maxwell,3941-12.html

 

The 970 and 980 sure didn't.

it was a 980Ti. Not reference model. Cannot find the review atm, will look later when i get back from work.

[Briggsy]

some reviewer stuck a tower cpu cooler on the 480 and was able to reach 1400mhz, but the chip alone was pulling up to 185 watts, which doesn't include the whole card power draw. I wonder if the card would pull 300 watts with 8+6 pin connectors and 1500mhz..  with just a 6-pin the card is definitely not being limited much by temps, at least we can now be relatively certain that custom AIB cards with just a 6-pin are not going to overclock much better than reference.

 

http://oc.jagatreview.com/2016/06/teaser-overclocking-amd-radeon-rx480-ke-1-4ghz-dengan-cooler-3rd-party/

Quote

google translate: The Variations of overclock-ability for the Radeon RX480 we do not know, and based on the 4 GPUs that we used (2 from AMD, 2 from PowerColor), only 1 GPU was capable of running at 1.4GHz clock. The rest varies between 1.33-1.35Ghz.

 

 

[AlwaysFSX]

1 hour ago, djdwosk97 said:

What now? (Also which Krait Z97/Z170/970? -- I assume 970).

Z97 iirc.

[vanished]

44 minutes ago, Knaj said:

Why can't companies just not lie...

We wouldn't have this bullshit going on then.

No pressure not to if they can lie and still sell product.

[Majestic]

7 hours ago, RagnarokDel said:

There's no way in hell that's accurate. Just on the temperatures alone, this would make the card meltdown.

Those are 1 peaks measured at 1 uS. Not enough to cause a meltdown, given the time of the spike. But it can cause interference, which is what Tom's hardware says.

However, when they started overclocking the Average wattage on the PCI-E went from an already exceeding 82, to 100W. Now that can damage mainboards, possibly. And neither can you compensate with a proper PSU.

6 hours ago, Dan Castellaneta said:

Of course you won't hear a WattGate, because you have all of the defenders of AMD.

Although, as @Curufinwe_wins has pointed out a few times, this doesn't mean anything for people running modern motherboards and PSUs, still something that should be noted.

This isn't like the 295X, you can't compensate with a different PSU. And this issue hasn't come up much, so I doubt many mainboards will be running super beefy PCI-E slot power phases.

5 hours ago, patrickjp93 said:

When they tried overclocking it. To me it seems a firmware issue.

Those 155W spikes were at stock, they went up to 200W during OC. And 100W avg.

4 hours ago, Revinval said:

The best thing about this issue is its only one if you buy the reference design! Of course its bad but even keeping the 6 pin and swapping some of the vrms would solve the problem (since most 6-8 pin power can easily push their minimum rating and then some).

Uhm, based on what? I haven't seen tom's hardware test one yet (or any other outlet).

3 hours ago, Prysin said:

it was a 980Ti. Not reference model. Cannot find the review atm, will look later when i get back from work.

Ok

[Energycore]

Out of all of you who take up the pitchforks and scream at AMD for violating PCI-E, how many of you know whether or not your power supply passes ATX spec on voltage regulation, ripple, hold-up time, inrush current and transient loads?

 

[Majestic]

3 minutes ago, Energycore said:

Out of all of you who take up the pitchforks and scream at AMD for violating PCI-E, how many of you know whether or not your power supply passes ATX spec on voltage regulation, ripple, hold-up time, inrush current and transient loads?

 

RM550x has one of the lowest, if not lowest, ripple of all consumer power supplies at the moment. 

https://nl.hardware.info/reviews/6698/11/corsair-rm550x--rm650x-review-sterke-middenklassers-ripple

 

So yes, I do.

[Energycore]

1 minute ago, Majestic said:

RM550x has the lowest ripple of all power supplies at the moment. 

https://nl.hardware.info/reviews/6698/11/corsair-rm550x--rm650x-review-sterke-middenklassers-ripple

 

So yes, I do.

That's one. And counting.

[Energycore]

5 hours ago, patrickjp93 said:

It's what AMD promised so... And it can perform like a Nano in a couple DX 12 titles.

I don't think AMD ever said that, people "leaked" benchmarks that suggested it and we all took it because we wanted to believe

 

We got played, not by AMD but by the people who released leaks.

[Stefan1024]

Voilating the spec is nit good.

But seriousely, this connectors are heavily underrated. If you look at a typical PCIe cable is has 2 8 pinn connecors (300 watts) and only 8 pinns on the PSU side so they have double the current per pinn and are totally fine.

 

Also the MoBo needs to be able to provide 75 watts on every 16x slot. Given a lot of the ATX boards have at least 4, this is 300 watts. And the bigger boars have an additional molex or 6 pin to the slots anyway.

