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

[vanished]

oh no I have started this up again... 

 

[stconquest]

9 minutes ago, zMeul said:

PcPer conducted their tests on a ASUS Rmapage V if I recall right

If you read the comments there is a lot of good information coming from Allyn.  Ryan jumps in to refer to his pick for testing, a H170 mobo.

 

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

 

June 30, 2016 | 07:40 PM - Posted by Ryan Shrout

"For our part, we are going to be plugging the Radeon RX 480 into a couple of older platforms and running it in some “bad case” scenarios…just to see what happens."

 

June 30, 2016 | 08:22 PM - Posted by Ryan Shrout

"How does an H170 board sound?"

 

[zMeul]

1 minute ago, stconquest said:

If you read the comments there is a lot of good information coming from Alyn.  Ryan jumps in to refer to his pick for testing, a H170 mobo.

I suggested him (Ryan) to test on a cheapo mobo - dunno if they updated the article

H170 aren't cheapo, they're mainstream

 

also, you need to take into account AMD still has PCie gen2 in their offerings; except APUs that have gen3 PEG

wasn't MSI who built a AM2 board that kept blowing up FETs when people OCed it ...

[stconquest]

2 minutes ago, zMeul said:

I suggested him (Ryan) to test on a cheapo mobo - dunno if they updated the article

H170 aren't cheapo, they're mainstream

 

also, you need to take into account AMD still has PCie gen2 in their offerings; except APUs that have gen3 PEG

wasn't MSI who built a AM2 board that kept blowing up FETs when people OCed it ...

I know, I just remembered you were on those boards too. 

[zMeul]

1 minute ago, stconquest said:

I know, I just remembered you were on those boards too. 

haven't checked back, I have some stuff to deal with and will probably check it later today

[LAwLz]

17 minutes ago, zMeul said:

PcPer conducted their tests on a ASUS Rmapage V if I recall right

Yes but they will do other tests as well.

This is a quote from their latest article:

Quote

For our part, we are going to be plugging the Radeon RX 480 into a couple of older platforms and running it in some “bad case” scenarios…just to see what happens.

Then you posted a comment saying:

Quote

"older platforms" doesn't mean less overbuilt - cheap mobo means cheap mobo
please consider my suggestion

and Ryan replied:

Quote

How does an H170 board sound?

Not sure if I agree that a H170 board is old or low end, but it's at least more low-end than a Rampage V or whatever they used.

 

 

14 minutes ago, laminutederire said:

As for the 750Ti, I couldn't care less, it doesn't matter what it does, we're talking about what happens with the 480 and how it's going to be fixed, so I didn't informed me very well on that matter, since it's off topic anyway.

It does actually matter what it does, because you were saying it had issues as well.

You can't accuse a card of breaking specifications and then go "oh but that card doesn't matter" when proven wrong.

 

But hey it's Nvidia so fuck logic and reasoning right? Let's all just throw shit at them without doing proper research. So much for the forum being Nvidia biased...

 

 

7 minutes ago, zMeul said:

wasn't MSI who built a AM2 board that kept blowing up FETs when people OCed it ...

Happened with some MSI AM3 boards as well.

(Edit: Holy crap that was about 5 years ago. Time sure flies)

[vanished]

3 minutes ago, LAwLz said:

So much for the forum being Nvidia biased...

I thought we were AMD biased?  So hard to keep up...  

[Dabombinable]

4 minutes ago, Ryan_Vickers said:

I thought we were AMD biased?  So hard to keep up...  

No no no, its Tek Syndicate who are AMD biased 

[zMeul]

13 minutes ago, Dabombinable said:

Tek Syndicate

GTFO!

out of curiosity, did they even benched the RX480? or they were like "nope, let's sell more DACs"

[Bleedingyamato]

9 minutes ago, LAwLz said:

Yes but they will do other tests as well.

This is a quote from their latest article:

Then you posted a comment saying:

and Ryan replied:

Not sure if I agree that a H170 board is old or low end, but it's at least more low-end than a Rampage V or whatever they used.

 

 

It does actually matter what it does, because you were saying it had issues as well.

You can't accuse a card of breaking specifications and then go "oh but that card doesn't matter" when proven wrong.

 

But hey it's Nvidia so fuck logic and reasoning right? Let's all just throw shit at them without doing proper research. So much for the forum being Nvidia biased...

 

 

Happened with some MSI AM3 boards as well.

