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

[Majestic]

UPDATE #4: Just watch this video instead:

 

 

Just saw this on Tom's Hardware review. It seems the 6-pin was used to sell a story of the card being more energy efficient than it actually was, or cut costs on places they shouldn't have.

 

Source:  http://www.tomshardware.com/reviews/amd-radeon-rx-480-polaris-10,4616-9.html

 

Quote

 It truly takes an observant eye to see why we bemoaned the lack of an eight-pin power connector and the power circuit's design.

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

 

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.

We’re also left to wonder what we'd see from a CrossFire configuration. Two graphics cards would draw 160W via the motherboard’s 24-pin connector; that's a tall order. Switching from the bars back to a more detailed curve makes this even more evident.

Quote

The smaller six-pin PCIe connector implies a more economical graphics card, but let’s be real here: an eight-pin connector would’ve been the way to go. The six phases could’ve been balanced for 30W each, totaling 180W. In such a configuration, two phases would have drawn a maximum of 60W through the motherboard.

We skipped long-term overclocking and overvolting tests, since the Radeon RX 480’s power consumption through the PCIe slot jumped to an average of 100W, peaking at 200W. We just didn’t want to do that to our test platform.

During the stress test:

 

Quote

Believe it or not, the situation gets even worse. AMD's Radeon RX 480 draws 90W through the motherboard’s PCIe slot during our stress test. This is a full 20 percent above the limit.

To be clear, your motherboard isn't going to catch fire. But standards exist for a reason. All of the components around the PCIe slot and along the path from the slot to the 24-pin ATX connector will suffer from the peaks. And depending on your platform's design, audio problems may also materialize.

 

_______________________

 

This is very reminiscent of the disregard AMD had with the specifications regarding the R9-295X. A card that went well over 500W when it was using connections designed for 375W. I think this is pretty serious and shouldn't be overlooked...

Especially considering, unlike the R9-295X, this is going to be a budget king. And people will be using cheaper motherboards that won't necessarily use the beefier construction of these testrigs. The 295X mostly got away with it because it was expensive and mostly used in overpowered systems.

 

UPDATE: AMD has responded about the issue on an AMA:

Quote

 

[–]gfxchiptweeter 236 punten een uur geleden 

Great question and I am really glad you asked.

We have extensive testing internally on our PCIE compliance and RX480 passed our testing. However we have received feedback from some of the reviewers on high current observed on PCIE in some cases. We are looking into these scenarios as we speak and reproduce these scenarios internally. Our engineering team is fully engaged

source: https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m

 

UPDATE1, member @LAwLz notes there are more outlets repoting the same:

 

On 30-6-2016 at 2:11 PM, LAwLz said:

@Majestic

A few more tests has come out and they all show the same result. It seems like it's an issue on all RX 480 cards, not just the one Tom's Hardware ended up getting.

So basically, if you don't have a high quality motherboard and a high quality PSU then it is not recommended to get the RX 480... So much for being a "value king".

 

 

The German website golem.de has tested it and their results shows the same thing as Tom's Hardware.

  Reveal hidden contents

 

The French site hardware.fr also went out, bought a card in a store and tested it. Both their review sample and the store bought card exceeds the PCIe limit.

 

 

A user at Reddit measured it as well, and got the same results.

 

Here is another German site, Heise, which also measured the PCIe power draw and got an average of 88 watts.

 

 

It is pretty safe to say that it is not an isolated issue caused by a fault card. All cards are most likely faulty, and it does not seem like it will be easily fixed.

 

 

These sources were gathered from this Reddit thread.

 

FOR AN ONGOING LIVEBLOG, CHECK REDDIT PAGE:

 

reddit.com/r/Amd/comments/4qfwd4/rx480_fails_pcie_specification/

 

UPDATE2: OC3D video:

 

 

UPDATE3: pcperspective investigated:

 

http://www.pcper.com/reviews/Graphics-Cards/Power-Consumption-Concerns-Radeon-RX-480

 

[Misanthrope]

Oh but Polaris is so efficient!

Seriously though, extreme buyers beware even though is unlikely people who run non gpu PCIE devices should be aware of this. Also does the new overclocking utility allows a hard voltage limit? That could be a temp fix as well.

 

Anyway I guess this is the 3.5 +5 of this generation: many will say it doesn't matter but it clearly does.

[raphidy]

God, I need some refill for my popcorn, the show is coming.

 

Guys who bought the rx 480, don't overclock. That's what you get when you buy new tech one day one. Prepare the beta testing.

[Trixanity]

Response from Raja:

Quote

We have extensive testing internally on our PCIE compliance and RX480 passed our testing. However we have received feedback from some of the reviewers on high current observed on PCIE in some cases. We are looking into these scenarios as we speak and reproduce these scenarios internally. Our engineering team is fully engaged.

