fallout for RX480 overdrawing power from PCIe slot - AMD got it's card delisted from PCI-SIG compliance list

[zMeul]

source: https://pcisig.com/developers/integrators-list

via: https://www.techpowerup.com/225081/high-pcie-slot-power-draw-costs-rx-480-pci-sig-integrator-listing

 

 

Quote

AMD's design of the Radeon RX 480 graphics card, which draws over 75W of power from the PCI-Express x16 slot, has cost it a product listing on the PCI-SIG Integrators List. The list is compiled for hardware devices implementing the various PCI-Express specifications to the letter. The RX 480 is off-spec, in that it overdraws power from the slot, as the card needs more power than what the slot and the 6-pin PCIe power connector can provide while staying within specs. According to these specs, the slot can provide up to 75W of power, and the 6-pin connector another 75W. The RX 480 was tested to draw more than this 150W power budget.

 

AMD cannot display the PCI-Express certification logo on the product, box or marketing materials

this, however doesn't mean that AIB partners can't retain the PCIe certification by designing their own boards in compliance with PCI-SIG's tech specs

 

in a world where certifications worth more than the actual product it's covering, system integrator and OEMs will think twice before allowing AMD's RX480 in their systems - why AMD did this is simply not comprehensible .. did they actually though no one would notice?!

 

Quote

PCI-SIG Integrators List includes all products that have successfully completed the rigorous testing procedures of the Compliance Workshop

 

---

 

where PCI-SIG stands on this?

well .. Al Yanes (president of the PCI-SIG) said at IDF that PCI-SIG doesn't consider creating a subset of tests for this particular problem, even after the RX480 blunder

well ... fuck you Mr Yanes  if this wasn't discovered, the RX480 would've kept it's certification - and if that isn't important, then why does the certifications exist in the 1st place

or are they silently giving AMD only slap on the wrist for their fuck up?

Quote

Behind the scenes, the high current draw of the Radeon RX 480 certainly seems quite to have caused greater excitement, was how to listen between the lines. Yanes emphasized, among other things, that standards for this are there to be respected.

 

source: http://www.heise.de/newsticker/meldung/Pikantes-Politikum-Die-Radeon-RX-480-und-die-PCIe-Spezifikation-3297029.html

Edited August 17, 2016 by zMeul

[themctipers]

Go global warming!

*argue about it being fake or something under me.*

[79wjd]

I thought 16.7.1 fixed the issue at stock speeds (and thus should comply with PCI Sig). Although I honestly don't see having a PCIE logo on the box meaning anything at all considering any user who knows what PCIE is will know it's still compatible and anyone who doesn't, won't know any difference. 

[TheRandomness]

Well

shit. 

At least the 470 is basically a 480 without this issue (with a little less performance).

[zMeul]

Just now, djdwosk97 said:

I thought the driver updated fixed the issue at stock speeds (and thus should comply with PCI Sig). Although I honestly don't see having a PCIE logo on the box meaning anything at all considering any user who knows what PCIE is will know it's still compatible and anyone who doesn't, won't know any difference. 

the driver update doesn't fix the card's design - if you buy the card and install the included original driver, the card will draw more from PCIe slot

 

this isn't about the end-user as is more about OEMs and other system integrators that care a great deal about certifications

[porina]

7 minutes ago, djdwosk97 said:

I thought 16.7.1 fixed the issue at stock speeds (and thus should comply with PCI Sig). Although I honestly don't see having a PCIE logo on the box meaning anything at all considering any user who knows what PCIE is will know it's still compatible and anyone who doesn't, won't know any difference. 

The original problem was talking too much power from the slot. The driver bought that in line, but shifted it to the 6 pin connector, pushing that out of spec instead. This was consider practically ok as the 6 pin connector has headroom above spec, but the PCIe slot was right on its limit.

 

I wonder, if AMD made the "compatibility mode" the default in the driver, would that be sufficient to regain certification? That mode ran the card at lower performance to bring things within spec limits.

