AMD confirms Power Issue for RX 480, blames memory bus, will patch

[lolfail9001]

My theory is that this entire "goof-up" was intentional so that everyone would be distracted away from the real news story, which is that the Polaris architecture is a complete turd; a sort of "Bulldozer" of GPUs.

Not really, Bulldozer would mean worse than Tonga, and P10 is definitely better arch-wise than Tonga.

 

[Palladium@SG]

It will be interesting to see how they handel public realations on this matter, if they want maredshare back, they do have to come up with something good. Especially since the 1060 is almost here, RTG is cornered, Raja do need something good and that sone, i wonder what the 490 is.

When has AMD PR and marketing not in shambles? How long do they need to realize that target market isn't exactly tech savvy to win them over solely by undercutting NV by price and perf/$, and a huge proportion buys NV cards by default because it's the "safe and reliable choice" like Apple? And now they want another screwup like PCIE power into the mix? My mind is blown.

 

[razor1]

Funny thou Razor, you are more optemistic than me, you are supposed to be a green Ferengi..

I've been in bad situations before in major companies at a fairly high level, just before ya get to director levels, so yeah I understand the ramifications of what is going on for Raja and the RTG group (I have also been in a company that changed their name and branding around because they failed in the previous quarter, I was brought on to fix that, wow I was there for 6 months as a consultant VP for production, and I didn't like it there lol). It takes a good 6 months to a year before your idea's are implemented into a corporate structure, and this is not in the tech world, it would take even longer there because the scope of things they are doing time wise are much longer.

I have also done consulting work for companies that go in an "optimize" other companies, it truth we are looking for cutting down and firing people that aren't needed lol. The shit isn't pretty but has to be done.

 

[ppi]

It will be interesting to see how they handel public realations on this matter, if they want maredshare back, they do have to come up with something good. Especially since the 1060 is almost here, RTG is cornered, Raja do need something good and that sone, i wonder what the 490 is.

It seems to me that nVidia is going for the kill ... with AMD getting cornered 3dfx style. They would probably get more profits from pushing more 1080s and 1070s to the market, but now they will divert their TSMC allocation to produce cheaper 1060s to ruin the RX480 launch. I suppose getting rid of competition is juicier goal.

 

[Rvenger]

It seems to me that nVidia is going for the kill ... with AMD getting cornered 3dfx style.

AMD has been killing themselves lately with these late launches and half baked product going on 2 generations. Something has to give.

 

[Meeho]

You GET VDROOP, and reviewers have noticed it!

Its just purely ignoring the facts that are presented to us if you didn't see that!

This is why the amp's are going up so much on the rails. The motherboard it trying to compensate.

Either this is a case of miscommunication or you're not making any sense. I'll try to explain one more time. You've stated that lowering power consumption to get in spec via downvolting is not enough because the lower volts would require higher amps, which is not the case. If you downvolt the GPU or the memory, you will reduce the power draw from the PSU and MBO and that could only lead to less vdroop. So, less amps from lower power draw + less amps from smaller vdroop. Downvolting to get in spec both power and amp wise, while keeping the stock base clock and boost, is certainly a possibility. It's a bad solution, though, because even if all the cards could keep the performance while being downvolted, you would still get possibly dangerous power draw if you would OC the card and that is not acceptable.

Well meeho isn't trying to defend the rx480, I know that, I think I should have been more clear with what I stated from the get go.

I didn't recognize myself in that comment anyway. There is nothing to defend regarding the RX 480 power issues.

 

[Meeho]

Now, I am confident that AMD will fix the power draw issue, I just don't know what effect, if any, their fix will have on performance. If, as stated elsewhere, they have a powerful programmable power control module ic chip, then a software fix may make all this moot. Or, maybe they'll have to send everyone an 8 pin adapter.

8pin would not solve this. 6pin alone is enough to provide the required power, especially the way they have it wired on the RX 480. Distribution between the 6pin and the PCIe slot is the problem and it remains to be seen if they can fix that via the control chip.

 

[SeymourGore]

I'm not very familiar with this whole 'out of spec' issue, but can we expect a measureable drop in performance once the issue is fixed?

