AMD: Speculation, Rumors, and Discussion (Archive)

[Razor1]

As long as there is no stability issues at those volts, that would be a very good option for AMD!

 

[renderstate]

So users that just got a 480 know better than AMD themselves how to run HW they had in the lab for 6+ months?

I don't know if I should be skeptical of these undervolting reports or think lowly of AMD lack of understanding of their own product!

I prefer to remain skeptical for the time being

 

[flopper]

Its very obvious AMD pushed the clocks/voltage at last minute (after PCIe certification) to be more competitive but they had no time to redesign the boards (already in production). And now they pay the consequence...

That opens up for a huge lawsuit.
I am sure dave baumann can settle that

 

[revan]

It's just luck of the draw. You know, not all cards are made even. Some will work with less voltage than other (depending of ASIC quality AFAIK)
Anyway, Damien proved that undervolting is an option for Radeon 480, and you can also obtain better performance with undervolting, belive it or not:
http://www.hardware.fr/articles/951-30/overclocking-quelle-marge-rx-480.html

 

[revan]

Well, not Damien but the boys from computerbase.de made the undervolt experiment, mea culpa:
https://www.computerbase.de/2016-06/radeon-rx-480-test/12/

 

[Clukos]

So users that just got a 480 know better than AMD themselves how to run HW they had in the lab for 6+ months?

Statistically it's probably better in AMDs eyes to have a slightly increased voltage curve to ensure that every GPU runs at the marketed 1266 core clock without problems rather than undervolting and pray that it works. That doesn't mean that most GPUs cannot undervolt though.

 

[Entropy]

Statistically it's probably better in AMDs eyes to have a slightly increased voltage curve to ensure that every GPU runs at the marketed 1266 core clock without problems rather than undervolting and pray that it works. That doesn't mean that most GPUs cannot undervolt though.

It's a question of performance distribution of chips/boards/client envrionments + margins. I would guess that dropping voltages by 10% and power limit by 15% or so would cost negligeably in terms of performance, and drop power (and noise) significantly. Thomas Ryan did something like that here. This is an important chip and marketsegment for AMD. It will be interesting to see how P10 performs say six months of driver/firmware tweaking and a chip revision down the line.

 

[Ryan Smith]

The following AMD statement just hit my inbox:

As you know, we continuously tune our GPUs in order to maximize their performance within their given power envelopes and the speed of the memory interface, which in this case is an unprecedented 8Gbps for GDDR5. Recently, we identified select scenarios where the tuning of some RX 480 boards was not optimal. Fortunately, we can adjust the GPU's tuning via software in order to resolve this issue. We are already testing a driver that implements a fix, and we will provide an update to the community on our progress on Tuesday (July 5, 2016).

 

[renderstate]

Will it invalidate reviews/affect performance?

 

[Ryan Smith]

Will it invalidate reviews/affect performance?

Quite possibly. It depends on the root cause. If power is somehow being wasted and it can be stopped (think leaking hose) then no. However if the issue is that power consumption is just being allowed to go too high (think a hose turned up higher) then yes, the card would need to be throttled back to avoid breaking power containment.

 

[Silent_Buddha]

Seems like what i thought might be true after all: https://www.reddit.com/r/Amd/comments/4qupw4/super_psa_all_rx480_owners_please_attempt_to/

480 users are undervolting + overclocking their GPUs and it seems to work so far. Might still be a lottery so who knows.

Hmmm, interesting, reading through that thread apparently some users have gotten cards that are factory undervolted [not underclocked] by AIBs and thus don't suffer the really high power draws.

The following AMD statement just hit my inbox:

Interesting. I'm interested to see if AMD will say what the root cause of the issue was. My first thoughts were that they were trying too aggressively to salvage dies for Rx 480. But that wouldn't make sense as Rx 470 isn't even out yet, so it doesn't seem like they have an unusually disproportionate amount of dies that could be used in 470 but not 480.

Hopefully a fix doesn't greatly impact the performance of the card. I'm trying not to be cynical and think that AMD knowingly pushed things farther than they should have just to get more performance in launch benchmarks.

[edit] whoops, I meant undervolting not underclocked. corrected it [/edit]

Regards,
SB

 

[Alexko]

The wording of that email suggests a rather small issue to me, but we'll see.

 

[kalelovil]

Have any sites tested the power consumption of the 4GB 7Gbps GDDR5 version (bios) of the RX480?

Edit: Nevermind. Anandtech has, it is within a few watts.

 

[xEx]

Quite possibly. It depends on the root cause. If power is somehow being wasted and it can be stopped (think leaking hose) then no. However if the issue is that power consumption is just being allowed to go too high (think a hose turned up higher) then yes, the card would need to be throttled back to avoid breaking power containment.

Thanks Ryan. Btw have to the ability to test AMDs new power optimizations technologies like dynamic clocking? It would be interesting to see if the card was actually fine but last minutes problems force AMD to disable or change the settings to a not optimal state and ended with this mess.

