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

[Dackzy]

1 minute ago, SteveGrabowski0 said:

I wouldn't put it past AMD to put out a BIOS and driver that limits power consumption to 150W after they have already gotten all the reviews out of the way at 165W.

They don't need to limet the power draw, they just need to put more load on the 6 pin and a 6 pin can handle way more than those 75w.

[zMeul]

27 minutes ago, Sintezza said:

But i´m interested in how they all measured the exact powerdraw from the pci-e slot.

where have you been all this time?!

read the article, they used osciloscope and other measuring equipment

[Majestic]

3 minutes ago, Trixanity said:

Can't we petition the admins to give him that title? That would be great.

Please leave the topic if all you're doing is attacking members.

[laminutederire]

1 minute ago, Majestic said:

The distinction, I repeat myself, is this:

 

GTX 1080

 

RX-480

 

 

You see the GTX1080 exceeding the 75W enveloppe anywhere? No. So you're not making the distinction. And you're missing the entire point of the post by not doing so.

Gosh the 1080 doesn't strain the motherboard but it strains the 8pin, and te same reasonement on the 8 pin strains just say that the issue is on the average. How hard is it to understand?

I'm just giving an explanation as to why the spikes aren't what matters, but the average matters. All of that to say that the issue with the RX 480 problem is that it has an average value above what's authorized. I don't see how it is contradictory with anything.

I don't see what is your point, but you don't seem to see mine either.

4 minutes ago, djdwosk97 said:

No, you can't infer that at all. First of all, the 1080 is still relatively new to the market and the people who have one more than likely also have a quality PSU. Secondly, you can't infer a single thing about it damaging the motherboard because the 1080 doesn't overdraw from the motherboard but rather the PSU.

The distinction is important because traces are a much lower gauge then the wires of a PSU. Continuous is 10-20w above while peak is up to 155w above -- 155w, even for an instant, may be enough to burn out the very small traces that are on a motherboard -- No one here knows enough about the size of the traces and the duration of the peak load to determine one way or another whether or not it's more than just a technicality of breaking spec or if it's actually going to risk damage to the motherboard. 

That's literally what I'm saying... they need to move the strain they need to the 6 pin connector.

[Dackzy]

2 minutes ago, laminutederire said:

That's literally what I'm saying... they need to move the strain they need to the 6 pin connector.

The same thing I have been saying in this thread....

[laminutederire]

13 minutes ago, djdwosk97 said:

No, you can't infer that at all. First of all, the 1080 is still relatively new to the market and the people who have one more than likely also have a quality PSU. Secondly, you can't infer a single thing about it damaging the motherboard because the 1080 doesn't overdraw from the motherboard but rather the PSU.

The distinction is important because traces are a much lower gauge then the wires of a PSU. Continuous is 10-20w above while peak is up to 155w above -- 155w, even for an instant, may be enough to burn out the very small traces that are on a motherboard -- No one here knows enough about the size of the traces and the duration of the peak load to determine one way or another whether or not it's more than just a technicality of breaking spec or if it's actually going to risk damage to the motherboard. 

 

I infer because of that explanation I posted earlier:

Quote

don't deny there is an issue, I'm just saying that the spikes aren't what we need to focus on.

The spikes shouldn't matter I think. The idea is that a power is measured in Watts, 1W is equal to 1J/s, therefore if you have a huge power in a very very limited time, you'll still will have produced the same energy, and that's the energy you need to look at when yiu want to know if everything is going to melt. As long as the spikes are short enough and not too powerful, there isn't too much energy to be dissipated before making the wires to melt.

I think their solution is to change the settings for it to draw the power from the PSU, since an 8 pin isn't so much powerful compared to a 6 pin, even if they have a massive rating difference.

 

 

[Majestic]

1 minute ago, laminutederire said:

Gosh the 1080 doesn't strain the motherboard but it strains the 8pin, and te same reasonement on the 8 pin strains just say that the issue is on the average. How hard is it to understand? I'm just giving an explanation as to why the spikes aren't what matters, but the average matters. All of that to say that the issue with the RX 480 problem is that it has an average value above what's authorized. I don't see how it is contradictory with anything.

