AMD: Speculation, Rumors, and Discussion (Archive)

[CSI PC]

Also looks like one youtube reviewer came across the issue when trying to build a low-priced budget PC, I think the price he built down to should had been higher IMO and with a better and more recent budget board, PSU could be debatable.
Although I assume he is thinking of those still on older PCs and GPUs and looking to replace those cards going back several years, still, too cheap with what he aimed at IMO and he is looking at the lower end of mainstream consumers whether right or wrong with this card.
That said it is pretty balanced even though it has that title.
From 1min45secs.

Cheers
Edit:
That said I am seeing posters asking if around $50 - $65 motherboards would be ok, the lower end of that range might have some considerations..

 

[SimBy]

PCI-E slot died with RX 480



https://community.amd.com/thread/202410

Interesting. Most of this posts are freshly created accounts. I'm sure mods will sip through IPs and check whats going on.

OTOH if they are legit accounts, shit storm will ensue.

 

[ninelven]

Also a lot of factory overclocked cards are not pass the PCIe compliance test, and this could be a problem for the whole industry, because for example ASUS don't want to say that the new super ROG card is a nice product, but you might void the motherboard warranty if buy it. These companies make too much money on the factory overclocked product. PCI-SIG might able to do something, but my wife said they just care about the reference products.

Well, I would think most (all at this point) of the custom cards will have a proper power connector, so it shouldn't be an issue.

 

[Kaotik]

That's Polaris 11 vs GTX 950

 

[ninelven]

Still not a good slide to make because it is purposefully misleading. The GTX 950 is a previous gen card (not only that but there is a process transition as well). Now, if you could already buy Polaris 11 at the time the slide was shown, and they knew with reasonable certainty their true competition would not be out for several months... OK. But that wasn't the case, which makes the ethics involved pretty dubious. And in fact, it actually hurts AMD's image more than anything to any objective person who understands the industry. While many companies make this same mistake, AMD's propensity for making these kind of mistakes concerns me a fair amount. It has been going on for a while (last several years), and it is not indicative of a healthy culture within the organization IMHO. I will probably write more on this later, but I think if they could recognize and address this detrimental mentality, it would actually prevent many of the seemingly needless stumbling blocks they continue to encounter.

 

[Gipsel]

edit: Might be late to the party...
Looking at the table 4.1 of PCI Express Card Electromechanical Specification Revision 3.0, it is quite clear and non-negotiable, that it's 5.5A (+/- 8%) for the 12 volt supply via the slot.

I discussed exactly that spec already in your quote of my previous post . And as you can read on top of the table, that are requirements, the slot has to be able to fullfill. In principle, nobody forbids some system designer to include higher quality slots than required by the spec. That would allow the card to use more if that is somehow communicated to the card. Before a higher power draw is negotiated, the card has to limit itself to the requirements set forth in your table. Something like this is actually described in the PCIe base spec (I posted a screenshot of a section with it). A card is allowed to use whatever is higher, the slot power limit set by the BIOS or what the form factor spec (i.e. the PCI Express Card electromechanical spec) allows.
From my quick glance over the ~900 page document of the base spec, it appears to be slightly inconsistent and I agree one can read it as if the "slot power limit value" in the slot capabilities register the BIOS can set for each individual PCIe slot in the system may include not only the power supply through the slot but also the supply through additional power plugs (even when I would deem this a somewhat strange way to specify a "slot power limit" for a mainboard). But frankly I can't be bothered to look in detail at the in total more than 1000 pages that makes the PCIe spec to figure that one out for sure, especially as it probably doesn't bear to many practical consequences.

The practically important question is, what are the actual current limits of the PCIe slots on normal mainboards. I have said already, there are some slots on the market, which barely fullfill the PCIe electromechanical spec. That would mean, with such a slot it can get problematic to draw +50% of the rated current (in case of OC) over the tiny contacts in them, especially if they are worn out a bit already, for instance by repeatedly exchanging the graphics card.

 

[ninelven]

Think about what you are saying.... Does it really make sense for the spec to say it is OK to pull 4x the current through these two wires here (at the Mobo) than from two identical wires directly connected from the power supply to the card? No.

 

[CarstenS]

I discussed exactly that spec already in your quote of my previous post .

Really? I was under the impression you mentioned a fairly old spec, not the most recent document „PCI Express Card Electromechanical Specification Revision 3.0“ from 2013 (which apparently wasn't in need for a 3.1-Update as the Base-Spec from roughly the same date, which has been updated in 2015.

