The 480: power consumption, PCI-E powerdraw

[2is]

Lol,somme people are still waiting for a driver to enable Async on Maxwell, almost one year that it was promised...

Hey, guess what, it s not enabled, Nvidia rep said at the time...

I think you're projecting a bit from all the years of having to wait for drivers months after a games release.

Only people who could possibly be waiting for Async Compute are AOTS players, so basically nobody.

It's about an obscure of a game as links showing FX8 edging out an i5.... If you're running WinRAR AND cinebench AND playing a game... So basically never

 

[Abwx]

I think you're projecting a bit from all the years of having to wait for drivers months after a games release.

You re stuck years behind, it s a long time that AMD drivers are more on point than their competitor s..

Only people who could possibly be waiting for Async Compute are AOTS players, so basically nobody.

Yes, because we can expect AoTs to be the only game using async in the coming months and years..

It's about an obscure of a game as links showing FX8 edging out an i5.... If you're running WinRAR AND cinebench AND playing a game... So basically never

Yet another urban legend..

http://forums.anandtech.com/showpost.php?p=38335025&postcount=48

 

[zinfamous]

I think you're projecting a bit from all the years of having to wait for drivers months after a games release.

Only people who could possibly be waiting for Async Compute are AOTS players, so basically nobody.

It's about an obscure of a game as links showing FX8 edging out an i5.... If you're running WinRAR AND cinebench AND playing a game... So basically never

strange argument considering it's a major feature of DX12 which is today and future games.

So something that nVidia either can't deliver for the next 4 or 5 years, or will suddenly become "an important and amazing feature!" for nVidia cards only when nVidia adopts it. Nevermind the delays and "lies" about it being available before and how far behind they were compared to AMD. That is the past!

Your argument reads exactly like the old VRAM argument with nVidia: "3gb is more than enough! No one really needs more than 2gb anyway!" in response to AMD running around with 4gb cards. Then, of course, we start seeing 4gb cards and nVidia fans criticizing any AMD card in that same class with less than 4gb: "underpowered, overpriced crap!"

 

[guskline]

IEC, thank you for the info on your mb, etc. I really hope AMD corrects this problem with a software fix.

I also hope that Tom's hardware and PC Perspective run the same test on an AIB RX480 with an 8 pin connector to see if the observed higher than spec PCI-E draw disappears.

 

[rainy]

Lol,somme people are still waiting for a driver to enable Async on Maxwell, almost one year that it was promised...

Hey, guess what, it s not enabled, Nvidia rep said at the time...

If I remember correctly, they have promised also DX12 support for Fermi and that not happened either.

 

[MrTeal]

On another issue, the actual PCB design itself is very stupid.

https://youtu.be/plC7tOYIqBw?t=53m55s

AMD split up the VRM power channels to separate power from 6pin and PCIE, ie, the 6 pin delivers power to half of the VRM banks and the MB delivers the other half.

This is not how GPUs are normally designed. They should have combined the power sources and draw extra via the 6pin. So if the card is OC, it will draw the extra via the 6 pin and not split 50-50 with the MB which will run over-spec.

So even if they fix this in stock config, when OC, it will exceed 75W from the MB.

It's just an awful design, like someone took a shortcut, when designing for Polaris 11 to run off PCIE slot power, and scaling it up to Polaris 10 SKU, they just added extra VRM phases and put those on the 6pin rather than re-designing it.

It is literally designed to not be stable OC at all, period, even if your PSU and rails are good. Push it too far, your system shuts down to OCP through the MB.

That actually is how GPUs are normally designed. It's not nearly as simple as you make it out to be; you can't just combine them all into one big plane and then draw extra from the 6 pin. Almost all GPUs segregate the PCIe slot and the power plugs because that's the only way you can control the amount of current each is delivering.
What AMD should have done was put all the core phases on a single 8 pin. Their choice to go with a single 6 pin wasn't a good one in this case as the TDP is marginal for a 150W card, but that hardly makes the PCB design very stupid. The PCB is designed in one of the few ways you could do it while still keeping both the slot and 6 pin power at 75W each.

 

[Shivansps]

Complete non sense, the MB PCIe has a capability of 300W for a 4 slots MB, if a permanent drain of 240W could cause a problem just imagine the GTX960 in SLI (as often recommended here) that should drain 500W peaks for two cards.