 

[SteveGrabowski0]

I can't believe how much power this card uses. When I saw the performance numbers this morning I was impressed, but that was back when I was thinking this was a card that typically uses 110W in gaming loads, not freaking 165W. So it's a slightly more power hungry GTX 970 released 21 months later. Except the 970 was 28 nm, so AMD hasn't gotten their performance per watt under control at all. Where does this leave Vega? 300W to match a 1080?

 

What a disgrace to see them use a 6-pin power connector out of spec to push the narrative of this being some great power efficient card. And then the bait and switch by advertising it as a $199 card and only sending $239 review samples with higher clocked memory to the reviewers.

[TheRandomness]

Why is this such a big issue? It draws more power than specs allow (10 more W..), so?

[Rune]

Trolllolololo

[Majestic]

4 minutes ago, TheRandomness said:

Why is this such a big issue? It draws more power than specs allow (10 more W..), so?

Because it peaks 155W over a bus designed for 75W. Meaning it will stress the mainboard momentarily for double it was designed for.

They note it won't catch fire, but it will cause perhiperal issues like sound or add-in cards.

 

But who knows when someone is using a cheap board and is overclocking it. Meaning it will average 100W over something designed for 75W. It could cause damage.

[raphidy]

1 minute ago, TheRandomness said:

Why is this such a big issue? It draws more power than specs allow (10 more W..), so?

Standards are here to protect the consumer from electric fault/damage. If you don't respect it you should be fined because you exceed the limit that can cause either damage to the product or even becoming hazardous. 

[vanished]

Quote

We have extensive testing internally on our PCIE compliance and RX480 passed our testing. 

lol sounds like you need better testing then!  

 

--

 

So what does this mean for those who were planning to crossfire them?

[GER_T4IGA]

5 minutes ago, TheRandomness said:

Why is this such a big issue? It draws more power than specs allow (10 more W..), so?

Becasue it is out of spec and if shits hits the fan it can probably kill some power delivery circuits on your mobo and the whole thing all together I suppose.

[Majestic]

3 minutes ago, Ryan_Vickers said:

So what does this mean for those who were planning to crossfire them?

Quote

We’re also left to wonder what we'd see from a CrossFire configuration. Two graphics cards would draw 160W via the motherboard’s 24-pin connector; that's a tall order. Switching from the bars back to a more detailed curve makes this even more evident.

I guess wait for a conclusive answer on that.

[Guest]

https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m?st=iq158yqz&sh=dff91e91
https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m?st=iq158yqz&sh=dff91e91
https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m?st=iq158yqz&sh=dff91e91
https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m?st=iq158yqz&sh=dff91e91

 

Are current rumors of RX 480 exceeding PCIe specifications (limited at 75W) true?! And - even if not, would custom RX 480 versions with 8-pin power connector (instead of 6-pin) have a lot more OC potential, since the card seems to be at ~150W maximum even with no overclock.

 

p.s.: my sombrero is very lovely! And thanks AMD for providing new driver support for my aging HD6870 even in 2016 Time to replace it tho!

[edit] for those of you not aware, some reviewers have noticed this potential problem. Others have not been able to replicate it with their review cards, even after being made aware of it. That is why I called it "a rumor".

• permalink
• embed
• save
• report
• give gold
• reply

 

 

[–]gfxchiptweeter 234 points an hour ago 

Great question and I am really glad you asked.

We have extensive testing internally on our PCIE compliance and RX480 passed our testing. However we have received feedback from some of the reviewers on high current observed on PCIE in some cases. We are looking into these scenarios as we speak and reproduce these scenarios internally. Our engineering team is fully engaged.

 

[Majestic]

1 minute ago, JelleDekkers said:

https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m?st=iq158yqz&sh=dff91e91
https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m?st=iq158yqz&sh=dff91e91
https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m?st=iq158yqz&sh=dff91e91
https://www.reddit.com/r/pcmasterrace/comments/4qfy9d/i_work_at_amd_the_time_has_come_to_ama_about/d4smb2m?st=iq158yqz&sh=dff91e91

Yeah, they gave a vague statement. I'll add it, but that's not really a solution.

[ace_cheaply]

3 minutes ago, Majestic said:

Because it peaks 155W over a bus designed for 75W. Meaning it will stress the mainboard momentarily for double it was designed for.

They note it won't catch fire, but it will cause perhiperal issues like sound or add-in cards.

 

But who knows when someone is using a cheap board and is overclocking it. Meaning it will average 100W over something designed for 75W. It could cause damage.

Quote

AMD's Radeon RX 480 draws 90W through the motherboard’s PCIe slot during our stress test. This is a full 20 percent above the limit.