And as you can read on top of the table, that are requirements, the slot has to be able to fullfill.

True, I can read that.

In principle, nobody forbids some system designer to include higher quality slots than required by the spec.

True, in principle they could. Maybe for some higher end boards they actually have higher tolerances. But I hope you would agree that for the very cost-conscious part of the market, especially in the OEM business where every single cent or fraction thereof counts, chances are pretty low that your assumption will hold true as well.

That would allow the card to use more if that is somehow communicated to the card. Before a higher power draw is negotiated, the card has to limit itself to the requirements set forth in your table. Something like this is actually described in the PCIe base spec (I posted a screenshot of a section with it).

I saw (and read that) as well. But as you yourself indicated, this passage is very much open to interpretation. In fact, I could not find any indication that this is superseeding the electomechanical specs and that the negotiable higher power values refer indeed to the power supplied by the slot alone.

When taking into considerung what it described in the section directly before (6.8, Power Budgeting Capability), I am inclined to attribute section 6.9 to the system-level power management wrt to thermals and power budgetting on a system level as well, where it would only make sense that this is not referring to the power delivered via the few thin 12 volt contacts of the slot itself. Especially when considering the wire diameter of a 6- or 8-pin connector and the associcated wire temperatures when unter full load.

I still wonder...

The practically important question is, what are the actual current limits of the PCIe slots on normal mainboards. I have said already, there are some slots on the market, which barely fullfill the PCIe electromechanical spec. That would mean, with such a slot it can get problematic to draw +50% of the rated current (in case of OC) over the tiny contacts in them, especially if they are worn out a bit already, for instance by repeatedly exchanging the graphics card.

Good.

 

[fellix]

Are there any recent motherboards with auxiliary on-board power plugs? In the past some MB vendors did equip their high-end models with an extra Molex connector on the board to feed power to the PCI-E slots directly from the PSU.

 

[silent_guy]

PCI-E slot died with RX 480

https://community.amd.com/thread/202410

Given the amount of motherboards out there that must be at the point of failure, this could just as well have happened with any new or old GPU. E.g., a marginal connector finally breaking down or so. But it's impossible to know for sure, and the reason it's better to be diligent and stay in spec, is to avoid sowing doubt into people's mind.

The only difference is that somehow violating them via the PCIe connector is universally accepted and ok while violating them over PCIe slot is not ok. I wonder who decides which violation is ok and which one isn't.

Have you looked at the difference in cross-section of the wires that go through a PCIe power connector vs the little pins of a PCIe motherboard connector? It's not a very hard case to make that one is close to critical than the other. Especially when, historically, the PCIe power connector has been exposed to abuse more often than the other.

As for proof, there is no problem until proven otherwise.

It doesn't work that way when there's an official spec that's being violated. From now on, every time a motherboard fails with an RX 480, people will suspect the latter to be the cause.

 

[Razor1]

Are there any recent motherboards with auxiliary on-board power plugs? In the past some MB vendors did equip their high-end models with an extra Molex connector on the board to feed power to the PCI-E slots directly from the PSU.

Yeah high end motherboards still have that, but "high" end lol, these cards are not made for a 200+ dollar motherboard. And certainly the bulk of the people buying this card aren't going to having such a motherboard either.

 

[DavidGraham]

Yet everyone seems to be ignoring actual technical knowledge and instead trying to learn something about circuits by looking and at and pasting here that PCIe electrical spec sheet.
The epitome of ridiculousness was probably when I saw a comment saying "It's increasing current output so it's even worse". Because apparently a better solution would be to increase the 12V feed to 14V and screw up all the 12V circuitry?
I'm a bit disappointed at this thread, really.

People with actual technical knowledge about the subject, said this is a serious issue, and wrote an article about it, go figure:
http://www.pcper.com/reviews/Graphi...ns-Radeon-RX-480/Overclocking-Current-Testing

 

[Grall]

Think about what you are saying.... Does it really make sense for the spec to say it is OK to pull 4x the current through these two wires here (at the Mobo) than from two identical wires directly connected from the power supply to the card? No.

Well, actually, yes.

Wires coming from graphics card are significantly higher gauge than the PCIe slot pins (these are like 1.5mm wide or something on that order.) So according to what some people have said, a good quality 6-pin molex plug is good for upwards of 200W of power delivery. If you were to continuously draw that much through the 12V pins out of the PCIe slot, you would undoubtedly fuck up the connector.