(Edit: Holy crap that was about 5 years ago. Time sure flies)

If a Z or X chipset is "high end" to most people I'd assume then that an H170 would be the upper end between it and the H110 chipset which if I'm not mistaken is the most basic/low end 1151 chipset.  So at least like you said an H170 is a better example of a low end/average user grade motherboard that is more representative of a larger portion of the total PC user base.  

 

I'd like to see testing of the 480 done with maybe some cheap H110 motherboards and see how they handle the power draw from the 480.  That or maybe some B150 boards too.

[stconquest]

7 minutes ago, Bleedingyamato said:

If a Z or X chipset is "high end" to most people I'd assume then that an H170 would be the upper end between it and the H110 chipset which if I'm not mistaken is the most basic/low end 1151 chipset.  So at least like you said an H170 is a better example of a low end/average user grade motherboard that is more representative of a larger portion of the total PC user base.  

 

I'd like to see testing of the 480 done with maybe some cheap H110 motherboards and see how they handle the power draw from the 480.  That or maybe some B150 boards too.

Well, have have a first confirmed failure of a mobo due to excessive power draw from the RX 480:  Some Foxconn AM2 board.  Did you watch that vid from the Science Studio?

 

The board is fine, the system would just shut down at GPU intensive moments.

[TheRandomness]

11 hours ago, Blake said:

go read https://www.reddit.com/r/Amd/comments/4qmlep/rx_480_powergate_problem_has_a_solution/

 

And clearly you haven't read it. otherwise you would also know this is a nothing story and never was, but the usual "DRAMA ALERT" that gets thrown around every time there is a new product release or some other polarised topic mentioned.

Because not everyone will click that link:

Quote

The motherboard it self sets the maximum allowed watt per slot in the "Slot Capabilities Register" which you can configure up to over 300 watt per slot. In the bits 7 to 14 "Slot Power Limit Value" you can set 250, 275, 300 and above 300 watt. This will be multiplied with bits 15 to 16 "Slot Power Limit Scale" in steps x1 ,x0.1, x0.01 and x0.001. So its up to the motherboard manufacturer and the power management on it how many watt the slot is capable of. The Specifications do define the protocol and not the hardware specs of the PCI-E slot. If a manufacturer uses better parts which can handle higher amps on the contacts and the lines, they can allow the devie in the slot a higher power consumption than 75 watt via these registers. Sadly most people doesnt even read the specifications and judge things they dont understand."

^^^

Apparently irrelevant.

[mariushm]

@TheRandomness

 

The author of that reddit post is incorrect, the text refers to the COMPLETE power consumption of a device. The maximum power drawn from the slot is still limited to 75 watts (5.5a x 12v  + 3a x 3.3v + ~0.4A x 3.3v aux / standby)

The card can report to the motherboard how much power it uses through those registers but motherboard has no real control over how much power the card will pull through pci-e 6pin or 8pin connectors, there's no way to control that. (well, maybe in future atx revisions we will have some bi-directional communication between power supply and motherboard but as it stands now, it's not possible)\

 

It's physically impossible for the pci-e x16 connector to carry more than 100-150 watts continuously, there's only 5 contacts with too narrow of a surface to carry more current x voltage.

Maybe in pci-e v5 we will have 20v or higher voltage used instead of 12v and then we may have higher wattage. Right now, we're limited by current amount.

[TheRandomness]

2 minutes ago, mariushm said:

@TheRandomness

 

The author of that reddit post is incorrect, the text refers to the COMPLETE power consumption of a device. The maximum power drawn from the slot is still limited to 75 watts (5.5a x 12v  + 3a x 3.3v + ~0.4A x 3.3v aux / standby)

The card can report to the motherboard how much power it uses through those registers but motherboard has no real control over how much power the card will pull through pci-e 6pin or 8pin connectors, there's no way to control that. (well, maybe in future atx revisions we will have some bi-directional communication between power supply and motherboard but as it stands now, it's not possible)

oh ffs then... The arguments shall continue.

[zMeul]

19 minutes ago, Bleedingyamato said:

I'd like to see testing of the 480 done with maybe some cheap H110 motherboards and see how they handle the power draw from the 480.  That or maybe some B150 boards too.

ADM boards too, since you know .. it's an AMD product

[laminutederire]

1 hour ago, LAwLz said:

 

Well that's the only time I talked about the 750Ti, and I just wanted to make the nvidia vs amd thing stop,  so sue me for trying to calm things down a notch and give this thread the possibility to go on topic again.

[patrickjp93]

8 hours ago, Ryan_Vickers said:

Hm I suppose I did  

But that brings it up to 76 W basically (not counting the +/- range this time)... did they just round it off then?

In all likelihood.