I don't see what is your point, but you don't seem to see mine either.

No you're goalpost shifting and are using non-sequiturs as 'proof'. 

 

You are arguing because the fluctuating draw on the PCI 6+2 pins are not an issue, therefor it won't be an issue on the PEG connector either. Non-flippin'-sequitur. Because those are two completely seperate things. Read up on EMC please, and how those spikes in ampere can cause noise on your system. It's not an issue on the PSU because it's got massive caps and it's shielded from the system. It does matter on the mainboard circuit.

 

[Sintezza]

15 minutes ago, zMeul said:

where have you been all this time?!

read the article, they used osciloscope and other measuring equipment

 

Yeah like i said i have revoked my claim on it being totaly bullshit.

However my opinnion towards that this is realy a serious issue, and users being in real danger for killing their boards, still has not changed.

the Asrock OC formula isnt a cheap board, so thats why i´m interested in what the actual cause of the burned traces on the motherboard was.

Its of course obvious that traces on a motherboard in general are way thiner then wires from a psu.

And there for can handle way less current.

But those cards are getting fully tested to meat pci-e and other standard before they left the factory.

Keep in mind that AMD does not produce those cards them selfs, they have a very reputable board partner for that.

 

I´m currently looking up as much info on this as possible.

But all i can say, i have inspected allot of motherboards over the years.

And i have seen something like what happened with Toms never happen, with a decent motherboard.

[laminutederire]

2 minutes ago, Majestic said:

No you're goalpost shifting and are using non-sequiturs as 'proof'. 

 

You are arguing because the fluctuating draw on the PCI 6+2 pins are not an issue, therefor it won't be an issue on the PEG connector either. Non-flippin'-sequitur. Because those are two completely seperate things. Read up on EMC please, and how those spikes in ampere can cause noise on your system. It's not an issue on the PSU because it's got massive caps and it's shielded from the system. It does matter on the mainboard circuit.

 

I've updated my precedent post explaining why the fact it spikes above the limit doesn't mean it will melt anything.

[Misanthrope]

13 minutes ago, Dackzy said:

They don't need to limet the power draw, they just need to put more load on the 6 pin and a 6 pin can handle way more than those 75w.

That's still not ideal: It just wouldn't affect much more than the pci cable and the card, maybe the PSU.

 

The only true solution would be for AMD to limit the voltage to a hard 150 cap or to order a recall of all cards. I suspect they will do neither and just downplay it and hope it goes away.

 

3.5 vram did went away too after a while.

 

[spartaman64]

4 minutes ago, laminutederire said:

I've updated my precedent post explaining why the fact it spikes above the limit doesn't mean it will melt anything.

with a quality motherboard it is probably fine but i can see this being a problem with people who buy 50 dollar h81 motherboards 

[Majestic]

3 minutes ago, laminutederire said:

I've updated my precedent post explaining why the fact it spikes above the limit doesn't mean it will melt anything.

Nobody except people who aren't reading the OP is talking about melting, it's interference.

However with OC'ing, the 100W average draw might damage the mainboard. I'm not sure about 86W.

 

Oh, and that 75W is over the complete interface, not per slot. So crossfire means it's doubling the max. tolerated draw.

[Dackzy]

1 minute ago, Misanthrope said:

That's still not ideal: It just wouldn't affect much more than the pci and the card, maybe the PSU.

 

The only true solution would be for AMD to limit the voltage to a hard 150 cap or to order a recall of all cards. I suspect they will do neither and just downplay it and hope it goes away.

 

3.5 vram did went away too after a while.

 

dude just stop. A 6 pin can handle 150w and a 1080 also takes more than the 150w from the single 8 pin, but I don't see you saying anything about that. It will be a ideal solution since it will be safe.

[Trixanity]

4 minutes ago, Majestic said:

Please leave the topic if all you're doing is attacking members.