Where are you getting that it draws 155w over a bus designed for 75?  

[Misanthrope]

10 minutes ago, SteveGrabowski0 said:

I can't believe how much power this card uses. When I saw the performance numbers this morning I was impressed, but that was back when I was thinking this was a card that typically uses 110W in gaming loads, not freaking 165W. So it's a slightly more power hungry GTX 970 released 21 months later. Except the 970 was 28 nm, so AMD hasn't gotten their performance per watt under control at all. Where does this leave Vega? 300W to match a 1080?

 

What a disgrace to see them use a 6-pin power connector out of spec to push the narrative of this being some great power efficient card. And then the bait and switch by advertising it as a $199 card and only sending $239 review samples with higher clocked memory to the reviewers.

To me it sounds like they're kind of cheating like they did with the Nano that has a very aggressive power scaling curve, only it's not working properly on the new drivers.

[Tieox]

All this talk is making me think my R9 390 may be the last AMD card I run for a while. 

[Guest]

2 minutes ago, Majestic said:

Yeah, they gave a vague statement. I'll add it, but that's not really a solution.

That's a clear fucking reply to me. Nothing vague about it.

[Majestic]

2 minutes ago, ace_cheaply said:

Where are you getting that it draws 155w over a bus designed for 75?  

Read the post and the source maybe? It's the peak load over the PCI-E bus.

[Guest]

2 minutes ago, Kierax said:

All this talk is making me think my R9 390 may be the last AMD card I run for a while. 

I gave this forum another shot today and it's driving me mad already.

[Majestic]

2 minutes ago, JelleDekkers said:

I gave this forum another shot today and it's driving me mad already.

I can tell from your rabid responses, "doe effe normaal man". Also, by vague I meant it's nothing past an aknowledgement of the allaged issue. They're not confirming or denying it.

[raphidy]

4 minutes ago, Kierax said:

All this talk is making me think my R9 390 may be the last AMD card I run for a while. 

I was thinking of selling my 390, now is just a shit show... I think I'll wait for 1070 to drop a bit.

[TheRandomness]

1 minute ago, raphidy said:

Standards are here to protect the consumer from electric fault/damage. If you don't respect it you should be fined because you exceed the limit that can cause either damage to the product or even becoming hazardous. 

The 295x2 disagrees. So does the 480 apparently. Plus, it's only 15W in total (ish)

2 minutes ago, Majestic said:

Because it peaks 155W over a bus designed for 75W. Meaning it will stress the mainboard momentarily for double it was designed for.

They note it won't catch fire, but it will cause perhiperal issues like sound or add-in cards.

 

But who knows when someone is using a cheap board and is overclocking it. Meaning it will average 100W over something designed for 75W. It could cause damage.

150W* > PCIe bus + PCIe power cables

*could* I can draw a total of 400ish W with my 290 when it's overclocked to hell with an 8 pin and a 6 pin. 400~ > 375. 15W difference, no harm done. In fact, I'm considering pushing the wattage higher to see how far I can go. No damage done to anything so far, the only thing stopping me being thermals (This'll be my overclock for gaming btw)

1 minute ago, Ryan_Vickers said:

lol sounds like you need better testing then!  

--

So what does this mean for those who were planning to crossfire them?

That they need a 650-750W PSU c:

 

BTW: I know this is kindof a 'my card does it so it's fine' post, but my point is that is the spec is broken by like 15W, is it going to set my house on fire?

[ace_cheaply]

16 minutes ago, Majestic said:

Read the post and the source maybe? It's the peak load over the PCI-E bus.

What they say makes no sense.

Quote

We skipped long-term overclocking and overvolting tests, since the Radeon RX 480’s power consumption through the PCIe slot jumped to an average of 100W, peaking at 200W.

The draw through the pci-e reached a peak of 200W, yet the card never draws more than 168, so even assuming none of the 168W comes from the 6 pin connector, that is still less than their peaks. We are to believe that without overclocking it reaches a max of 300W?  200W peak through the pci-e, and 100 from the 6 pin?  I don't believe it. If it's true, it should be repeatable and we'll know soon enough. 

 

I might not be getting a 480 after all. 

[vanished]

1 minute ago, TheRandomness said:

That they need a 650-750W PSU c:

 

BTW: I know this is kindof a 'my card does it so it's fine' post, but my point is that is the spec is broken by like 15W, is it going to set my house on fire?

Probably not, but it's the principal.  You simply don't do this.  Specifications are there for a reason.  What if your USB ports put out 6 volts instead of 5?  There's also something to be said for integrity in advertising.  What if your car consumed 20% more gas than it claimed?  Plus, OP made a great point about many people who will probably run these probably have very cheap motherboards which won't handle this as well as some might.