Anyway your post is noticed so in a few weeks i can remind you how fuddy it was retrospectively...

You did not even know that CPU pull power from the EPS plug instead of the ATX one and you now want to discuss me PCI-E specs?
The boards can supply 300W? great! point me at the 2x8PINs on the MB in order to provide that amount of power.

Im having trouble to find them. That ATX does not have enoght 12V lines or grounds to do that. Take 2x8 pins plugs and start counting wires, then tell me.

And just stop bringing up a single GTX960 peak power use, peaks are not gona cause heat as long they are short, you need to stay over the limit for most of the time on continuous usage to do that.

Im not sure why i even bother, its like talking to a wall, MBs cant comply with PCI-E power standards and everyone is expected to respect REV 1.0 specs, so you cant ensure a thing, it may blow up something or may not, its highly likely that nothing gona happen, but im not going to put my hands on the fire for that like you are doing.

I whould expect that everyone whould be on the same boat here, but no.

 

[Erenhardt]

Wright. Let me get 300Amps through your system just for a 0.1 sec. The average will be 360W, which is nothing.

I will send you something to plug into your PC, OK?

 

[Azix]

You did not even know that CPU pull power from the EPS plug instead of the ATX one and you now want to discuss me PCI-E specs?
The boards can supply 300W? great! point me at the 2x8PINs on the MB in order to provide that amount of power.

http://www.gigabyte.com/fileupload/product/2/4485/8666_big.jpg[/IMG

Im having trouble to find them. That ATX does not have enoght 12V lines or grounds to do that. Take 2x8 pins plugs and start counting wires, then tell me.

And just stop bringing up a single GTX960 peak power use, peaks are not gona cause heat as long they are short, you need to stay over the limit for most of the time on continuous usage to do that.

Im not sure why i even bother, its like talking to a wall, MBs cant comply with PCI-E power standards and everyone is expected to respect REV 1.0 specs, so you cant ensure a thing, it may blow up something or may not, its highly likely that nothing gona happen, but im not going to put my hands on the fire for that like you are doing.

I whould expect that everyone whould be on the same boat here, but no.[/QUOTE]

A motherboard with 4 x x16 slots has to be able to support 300W. At least if it's supposed to be able to handle cards in all those slots. Question is if the PSU 24 pin can handle it.

the 12V pins on the motherboard 24 plug can handle it up to 13A and there are 5 of them iirc. depends on the quality. For that kind of board I would assume its that high.

 

[MrTeal]

A motherboard with 4 x x16 slots has to be able to support 300W. At least if it's supposed to be able to handle cards in all those slots. Question is if the PSU 24 pin can handle it.

the 12V pins on the motherboard 24 plug can handle it up to 13A and there are 5 of them iirc. depends on the quality. For that kind of board I would assume its that high.

Why do you assume that's the case? You can have a dozen electrical outlets on your breaker, and each can supply a hair dryer at 1200W. If I tried to plug in 10 1200W hair dryers at once, things wouldn't work well.

Asus made a very nice X79 board with 8 x16 slots and no supplementary connector, you definitely would run into issues if you tried to pull 50A through the two ATX 24pin 12V wires.

There's 2 12V pins on an ATX 24 pin connector, and if they use the best quality terminals (the Minifit Plus HCS) the contact rating is 8A (with 16 gauge wire, most PSUs are 18ga. so the rating is 7A).

 

[rainy]

Moja sztuka którą kupiłem głównie do testów bez problemów trzyma te same częstotliwości jakie ustawiłem w Wattmannie także po obniżeniu napięcia.
Aktualnie tabela kończy się na 1290MHz z napięciem 1.000V Pobór mocy średni 135,4W.
Katuję ją 24h na dobę programami do minigu i żadnych problemów. A minęła 3 doba.

It was said by Polish user: my RX 480 have no problem with keeping clocks after setting them in WattMan (also after undervolt).
Table ending at 1290 MHz with vcore 1.0V, average power consumption 135.4W.
He has "torture" the card for 3 days with various mining programs and no problem at all.
http://pclab.pl/news70435.html

 

[MrTeal]

Also, people here might find this interesting.
http://suddendocs.samtec.com/notesandwhitepapers/simplified-pulse-current.pdf
Unfortunately it doesn't go down to 1% or lower as you'd like to see given how short these high current pulses are (though you could do that yourself in Excel if you really wanted), but it should be apparent that as long as the RMS current is kept within spec, short excursions of a couple times over the connector rating aren't an issue.