[LAwLz]

17 minutes ago, laminutederire said:

Well that's the only time I talked about the 750Ti, and I just wanted to make the nvidia vs amd thing stop,  so sue me for trying to calm things down a notch and give this thread the possibility to go on topic again.

Excuse my French but are you dense? How in God's name do you think that accusing Nvidia of the same thing, without properly researching it, would stop the Nvidia vs AMD flamewar? 

 

It's like trying to pour gasoline on a fire in order to put it out. Even if your statement was true (which it isn't) you would still end up with the exact opposite effect than what you tried to accomplish. 

 

Wanna know how to make the Nvidia vs AMD flame war stop? By accepting facts that AMD fucked this up. The only reason this conversation has gone on this long is because ignorant people are desperately trying to defend AMD. If everyone instead went "yeah AMD fucked up. Let's see if the custom cards will have the same issue" then the thread would have been over in a single page. Bringing up Nvidia does nothing but make you look like a massive fanboy and make the topic go on even longer. Your ignorant response has now lead to something like 6 more posts. Even if your post wasn't based on misinformation we would still have ended up with people responding to your post, thus continued the thread for even longer. 

 

This thread isn't about Nvidia so your response would not have "gone on topic" again even if it was correct. In fact, it's people like you who try to make the thread go off topic. This thread is about the RX 480, not the 750 Ti. 

[Khvarrioiren]

9 minutes ago, LAwLz said:

By accepting facts that AMD fucked this up. The only reason this conversation has gone on this long is because ignorant people are desperately trying to defend AMD. If everyone instead went "yeah AMD fucked up.

Some people are too forgiving, sadly. A fuck up is a fuck up, no matter how much you sprinkle it with glitter and hopes for a better tomorrow.

[patrickjp93]

3 hours ago, laminutederire said:

Mathematics doesn't prove much in real life. In that case, they allow us to use a model which happen to be close enough to reality, as experimentation prove it to be acceptable to a certain degree.

I know that mathematics are so much more intense and beautiful when they aren't done for physics though.

I'm afraid computer scientists have much to say to the contrary. You don't have to test that an algorithm is correct if you can mathematically prove it is, which is usually just a matter of a double-ended proof by contradiction or contra position with some induction mixed in.

 

Then you have Amdahl's Law and Gustafson's Law which were proven and didn't require testing for those who were literate with the concept of infinity and series (Calculus II basically).

 

Then you have NP-Completeness and the polynomial hierarchy which is extremely useful for backing yourself up when you say a particular problem cannot have an efficient (read: polynomial time) solution and get a 100% correct answer. Enter polynomial-time approximation schemes (PTAS) and their proofs of time scaling with accuracy.

 

Then there's the randomized algorithm proofs of completeness saying even with randomness used to drive key parts of decisions in an algorithm, it will finish within a given amount of time guaranteed.

 

Mathematics that is provable is far more valuable than mathematics requiring modeling, primarily because it cuts down on the amount of time you need to spend testing your models.

 

[patrickjp93]

7 hours ago, Ryan_Vickers said:

Actually that raises an interesting point.  We're all flipping out over how much power is being drawn through the PCIe slot, but if we're going to be "proper" about this, shouldn't we independently test the 12 v and 3.3 v rails?  Even if a card is only drawing, say, 74 W though the slot, if 100% of that is coming off the 12 v rail, it's still breaking spec, no?

Yes, and @LAwLz covered that.

[Prysin]

@LAwLz

This thread keeps going, because a certain OP has a lot of salt for a certain product. And a certain fetish for quoting specifications before he understands them and or quote a actually active and valid standard.

 

The PCIe thing IS an issue. I think most of the thread have moved past that. Now it is mostly circle jerking and fanboys going at eachother.

 

Even among those discussions its mostly Nvidiots throwing gasoline at the fire and people coming back at them with proof that Nvidia has had problematic products too.

 

Nvidias products were fixed. BUT HOW lONG DID IT TAKE?! Since nobody seems to have cared during the issues Nvidia had, nobody has been vigilant about their products and noticed when they fixed it.

 

Thing is. We learned of this issue with AMD Rx 480 cards 2 days ago tops.

It is treated and presented by many people like this has been a 2 year long issue that hasnt even been considered to be adressed.

 

Its not a big deal if AMD has a fix for consumers within couple of days or a week at worst. It would take a lot of bad luck to wreck your mobo that fast. Another thing is if this goes on for MONTHS due to legal "slithering" and avoiding of problems and cost of recalls/redesign.