 

[variant]

LegitReviews undervolted it and ran performance tests.

AMD Radeon RX 480 Undervolting Performance - Legit Reviews

By taking a closer look at the clock speeds and voltages on the cards we can see why we are getting better performance while gaming. At stock settings the AMD Radeon RX 480 runs at it’s peak boost clock speed of 1266Mhz for about 18 seconds on this particular game title before retreating down to as low as 1211Mhz for the rest of the benchmark run. The voltage on the card was reported by GPU-Z as being 1.1187 Volts for most of the first portion of the run before dropping down and jumping around for the remainder of the benchmark. Note that one of the polling times saw the VDDC running at 1.3125V, which is a pretty high spike that is well over the 1136mV the card is set to by default. By undervolting the card we were able to keep the clock speed at the top boost clock of 1266MHz for the entire duration of the benchmark run and that meant better performance, lower GPU temperatures and less power consumption. We noticed 10-30W power reductions at the power outlet meter by undervolting the Radeon RX 480! A one voltage fits all approach obviously leaves some room for adjustment!

It appears that AMD was pretty easy going with regards to the higher P-state voltages as we were able to reduce ours from 1.15V to 1.05V with the new AMD WattMan software and still had full gaming stability. By lowering the power level down it increases the efficiency of the GPU and for gamers that means you get to hold the boost clock speed longer and get better performance and use less power. Having the ability to turn all the P-State knobs is pretty nice and we are already seeing the new WattMan software paying off. Being able to get a small performance boost while lowering the power consumption and heat levels was pretty nice and it’s something you might want to try out if you happen to have an AMD Radeon RX 480 graphics card. It’s easy to geek out and enjoy seeing performance go up in benchmarks while seeing double digit Wattage drops at the wall!

 

[travm]

Undervolt to increase performance. What will they think of next.
someone(s) screwed up pretty bad. Looks like half decent hardware, just terrible finish.

 

[Araxie]

LegitReviews undervolted it and ran performance tests.

AMD Radeon RX 480 Undervolting Performance - Legit Reviews

yeah man we get the point, that's one of the marvelous things of FinFet you know? it was widely know since years ago when intel introduced ivy bridge, with those chips a 3770K could be able to overclock up to 4.2ghz undervolted from stock clocks. yeah yeah, nothing great. the thing it's when you needed to run a little more than 4.2ghz for example 4.4ghz then you needed not only to go to stock volts but overvolt it by an ample margin in comparison to was able to achieve undervolted so a fast example.. stock chip 1.16v stock setting 3.7ghz when loaded... undervolted to 1.06v able to reach 4.2ghz still stable... then needed 4.4ghz? at that point most needed to go over 1.22v to be able to reach stable that kind of speed. needed 4.5ghz then the jump was over 1.25V. needed 4.8ghz then you instantly was already in the 1.35v area..

Fin fet Scale horribly with voltages but, you need from the design point ensure you are adding stability to 100% of the chips.. not only those able to reach certain speeds at certain amount of voltages. that what's called stability headroom because not every application stress the chip on the same way, you have to add a stock voltage based in a +/- % tolerance to stress and leakage, nobody put a stock voltage based in a arbitrary mind, that's what validation process are made for... not every chip behave the same way with regard of voltage scaling, you can't run to anyone and say "hey they found that undervolting help to solve the issue" then someone do that and receive instabilities...

 

[Palladium@SG]

Undervolt to increase performance. What will they think of next.
someone(s) screwed up pretty bad. Looks like half decent hardware, just terrible finish.

Apparently, AMD bothered to put in incredibly overbuilt VRMs which can supply an ungodly 600A current to the GPU even with the ability to shift power loading from the slot to the 6-pin, and the GPU is heavily overvolted for stock.

Everything looks to be in place for a perfect launch but they juuuuuust can't be bothered. Not sure if "stupid" is a good enough word for this debacle.

 

[Dreamerbydesign]

My theory is that this entire "goof-up" was intentional so that everyone would be distracted away from the real news story, which is that the Polaris architecture is a complete turd; a sort of "Bulldozer" of GPUs.