Come now, attacking members? Please.

 

Also, it seems you've abandoned all reason if you point to one quote and saying "all you do". You can't just go about making up statistics.

 

And that's not attacking. If that's attacking, then you've clearly not spent enough time on the internet and certainly not been around people for any length of time. It's called human interaction.

 

If we include all my posts in this topic and you point out one instance of me providing some playful banter, how can that be construed as representative of my behavior in this thread?

 

Let's not turn this into a bastion for SJWs. That would be a tragedy.

Let's return to the topic at hand, shall we?

[79wjd]

1 minute ago, Dackzy said:

dude just stop. A 6 pin can handle 150w and a 1080 also takes more than the 150w from the single 8 pin, but I don't see you saying anything about that. It will be a ideal solution since it will be safe.

With a decent PSU it's not a concern, but a 480 is a mainstream card and will likely end up in quite a few systems with fairly questionable PSUs. 

[zMeul]

9 minutes ago, Sintezza said:

But those cards are getting fully tested to meat pci-e and other standard before they left the factory.

apparently they were not, or so both AMD's Raja and Robert claimed - it was news to them that RX480 draws more than PEG PCI-SIG specs, 75W

it's somewhere in this thread, don't ask me to quote it .. go back and read all pages 

 

Quote

Keep in mind that AMD does not produce those cards them selfs, they have a very reputable board partner for that.

it doesn't matter who manufactures their cards, Foxconn or some other chinese FAB

it's AMD that designed them, and it's AMD's duty to inspect the manufacturing process

 

Quote

decent motherboard

you need to take into account what market segment AMD is targeting this card at

it's not the people who'll spend hundreds of $ on medium to hi-end mobos, but to the people who'll most likely buy cheap mobos with few power phases and thin PCBs

[mariushm]

According to the Tom's Hardware, the VRM feeding the GPU chip is a six phase design, where three phases are connected to the pci-e 6 pin connector, and three phases are connected to the motherboard slot.

 

If that's the case, on average, there should be an even current draw from both sources. It's just bad luck that at some points for a few ms of time, the gpu boosts its frequency and therefore demands more power and it just so happens those three phases connected to the motherboard slot are active at that moment, pulling more current than the maximum recommended by the standard.

 

AMD should have connected only two power phases to the motherboard slot and the other four to the pci-e connector... but my guess is AMD initially designed the board around the 1080 Mhz base frequency, where maybe the total power consumption averages at around 150w instead of 165w as it is now, and probably with this configuration the current draw from motherboard would have been less than 80w average.

 

Some OC-aimed non reference designs are rumored to come with a 8pin pci-e connector and possibly one pci-e 6 pin connector reserved just for memory overclocking. So, you'd have 150w + 60w for the gpu chip, and up to 75w for memory alone, plenty of reserves to try overclocking to 1500 Mhz or thereabouts.

 

@Sintezza : a very simple method would be to use pci-e x16 riser cards / extenders, here's an example of such cable: https://www.amazon.com/Express-Riser-Extender-Flexible-Extension/dp/B008BZBFTG

 

 

You have the pci express slot pinout here: https://en.wikipedia.org/wiki/PCI_Express#Pinout

Basically 5 out of the first 6 pins in the slot are 12v , so they can simply make a cut in those wires and connect a measurement tool (to log the current and voltage) between those pins of the riser cable and wires going into the motherboard slot (or they could just work around the motherboard and connect those pins to the power supply but then the measurements wouldn't be fair and they wouldn't notice if the current consumption affects the motherboard in some way, let's say for example if they want to measure the temperature of the motherboard surface for hot spots).

The equipment they use is quite high end and capable of measuring the instantaneous voltage and current at least a few hundred times a second, which gives them the ability to show when the card pulls much more current than normal for very brief moments.

 

Of course, just by introducing some equipment between video card and motherboard/power supply introduces some power losses but with good quality equipment these can be ignored (the deviations/errors are usually less than 0.05% from real values so maybe a few tens of a watt error)

[laminutederire]

2 minutes ago, djdwosk97 said:

With a decent PSU it's not a concern, but a 480 is a mainstream card and will likely end up in quite a few systems with fairly questionable PSUs. 