[Misanthrope]

I got an update on my status: I actually went to the PC store yesterday and checked the stock for the 480. The sapphire 8gb reference was out of stock, the gigabyte one had a 30% Price Premium (On top of the usual import Premium of 30% so it was actually 360 USD: good luck with that) but then a wild 1070 appears! Caught!

[Trixanity]

@LAwLz You're just as guilty as everyone else. Don't pass the blame off. It doesn't befit you.

 

And just as a response to your previous reply to me:

You jumped in with both legs and it was glorious. It's not misinformation if you don't actually read it. Don't be mad.

Also, spikes are measured in microseconds, not seconds (if this thread is to be believed). Several seconds of high power draw are not spikes and are still bad, mkay? But let's leave it. No one likes that little tid bit. 

 

When all is said and done: it seems that even AMD's blunder won't burn motherboards. It will at worst destroy your PCIE slot (if my interpretation of the data is correct). It's however more likely to make your board shut itself off to prevent damage as have been demonstrated.

 

I will say this: these shitstorms are fun.

 

Now back to the topic at hand:

 

A solution will be presented sooner or later. It's just a matter of what form it will take.

I still wonder why AMD wired the card to draw power equally from the bus and the aux. It was a bad decision given the power target. The RX 470 will probably do just fine (and the board should be pretty much the same). Given what I've heard of the board design otherwise, I don't think it'll be an issue if power were to be limited on the bus and all the remaining power going through the aux as the board is apparently very over-engineered (blunders aside). So what AMD should have done is limited the bus power to 50-60W and put an 8-pin on it and rebuffed all negative assertions about power consumption and said it was for overclocking but of course avoid all mentions of "overclocker's dream". That way everyone would have been happy and it would be a good pairing with WattMan as well. AMD gambled on the 6-pin and lost.

It's possible they increased stock speeds (relatively) shortly before launch to boost performance causing higher consumption than it was targeted for but for whatever reason stuck with it. I don't quite buy that but it's possible.

 

What's interesting is that the card apparently undervolts really well but doesn't benefit in power consumption at default power target because it simply boosts performance instead due to bigger headroom. However when you boost the power target +20% (max) you get 30W reduced power compared to default voltage at max power target. Disclaimer: these were measured at the wall aka not as accurate.

This could mean that if changes were made to the software, you could allow a lower-than-default power target (at the expense of a little performance) and undervolt it to achieve lower power consumption possibly alleviating the problem although still not a good fix in my opinion.

 

Edit:

Forgot one more point. I also think it was a big gamble to go with GloFo for GPUs and I'm not sure it paid off.

TSMC seems to make superior GPUs and no, not based on Pascal vs Polaris. The initial impression is that GloFo's didn't have the power and performance characteristics necessary for a GPU. I might be wrong here.

However, AMD might also not have had a choice. TSMC seems overtaxed as it is. They can barely keep up with Nvidia and other customers' demands. Add AMD to the mix and everything would probably have fallen apart. Going with GloFo allowed AMD to get a significant amount of stock at launch because apparently GloFo does fuck all but wait for AMD to place orders at empty fabs, so they had the capacity to produce if not the skill.

 

All jokes and exaggerating aside, I do think it might come back to haunt them that they dropped TSMC this round. I wonder who will make Vega and Navi. That would be interesting to know.

[mariushm]

Global Foundries has a basically identical fabrication line to Samsung , it's a "copy smart" design, whatever Samsung changes to their 14nm process Global Foundries copies and adapts to their process and the other way around, so they can make chips in both locations with minimal changes.

TSMC also makes processors for Apple, and they have contracts for lots of quantity which makes TSMC give others lower priority, so if both nVidia and AMD would have used TSMC again...

 

When judging how well this Global Foundries work, also keep in mind that AMD may have binned the chips and put the best ones aside, using just the slightly lower grade chips in the reference models. They could also keep some slightly better chips binned for the custom RX 480 versions which will overclock better and will be cooled better (if the video card manufacturers don't bin the chips themselves)

 

There's rumors AMD has a contract with Apple to deliver a "custom" version of the Polaris 10 , which could very well be just these better binned Polaris 10 chips and maybe with some other features enabled.

 

AMD may also lie or hide the fact that the chip may in fact have more than 36 CU as their documentation says, I personally sincerely doubt they were so confident in the manufacturing process to think they're get such a high percentage of chips with 100% working CUs, or that they didn't choose to increase the number of working chips by disabling non-functional CUs .. I suspect the actual chip has 40 CUs or even more but AMD hides the fact that 4 or more are disabled. Maybe the RX 490 or the Apple custom version will have more than 36 CU working?