Either way they are not selling GPU's with issues to consumers in the know. Sure they will make a few fan boy sales and plenty of un informed buyer sales, but anyone who does an research on tech related products will not be buying this card.

 

[c3k]

"Debacle" is apt.

 

[Chimpee]

yeah man we get the point, that's one of the marvelous things of FinFet you know? it was widely know since years ago when intel introduced ivy bridge, with those chips a 3770K could be able to overclock up to 4.2ghz undervolted from stock clocks. yeah yeah, nothing great. the thing it's when you needed to run a little more than 4.2ghz for example 4.4ghz then you needed not only to go to stock volts but overvolt it by an ample margin in comparison to was able to achieve undervolted so a fast example.. stock chip 1.16v stock setting 3.7ghz when loaded... undervolted to 1.06v able to reach 4.2ghz still stable... then needed 4.4ghz? at that point most needed to go over 1.22v to be able to reach stable that kind of speed. needed 4.5ghz then the jump was over 1.25V. needed 4.8ghz then you instantly was already in the 1.35v area..

Fin fet Scale horribly with voltages but, you need from the design point ensure you are adding stability to 100% of the chips.. not only those able to reach certain speeds at certain amount of voltages. that what's called stability headroom because not every application stress the chip on the same way, you have to add a stock voltage based in a +/- % tolerance to stress and leakage, nobody put a stock voltage based in a arbitrary mind, that's what validation process are made for... not every chip behave the same way with regard of voltage scaling, you can't run to anyone and say "hey they found that undervolting help to solve the issue" then someone do that and receive instabilities...

Sounds like their is a huge variation in the binning that AMD did use this method right from the get go.

 

[Araxie]

Sounds like their is a huge variation in the binning that AMD did use this method right from the get go.

which given kyle's editorial and statements is nothing strange.... the chip have issues, the PCB have issues, when you mix both you have the problem AMD is actually facing..

 

[NKD]

Either way they are not selling GPU's with issues to consumers in the know. Sure they will make a few fan boy sales and plenty of un informed buyer sales, but anyone who does an research on tech related products will not be buying this card.

GPU itself does not have issue. Power draw is the issue. If they can't fix it, good! they have to make it right. So for now stay tuned. I didn't buy this card but if they fix the power draw I certainly wouldn't hesitate to recommend it.

 

[primetime]

GPU itself does not have issue. Power draw is the issue. If they can't fix it, good! they have to make it right. So for now stay tuned. I didn't buy this card but if they fix the power draw I certainly wouldn't hesitate to recommend it.

I think the gpu issue Araxie is referring to is the initial intended performance?.....As in they were targeting a certain segment at much reduced power need? IDK what they got and what they wanted are two different things. A few people mentioned it was already 7 revisions in and that in itself sorta gives an impression their was more than typical complications

 

[razor1]

Well its not a GPU architectural issue Araxie is talking about, its an issue with the node and the voltage variances on different chips.

 

[Algrim]

Based on power draw, the 470 might have been the intended target.

 

[razor1]

stock voltage for the 1070 and 1080 are much lower than the rx 480 (and the 1070 compared to the 1080 voltage levels are quite similar in stock) so its possible that the 470 is what they were hoping for as their base line,

 

[NKD]

Well its not a GPU architectural issue Araxie is talking about, its an issue with the node and the voltage variances on different chips.

Yea been saying this. When you have so many revisions of the chip and its obvious they made improvements with each one, looks like they are still perfecting the new node. I think in the end it may be a good thing vega is not here for another 5-6 months cuz that would have been a fail if they were trying to push that out right now.

 

[razor1]

5 to 6 months they will only be able to do 2 re spins lol, so time wise it won't solve anything if there is a fundamental issue with the architecture, maybe on the node yes. This is of course if Vega has the same issues as Polaris, which we don't know as it can be on a total different fab and architectural might have been done with another design team.

 

[bluesynk]

Not for nothing but if this card was a turd this would be a non issue. Amd was close to something here, like a new 9700 and just missed, and I think it scared some people.

 

[travm]

Not for nothing but if this card was a turd this would be a non issue. Amd was close to something here, like a new 9700 and just missed, and I think it scared some people.