 

[silent_guy]

People with actual technical knowledge about the subject, said this is a serious issue, and wrote an article about it, go figure:
http://www.pcper.com/reviews/Graphi...ns-Radeon-RX-480/Overclocking-Current-Testing

I think the money quote is this:

One vendor told me directly that while spikes as high as 95 watts of power draw through the PCIE connection are tolerated without issue, sustained power draw at that kind of level would likely cause damage.

 

[3dilettante]

Interesting. Most of this posts are freshly created accounts. I'm sure mods will sip through IPs and check whats going on.

OTOH if they are legit accounts, shit storm will ensue.

There is a somewhat remote chance of bias or ill will, but a likely scenario is the Law of Large Numbers coupled with confirmation bias.

With regard to breaking PCIe spec:
Will there be problems? Yes.
Will they be because of breaking the spec? There's a universe of causes besides that, but that doesn't mean it doesn't get the blame.

With thousands of systems, something is going to go wonky somehow.
Before the story about Polaris exceeding spec coming out, a lot of people would have shrugged it off as a bad motherboard--which it actually could be since having a device wobbling at the margins of spec isn't necessarily an insta-kill.
It could even be blamed on a card needing an RMA in some situations.
With the story out, the assumption on everyone's mind is a bad architecture, and people with new accounts feel they have a reason to contribute when they would have been silent before.

 

[ninelven]

Well, actually, yes.

EDIT: OK, I see what happened here. You misinterpreted my post exactly backwards....

 

[xEx]

Super disappointed with the results. not worth buying a 100 dollars liquid cooler for the reference card.

 

[Deleted member 13524]

Still not a good slide to make because it is purposefully misleading. The GTX 950 is a previous gen card

Curious. What card is currently replacing the GTX950 in nVidia's portfolio?
You should e-mail nVidia with that information because their website doesn't show any card newer than the GTX950 for the same price and performance range. They might thank you and update their portfolio based on your information.

People with actual technical knowledge about the subject, said this is a serious issue, and wrote an article about it, go figure:
http://www.pcper.com/reviews/Graphi...ns-Radeon-RX-480/Overclocking-Current-Testing

It seems neither you or silent_guy actually read the article. The motherboard manufacturer's rep said there would be problems if there is a sustained power draw of 95W, though he didn't state after how long it would actually make any perceivable damage on the motherboard.
Regardless, the sustained power draw is 82-83W, not 95W. One is 7W above the standard, the other is 20W.

Unless you're somehow trying to blame AMD on consequences of overclocking, which would be funny as hell.
Not unexpected coming from the haters circlejerk of pages 171-172 though.

 

[seahawk]

That's Polaris 11 vs GTX 950

Which must be wrong, because no test ever has shown a standard clocked 950 to consume anything close to 140W.

Curious. What card is currently replacing the GTX950 in nVidia's portfolio?
You should e-mail nVidia with that information because their website doesn't show any card newer than the GTX950 for the same price and performance range. They might thank you and update their portfolio based on your information.


It seems neither you or silent_guy actually read the article. The motherboard manufacturer's rep said there would be problems if there is a sustained power draw of 95W, though he didn't state after how long it would actually make any perceivable damage on the motherboard.
Regardless, the sustained power draw is 82-83W, not 95W. One is 7W above the standard, the other is 20W.

Unless you're somehow trying to blame AMD on consequences of overclocking, which would be funny as hell.
Not unexpected coming from the haters circlejerk of pages 171-172 though.

If you play with Wattman is quickly is around 95W. And then there old motherboards and cheap motherboards...

 

[Malo]

It seems neither you or silent_guy actually read the article. The motherboard manufacturer's rep said there would be problems if there is a sustained power draw of 95W, though he didn't state after how long it would actually make any perceivable damage on the motherboard.
Regardless, the sustained power draw is 82-83W, not 95W. One is 7W above the standard, the other is 20W.

Unless you're somehow trying to blame AMD on consequences of overclocking, which would be funny as hell.
Not unexpected coming from the haters circlejerk of pages 171-172 though.

In that article he's testing stock settings as well and gets similar results

This graph shows that result, running Metro: Last Light at 4K with the Radeon RX 480 at stock settings. The green line is the amperage being used by the +12V on the motherboard PCI Express connection and the blue represents the same over the 6-pin power connection. The motherboard is pulling more than 6.5A through the slot continuously during gaming and spikes over 7A a few times as well. That is a 27% delta in peak current draw from the PCI Express specification. The blue line for the 6-pin connection is just slightly lower.