[79wjd]

1 minute ago, porina said:

The original problem was talking too much power from the slot. The driver bought that in line, but shifted it to the 6 pin connector, pushing that out of spec instead. This was consider practically ok as the 6 pin connector has headroom above spec, but the PCIe slot was right on its limit.

 

I wonder, if AMD made the "compatibility mode" the default in the driver, would that be sufficient to regain certification? That mode ran the card at lower performance to bring things within spec limits.

Overdrawing from the PCIE connector wouldn't break PCI SIG compliance (I don't think). 

[zMeul]

1 minute ago, djdwosk97 said:

Overdrawing from the PCIE connector wouldn't break PCI SIG compliance (I don't think). 

yes it will, there's a certification for that too 

[porina]

http://www.tomshardware.com/reviews/amd-radeon-rx-480-polaris-power-fix,4668.html

 

Just looked back at earlier testing, even the compatibility mode still takes more than 150W total. Did pcper also remeasure? I doubt it'll be much different...

[AresKrieger]

3 minutes ago, djdwosk97 said:

Overdrawing from the PCIE connector wouldn't break PCI SIG compliance (I don't think). 

This is true but they would need to do a proper retest this time with some sort of overseer, additionally that breaks a different specification, needless to say they should have just put the god damn 8pin on it.

[DXMember]

so they de-list the reference model only or do they delist Polaris 10 chip as a whole?

Don't they like release a revision 2 of the board that ships with vBIOS ­µcode update?

what are the consequences of being delisted from PCI-SIG, some system integrated just look at it with a raised eyebrow?

[Misanthrope]

1 hour ago, djdwosk97 said:

I thought 16.7.1 fixed the issue at stock speeds (and thus should comply with PCI Sig). Although I honestly don't see having a PCIE logo on the box meaning anything at all considering any user who knows what PCIE is will know it's still compatible and anyone who doesn't, won't know any difference. 

Yes but I'm pretty sure the specification doesn't allows patching after the fact with a driver. They'd require a complete recall and for AMD to flash the bios of the cards so this is done at the hardware level  (Not really but with the default bios in the card regardless of OS drivers, you know what I mean) so it's compliant with the specification even in the event of people installing the card with no internet at all and with the drivers provided in box.

 

Or maybe just the drivers in the box though arguably that still leaves the card potentially overvolting if somehow using other drivers legitimately i.e. Linux open source drivers for AMD are actually usable but would probably not include the fix.

 

Regulations are intentionally picky like that.

[DXMember]

1 hour ago, AresKrieger said:

This is true but they would need to do a proper retest this time with some sort of overseer, additionally that breaks a different specification, needless to say they should have just put the god damn 8pin on it.

jlhgurfoiuyhedfwhiulfwe lhiufew ihlfuewahifueawlhiulafwehiuaflew huifaewfwehuai fwaeihufewahliu\

it already has an 8-pin

[Prysin]

1 hour ago, zMeul said:

the driver update doesn't fix the card's design - if you buy the card and install the included original driver, the card will draw more from PCIe slot

 

this isn't about the end-user as is more about OEMs and other system integrators that care a great deal about certifications

althought that being said, W10 will occasionally ninja install drivers if they are WHQL certified..... So AMD could have the fix be pushed by MS as a ninja update.

[Prysin]

10 minutes ago, DXMember said:

so they de-list the reference model only or do they delist Polaris 10 chip as a whole?

Don't they like release a revision 2 of the board that ships with vBIOS ­µcode update?

what are the consequences of being delisted from PCI-SIG, some system integrated just look at it with a raised eyebrow?

reference model only from what i can see in the article

[AresKrieger]

 

13 minutes ago, DXMember said:

jlhgurfoiuyhedfwhiulfwe lhiufew ihlfuewahifueawlhiulafwehiuaflew huifaewfwehuai fwaeihufewahliu\

it already has an 8-pin

No, it doesn't

[DXMember]

1 minute ago, AresKrieger said:

 

No, it doesn't

that's a mini 8-pin

l2e

[Misanthrope]

8 minutes ago, Prysin said:

althought that being said, W10 will occasionally ninja install drivers if they are WHQL certified..... So AMD could have the fix be pushed by MS as a ninja update.