It's not that mainstream. it'll be popular but it remains the 1080p gaming best cards (since it can easily do 1080p@60 on ultra settings, for the lowest price possible), but the 470 should be even more widespread.

Doesn't mean they shouldn't fix it though.

[Dackzy]

1 minute ago, djdwosk97 said:

With a decent PSU it's not a concern, but a 480 is a mainstream card and will likely end up in systems with fairly questionable PSUs. 

Even a lot of prebuilds would have a PSU that would be fine, it is really only those really cheap china ones that will have a problem and if you have one of those then there is a bigger chance that it will kill your system.

[Sintezza]

2 minutes ago, zMeul said:

apparently they were not, or so both AMD's Raja and Robert claimed - it was news to them that RX480 draws more than PEG PCI-SIG specs, 75W

it's somewhere in this thread, don't ask me to quote it .. go back and read all pages 

 

it doesn't matter who manufactures their cards, Foxconn or some other chinese FAB

it's AMD that designed them, and it's AMD's duty to inspect the manufacturing process

 

Yeah i know that Raja picked it up as news, or atleasts thats posted in this thread.

Like i said, if we are going to hear more about it, and if we going to see more sources, or people comming up with burned out motherboards.

Then we know that its a serious issue.

If we dont hear anything from it again, then it doesnt seem to be as big of an issue as blown out of porportion right now.

[Misanthrope]

4 minutes ago, Dackzy said:

dude just stop. A 6 pin can handle 150w and a 1080 also takes more than the 150w from the single 8 pin, but I don't see you saying anything about that. It will be a ideal solution since it will be safe.

8 Pin can handle 150 and possibly more. 6 pin should be 75 watts. Can you exceed 75? Sure. In most cases it will be fine. It's still fucking wrong for AMD to sell this card as a 150tpd 6 pin only card when they fucking knew it used way more. It's basically dishonest fucking bullshit to tell us "This card only needs a 6 pin!" when it clearly doesnt.

 

Yes Nvidia does it too, they're fucking assholes too. What you're advocating is to just get used to both companies flat out fucking lying and breaking specs. No, everyone should be called on it. In this specific thread we talk AMD but that's no excuse for Nvidia for you to pretend is bias. Is not an AMD bias is just staying on fucking topic.

[Prysin]

@Majestic

It was the 980Ti ref.

 

http://www.tomshardware.co.uk/nvidia-geforce-gtx-980-ti,review-33214-7.html

 

spikes of 80w under gaming load. Sure, not constant. I think PCPer also mentioned it being VERY bad under OC in one of their weekly podcasts.

[Dackzy]

2 minutes ago, Misanthrope said:

8 Pin can handle 150 and possibly more. 6 pin should be 75 watts. Can you exceed 75? Sure. In most cases it will be fine. It's still fucking wrong for AMD to sell this card as a 150tpd 6 pin only card when they fucking knew it used way more. It's basically dishonest fucking bullshit to tell us "This card only needs a 6 pin!" when it clearly doesnt.

 

Yes Nvidia does it too, they're fucking assholes too. What you're advocating is to just get used to both companies flat out fucking lying and breaking specs. No, everyone should be called on it. In this specific thread we talk AMD but that's no excuse for Nvidia for you to pretend is bias. Is not an AMD bias is just staying on fucking topic.

I only brought the 1080 up because nobody seemed to make a thread about that lie they did. Yes it is wrong and I am sure that on the none ref cooled cards that we will see a 8 pin and it is not like it is the end of the world.

[zMeul]

PCI Express Card Electromechanical Specification Revision 3.0 - July 2013 http://download.csdn.net/detail/jianjunzaixian/9229521

[zMeul]

10 minutes ago, Misanthrope said:

"This card only needs a 6 pin!"

of course it only needs 6pin when the rest is drawn via the PEG