Cant wait to see the custom boards. If history has taught us anything about AMD's graphics division, they might reasonably get 10% more performance out of mature drivers.

 

[razor1]

AMD Radeon RX 480 Power Measurements Repeated And Clarified

follow up to Tom's

 

[Sith'ari]

Yeah, measurements repeated by Tom's and results remained the same.
What about the patch? wasn't today the day that AMD was claiming that it going to release it?

 

[Mav451]

Can probably give AMD till COB to respond.

 

[Araxie]

Yeah, measurements repeated by Tom's and results remained the same.
What about the patch? wasn't today the day that AMD was claiming that it going to release it?

it could be at any hour of the day.. Im not hoping a fix for today but a more official statements on how they want to fix the issue..

 

[Daniel_Chang]

What about the patch? wasn't today the day that AMD was claiming that it going to release it?

Patch is unlikely. They stated that they would have more information sometime today. They didn't guarantee a working solution today.

 

[Araxie]

Patch is unlikely. They stated that they would have more information sometime today. They didn't guarantee a working solution today.

they also stated a Fix was in upcoming drivers for today but that's highly unlikely in my opinion...

 

[Palladium@SG]

they also stated a Fix was in upcoming drivers for today but that's highly unlikely in my opinion...

My bets are on a firmware fix for those still want to keep the card and a recall/refund for those who don't.

 

[Araxie]

My bets are on a firmware fix for those still want to keep the card and a recall/refund for those who don't.

If and only if, the solution can be fixed by BIOS. Im genuinely interested on how they are planing to fix this thing without crippling the performance of the card.

 

[Daniel_Chang]

they also stated a Fix was in upcoming drivers for today but that's highly unlikely in my opinion...

AMD's statement:

As you know, we continuously tune our GPUs in order to maximise 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.

They didn't promise a driver today. They stated that they are working on one, and will tell us the progress today.

 

[Osjur]

It is possible to temporary fix this problem with newest MSI Afterburner by programming the voltage regulator through I2C bus if you wan't to do it yourself. There's no BIOS flashing tool which supports Polaris atm so bios patch is not possible.

1. Open command-line and go to MSI afterburner directory.
2. (optional) Test the I2C bus with command: MSIAfterburner /ri06,08,0D (if answer is 20 or 44, the bus is communicating properly)
3. input MSIAfterburner /wi06,08,1E,BB
4. input MSIAfterburner /wi06,08,1F,B0

5. Now your card power distribution ratio has changed to favor 6-pin plug. Difference is about 10w.

TEST AT YOUR OWN RISK

Credit goes to The Stilt who figured this out.

 

[Araxie]

AMD's statement:



They didn't promise a driver today. They stated that they are working on one, and will tell us the progress today.

TLDR

Spoiler

 

[Araxie]

It is possible to temporary fix this problem with newest MSI Afterburner by programming the voltage regulator through I2C bus if you wan't to do it yourself. There's no BIOS flashing tool which supports Polaris atm so bios patch is not possible.

1. Open command-line and go to MSI afterburner directory.
2. (optional) Test the I2C bus with command: MSIAfterburner /ri06,08,0D (if answer is 20, the bus is communicating properly)
3. input MSIAfterburner /wi06,08,1E,BB
4. input MSIAfterburner /wi06,08,1F,B0

5. Now your card power distribution ratio has changed to favor 6-pin plug. Difference is about 10w.

TEST AT YOUR OWN RISK

there's a BIOS editor for polaris however I don't see anywhere a Option for balancing the phase load tables.

 

[Osjur]

Because it hasn't been implemented yet would be my guess. Also all the earlier bios modifications are usually being made with HEX editor.

And I also guess you can't flash the bios with that program.

 

[NKD]

TLDR

Spoiler

just says fix in upcoming driver. Progress Update tuesday the 5th. I mean how you are connecting the two is outside of my comprehension. It is pretty clear those two are not connected.

 

[Daniel_Chang]

TLDR

Looks like you also didn't read your own post. Says the same thing as mine.

Driver update being worked on.
Status provided on Tuesday.
Driver NOT released on Tuesday.