Still leaves Linux users open to issues.

 

Arguably it's like a microscopic minority: 5% or so of all PC users and of them maybe 5% of them have a combination of and old/crappy motherboard and a 480 so that's what, 0.005%?

 

Thing is specifications are built like that to avoid liability. For them 0.005% might as well be 20% since they'd be liable for any and all damages if they don't revoke the certification. So even in this once in a blue fucking moon cases they have to cover their asses legally, at least that's what I think it's going on here.

[DXMember]

Just now, Misanthrope said:

Still leaves Linux users open to issues.

 

Arguably it's like a microscopic minority: 5% or so of all PC users and of them maybe 5% of them have a combination of and old/crappy motherboard and a 480 so that's what, 0.005%?

 

Thing is specifications are built like that to avoid liability. For them 0.005% might as well be 20% since they'd be liable for any and all damages if they don't refuse certification. So even in this once in a blue fucking moon cases they have to cover their asses legally, at least that's what I think it's going on here.

linux pplz probly don't play Metro 2033 4k ultra, do they?

[Misanthrope]

2 hours ago, DXMember said:

linux pplz probly don't play Metro 2033 4k ultra, do they?

Hey if I was smart about it and could somehow burn my house down to sue AMD and the certification board by running the 480 on linux with a 7 year old motherboard, I'd sure tell people to leave the house, have the fire dept. on speed dial and begin Metro 4k! *

 

 

 

 

 

 

*The feelings of Misanthrope here are in no way representative of any encoruagement to commit insurance fraud, we at the LTT community frawn upon and such illegal activity.

[Prysin]

4 minutes ago, Misanthrope said:

Still leaves Linux users open to issues.

 

Arguably it's like a microscopic minority: 5% or so of all PC users and of them maybe 5% of them have a combination of and old/crappy motherboard and a 480 so that's what, 0.005%?

 

Thing is specifications are built like that to avoid liability. For them 0.005% might as well be 20% since they'd be liable for any and all damages if they don't revoke the certification. So even in this once in a blue fucking moon cases they have to cover their asses legally, at least that's what I think it's going on here.

true, it reeks of ass covering.

 

3 minutes ago, DXMember said:

linux pplz probly don't play Metro 2033 4k ultra, do they?

can Linux even do 4k?

[DXMember]

2 minutes ago, Prysin said:

can Linux even do 4k?

ha

ha ha

[samcool55]

Well what a bunch of idiots at the pci sig thing then >_>

It's their job to call out manufacturers for such thing when they are testing their stuff before they approve it.

 

Is there a place where you can check their test results about the products they certified?

[captain cactus]

In that case also remove the PCI-SIG certification from the R9-295X2. It draws 500W or more from 2 8-pins, pulling them out of spec as well (quite massively, actually, like, 125W too much).

 

It's not so much about what the card's pulling (you can easily pull 150W or more from a 6-pin with a good PSU), but rather the fact that it's not a standard 6-pin layout. The top-middle pin is supposed to be a sensing pin sensing if the plug's inserted correctly (and won't turn on the GPU if it senses a bad mount/dead cable), but on the reference RX480 it's another ground pin. That's what's out of spec about it.

[Misanthrope]

Just now, samcool55 said:

Well what a bunch of idiots at the pci sig thing then >_>

It's their job to call out manufacturers for such thing when they are testing their stuff before they approve it.

 

Is there a place where you can check their test results about the products they certified?

They could be. Or AMD could have sent an "engineering sample" to them that was quite tamed down instead of the more voltage hungry retail unit.

 

They should be testing the final products randomly off the production line though so It's probably safer (as it usually is) to attribute this to incompetence than to malice.