AMD: Speculation, Rumors, and Discussion (Archive)

Not having aging compensation in the architecture at all would per AMD lead to reduced clocks, which in this case would seemingly only help efficiency. That's not to say that having in the architecture but kludging it off might not cause problems.
Boot time calibration would leave some power efficiency on the table, but at the same time in theory not having it at all would leave Polaris where every other board happens to be.

However, depending on the implementation of these methods, they might have been turned off by reverting to some higher catch-all voltage offset or turning them off leaves a reduced set of VRM or clock settings if final clocks and voltages were designed to be set by final offsets provided by those measures. If not architecturally required, AMD might have rushed a BIOS with some hacky settings hard-coded.

Some of the undervolting experiments (small sample size) seem to show a fair amount of cushion that would presumably have been taken advantage of if all these fancy measures were successfully used.
This might be a continuation of AMD's tendencies with every other first iteration of a product in the last 3-4 generations of CPUs, APUs, and GPUs. Their advanced features do depend more on fine-tuning, bug fixing, validation, and good physical characterization. AMD hasn't gotten that right on a first try for years.
That's not necessarily a bad thing if handled differently. In the old days Polaris would have been a comparatively quiet pipe cleaner launch, and it would have provided feedback for a second round of bug-fixed products. AMD seems to have put about as much effort into advancing GCN as it if were a pipe-cleaner, its marketing and executive bluster aside.

 

I think that's whole-system.

 

Those were full system power draw numbers. If you check the videos, the numbers are shown in power-meters directly connected to the PSUs.

95W spike is okay, 95W continuous is the result of overclocking therefore not AMD's responsibility.

 

The rep quoted the 95w figure because it's what the X480 was drawing!
And yes, I SHOULD and WILL blame the consequences of OCing on AMD, when other cards stick to spec just fine when OC'ed.

 

So a guy writing an article about how to run a card out of spec is shocked when he finds the card running out of spec?

 

It's full system consumption, not just the video card, and that's actually quite low number for it
(and yes, it has been shown live, too, at least on 2 different occasions)

 

95W was attained through overclocking.
You clearly didn't read or comprehend the article.

What company takes the blame from homemade overclocking?
Is this a troll post?

 

Who's the responsible of the PCI-E certification?

 

I remember the days when overclockers had to fiddle with jumpers, pencil mod their CPUs, lapped their dies, installed coolers that could and would take parts of their fingers off, and did so with the knowledge they could very readily turn part of their rig into magic smoke.
I don't quite reach back to the days of replacing clock crystals.

It's probably a good idea to implement some secondary fail-safes or additional warnings for serious overdraw in an OC situation, but there's a reason why warranties turn to magic smoke the instant you take anything past spec.

That there is some more mild overdraw at stock is iffy to me.

 

Is there a disclaimer for using wattman? I heard you can just set clocks and voltages to unrealistic numbers in it. I don't think +50%power limit is nearly as bad as setting core voltage to >2v..

 

Hmm I think its done internally and then the results are sent to PCI-SIG for them to verify on samples given to them.

 

In the articale a rampage extreme were having trouble sustaining the voltage of the 12v rail....a fking rampage extreme...we would have to wait to see how a cheap Mobo handles a CF config but in the article when you OC the 480 it goes 45% out of spec.

 

Totz, the 480 is a very decent piece of GPU silicon, but it's AMD that screwed up on the pre-launch messaging, setting massively wrong expectations, and the PCIe power issue, where it's a clear cut violation of some spec that may or may not have consequences.

It's completely in line with expectations that you think we should give them a pass on that.

 

If it is so, someone should be fired...

 

And yeah finally people on forums saying that the RX are killing boards... lol this is going to be viral

 

the power draw issue, Can open up for a huge class action suit.
amd dont need it

 

They mostly care about the communication part of the standard not so much about the power consumption. Its mainly to give list on compatible devices to OEMs

 

How many reports, and how many had this happening while running at stock?
I haven't seen reports of millions of motherboards blowing up, so how many are in this class?

 

They should have been fired months ago when the "Overclockers Dream" turned out to be a dud .....

 

There is a difference between PCI express spec and ATX12V V2.x spec and also the Molex connectors, the latter two specs are not applicable to the PCI express slots.
Therefore it is not unexpected that the PCI express slot is close to the sustained 5.5A current specification as it is integral to PCI Sig for 66W through the 12V, and different to the ATXV12 spec and standards that can be substantially higher IF following the component standards I mentioned earlier.

If looking at 150W and higher, these fall into the 150 W-ATX Specification/250W to 450W configurations part of the ATX12V publications, it is not just the PCI express you look at.
And in these publications it also specifies HCS that you will not find on lower mainstream boards and PSU.

Again there are two primary considerations;
1. The PCI Express slot that has tighter values in terms of manufacturers implementations giving 66W over the 12V and 9W from 3.3V - yeah 85W sustained and shown by Tom's Hardware and some extent at PCPer is eye raising and compounded as it should be acceptable to some systems but not all, also would not want it to be any higher though as it is already pushing it to the extremes.
2. The ATX12V 24-pin connector mostly in context of 2x480, but also potentially when looked at from the budget motherboard and PSU focus with a single 480.
2. a) This is shared with all PCI express slots and other devices that do not have their own separate power.

OCing or adjusting the power target (even without OC) would be applicable to both situations and not necessarily just budget boards.
Only the ATXV12/Molex connectors could be deemed to have a rating closer to 8A per 12V contact for the PEG connectors (total 192W for 6-pin and 288W for 8-pin), and a bit less at 6A for the ATX 24-pin 12V contacts (gives total 144W) if it is not HCS (that was specified under ATXV12 v2.2) - There is also a higher threshold for moderate sustained duration bursts beyond max current for the ATX 24-pin and PEGs.


Just to raise a separate point,custom AIB will not overcome this problem by implementing an auxiliary PEG 8-pin/2x6-pin/etc configuration unless they also re-define the power distribution and ease preassure off the PCI Express slot, this is going to be critical with AIB partners as it is more than likely they will at a minimum increase the power targets and possibly some small OC.

CHeers
Edit:
And to re-iterate, if the 480 is left in standard form it will be fine for most including many mainstream consumers (IMO just needs a careful look at when matched to the lower end of mainstream motherboards and PSU).
That means not increasing power target and not increasing clocks, and using 2x480 only with very good motherboard and PSU.