Its actually very sad all of situation of the "powergate".

 

[CSI PC]

It is not necessarily the clocks per se that are the problem.
It is the power target/voltage envelope they applied by default.
You can get pretty close performance to defaults just by lowering the voltage values and in turn reduces the power consumption.

Question that will never be answered is why the default settings and what were they focused on for their expected ideal parameters.
I would still recommend this card over a 970, just reduce the voltage a bit and no to increasing clocks/power target.
Cheers

TBH I think it comes back to this and the performance envelope parameters (and clocks indirectly), along with how the power management solution on the card is working in terms of distributing demand.

Now that said, after a solution is implemented would it make the 480 also applicable for OCing or importantly 2x480 mGPU setups in a PC.
Great that a solution will be implemented, just needs to be made clear if it really only applies to using a single card in standard form, so as to set expectations as quite a few it seems are interested in mGPU or OCing.
This may be a consideration showing in those situations a custom AIB card is the only feasible solution.
Cheers
Edit:
Also forgot to re-iterate that need to see if they can configure the power distribution, but depends upon flexibility of their power management solution and process involved.
Otherwise they will potentially end up with less performance if just focusing on the performance envelope.

 

[CarstenS]

The wording of that email suggests a rather small issue to me, but we'll see.

I hope it's not just the too high idle power. They wanted to fix that in a driver.

About the PCIe spec being violated, AMD sadly never bothered to get back at us. But we're probably just too small and unimportant.

 

[Ryan Smith]

I hope it's not just the too high idle power. They wanted to fix that in a driver.

About the PCIe spec being violated, AMD sadly never bothered to get back at us. But we're probably just too small and unimportant.

I don't think they've gotten back to anyone before this. This was a press blast sent out by the EMEA office to everyone.

 

[Grall]

So basically what we have here is a very strange behavior indeed, a group of people willingly determined to look the other way on any out of line behavior from AMD. Just because they are the underdog. A card that is out of spec so what? It's only damaging when OC'ed! as if this is suddenly a good thing, instead of talking about how good a card overclocks, now we talk about how it's the user's fault if he damages his hardware! because hey, It's AMD, they need our support, right?

I agree wholeheartedly; AMD needs our support , and breaking specs is BAD, very bad. Specs are there for a reason, and they shouldn't be violated.

While your rant isn't wrong - weak market players all wither and die (or get bought out), I don't think we'll ever see a new competitor to NV rise up in this marketplace. It's AMD or bust, because the required investment is simply too huge. Even Intel backed down and gave up the idea of entering the discrete graphics market after years of R&D into larrabee, and they have more genius engineers and money to throw at a project than most. If they couldn't do it, I doubt anyone else will bother to try, ever.

Exactly they did it through PCIe molex connector but violated it non the less. Or how do you explain 295 having only 2x8 pin and consuming up to 600W. 0 outrage. Heck it wouldn't surprise me if every Hawaii GPU is out of spec.

So if the specs are there for a reason as now everyone is saying, they are probably there for the same reason when it comes to power being delivered over the connector.

There's a big difference between going out of spec on the PCIe slot and doing the same on the aux PCIe molex connectors. For starters, you can't blow your mobo by drawing too much from the molex connectors. Second, molex connectors have a huge amount of headroom built into them - each specced for close to 300W, and they can undoubtedly withstand even more than that before Bad Things happen, simply because they're massive compared to the piddly power pins in the PCIe slot.

Your argument that we shouldn't care now about this spec violation because nobody allegedly cared in the past about other alleged spec violations is as irrelevant as it is dumb. Sorry, but it's true. By the way, what evidence do you have that every hawaii card violates PCIe specs? That's quite the accusation, you know. I don't think you've got shit, by the way.

The following AMD statement just hit my inbox:

Interesting. A software fix would be the ideal solution. Too bad that AMD's already tarnished reputation has already been further dinged by this whole affair though. As a (relatively) faithful fan of the red team, that gives me a sad.

 

[Entropy]

I think the PCPer investigation pretty much conclusively shows what is going on.
The card demonstrates an anomalous behaviour in how it draws power from the PCI-e connector and the Molex respectively, in that the power draw is just split between them rather than the typical behaviour of both nVidia and AMD of a constant limit over PCI and letting the other feed vary.
Now, this would still stay within spec with a small downward adjustment of voltages, however a preferable solution would be to fix the behaviour to the normal one. I can't really see why the card would be hardware designed for this atypical behaviour, so it may well be that this can be fixed via a software update on some level.
It would be the better solution. If it requires a card bios update, it may take a bit longer to propagate to the end users.

 

[Aaron Elfassy]

I've owned multiple phenom cpus, laptop apus, and amd gpus going back to the 5870. Almost all of them could be undervolted to improve power efficiency, heat and max clocks from throttling (especially on the Llano apu). They use very conservative voltage profiles, which is a shame. Undervolt and enjoy!