 

Summary: you only runn in problems with 4 way CF. And this configuration is stupid with a midrange card.

[Majestic]

4 minutes ago, Stefan1024 said:

 

Like it has been said a couple of times now, it's a budget card. Thus meaning the motherboards it will be used on won't be Z170 Mpowers, but H110/H170 boards.

 

Also, the 155W spikes will cause the same issues on your heavy-duty board in terms of perhipherals and/or soundcards.

 

Also, i'm not familiar with the way PCI-E power is layed out, but if the phases are placed per-slot, having more slots doesn't change the distribution if the single slot is overloaded. I'm not sure it's a sort of "rail" they all tap off of. That would make the pathing too heavy duty.

[stconquest]

3 minutes ago, Majestic said:

Like it has been said a couple of times now, it's a budget card. Thus meaning the motherboards it will be used on won't be Z170 Mpowers, but H110/H170 boards.

 

Also, the 155W spikes will cause the same issues on your heavy-duty board in terms of perhipherals and/or soundcards.

Proof?  Not saying you are wrong, just asking you to verify a claim.

[Majestic]

10 minutes ago, Stefan1024 said:

Also the MoBo needs to be able to provide 75 watts on every 16x slot. Given a lot of the ATX boards have at least 4, this is 300 watts. And the bigger boars have an additional molex or 6 pin to the slots anyway.

Quote

AMD’s Radeon RX 480 draws an average of 164W, which exceeds the company's target TDP. And it gets worse. The load distribution works out in a way that has the card draw 86W through the motherboard’s PCIe slot. Not only does this exceed the 75W ceiling we typically associate with a 16-lane slot, but that 75W limit covers several rails combined and not just this one interface.

 

 

4 minutes ago, stconquest said:

Proof?  Not saying you are wrong, just asking you to verify a claim.

Quote

With peaks of up to 155W, we have to be thankful they're brief, and not putting the motherboard in any immediate danger. However, the audio subsystems on cheaper platforms will have a hard time dealing with them. This means that the "you can hear what you see" effect will be in full force during load changes; activities like scrolling may very well result in audible artifacts.

Maybe some more expensive boards will be better protected against this, maybe they won't.

[Stefan1024]

3 minutes ago, Majestic said:

Like it has been said a couple of times now, it's a budget card. Thus meaning the motherboards it will be used on won't be Z170 Mpowers, bug H110/H170 boards.

 

Also, the 155W spikes will cause the same issues on your heavy-duty board in terms of perhipherals and/or soundcards.

The ATX specs are still heavily underrated, at least when it comes to the connectors.

Also proper designed regulator should be able to compensate spikes as long as they are not very shot (10 us) and all over the place.

A 200 ms long "spike" is like two steady states for a voltage regulator.

[stconquest]

3 minutes ago, Majestic said:

Maybe some more expensive boards will be better protected against this, maybe they won't.

...and that is why I ask.

 

I read all that stuff too BTW, and I don't take this report from TH lightly either.  ty for humoring me.

[Majestic]

1 minute ago, Stefan1024 said:

The ATX specs are still heavily underrated, at least when it comes to the connectors.

Also proper designed regulator should be able to compensate spikes as long as they are not very shot (10 us) and all over the place.

A 200 ms long "spike" is like two steady states for a voltage regulator.

Read my second post, also this isn't ATX specification but PCI-SIG specification

[alamox]

i hoped that AMD wouldn't be stupid enough to push the clock  way to far looking for performance, so that they end up out of the card's power envelope or limited OC headroom.

and they went and did just that, to a ridiculous extent, the 480 probably was probably designed for something like 1Ghz base clock, and performance wasn't good enough.

[DXMember]

14 hours ago, Crazy Ginger said:

Probably not, but it could damage your motherboard and processor, and other pics devices like for instance, a soundcard, network card, or even an expensive pcie NVME drive.  Basically, if it does damage your mobo, you better hope that only the mobo dies, as a damaged motherboard can easily damage anything connected to it.  If a short circuit occurs, theres a chance that it will send too many volts to your processor.

over power and short circuit are different things, both should have apropriate protection in place that allows for a safe shut down

[DXMember]

5 hours ago, Revinval said:

The best thing about this issue is its only one if you buy the reference design! Of course its bad but even keeping the 6 pin and swapping some of the vrms would solve the problem (since most 6-8 pin power can easily push their minimum rating and then some).

the VRMs on the board are just fine, those are practically the same VRMs that Fury cards have, the problem is probably the controller that doesn't balance the load according to spec