 

[Erenhardt]

nvm

 

[Zstream]

Snip

That's not how it works at all... :thumbsdown: Tired of debating with people like yourself that refuse to understand or learn.

 

[.vodka]

It was said by Polish user: my RX 480 have no problem with keeping clocks after setting them in WattMan (also after undervolt).
Table ending at 1290 MHz with vcore 1.0V, average power consumption 135.4W.
He has "torture" the card for 3 days with various mining programs and no problem at all.
http://pclab.pl/news70435.html

That's at least a 0.15v undervolt (there have been screenshots of boost voltages higher than that) with a 24MHz OC over stock 1266MHz boost clocks. >0.15v, that's missing the mark by a mile :\


See how power consumption now falls within the 150w envelope? We'll see if tomorrow's fix undervolts the cards a bit to mitigate the issue.

 

[2is]

You re stuck years behind, it s a long time that AMD drivers are more on point than their competitor s..



Yes, because we can expect AoTs to be the only game using async in the coming months and years..



Yet another urban legend..

http://forums.anandtech.com/showpost.php?p=38335025&postcount=48

More on point with drivers? Making up your own reality like your own pcie standards are we? Are you still looking for that infamous 960 that is out of pcie spec? How's that working out?

 

[2is]

strange argument considering it's a major feature of DX12 which is today and future games.

So something that nVidia either can't deliver for the next 4 or 5 years, or will suddenly become "an important and amazing feature!" for nVidia cards only when nVidia adopts it. Nevermind the delays and "lies" about it being available before and how far behind they were compared to AMD. That is the past!

Your argument reads exactly like the old VRAM argument with nVidia: "3gb is more than enough! No one really needs more than 2gb anyway!" in response to AMD running around with 4gb cards. Then, of course, we start seeing 4gb cards and nVidia fans criticizing any AMD card in that same class with less than 4gb: "underpowered, overpriced crap!"

Your argument reads like the same ones I've heard for a decade. Looking to the future where AMD products will suddenly perform better then they are now. In the last 15 years I've heard this, it's happened exactly once. And that's with the 7950/7970 vs 670/680.

Here are the facts, nvidia outperforms AMD in every game, maybe you can find an outlier somewhere but that's pretty much the reality. Until that actually changes, you're living in a world that doesn't actually exist if you think AMD is better die to a feature that still has them being 2nd best.

So you can taunt async compute all you want... You're still slower.

 

[MrTeal]

That's not how it works at all... :thumbsdown: Tired of debating with people like yourself that refuse to understand or learn.

Well, I wasn't replying to you, but if you're going to say that it's not how it works at all, at least have the courtesy to quote the part you're disagreeing with.

 

[JDG1980]

I love the fact that if AMD advertises features that don't seem to exist in the shipping product is because they haven't enabled them yet, not because they are lying.

You see..we must understand AMD cannot possibly lie, they just don't enable. It's yet another thing they forgot to do before shipping the (not so)final product. It's just life, this stuff happens..

*plonk*

This message is hidden because renderstate is on your ignore list.

Ahh, that's better.

 

[William Gaatjes]

That actually is how GPUs are normally designed. It's not nearly as simple as you make it out to be; you can't just combine them all into one big plane and then draw extra from the 6 pin. Almost all GPUs segregate the PCIe slot and the power plugs because that's the only way you can control the amount of current each is delivering.
What AMD should have done was put all the core phases on a single 8 pin. Their choice to go with a single 6 pin wasn't a good one in this case as the TDP is marginal for a 150W card, but that hardly makes the PCB design very stupid. The PCB is designed in one of the few ways you could do it while still keeping both the slot and 6 pin power at 75W each.

I agree it is a bit on the edge when it comes to the maximum available electrical power. But now more and more positive signals come from the field that undervolting does not harm stability but it does lower the amount of power needed to run the card.
It seems to me that the BTC is not used or there is a software bug in the calibration or the core voltage is not set appropriate.


The 6+1 phases smps controller IR3567 has some great features. It can turn of phases when there is not much power needed. That makes me think, to make it work i guess it is possible to put one phase on the pcie card connector so that there is always 12V to power the core. And put the other 5 phases on the pcie power connector (8 pin version).
That way, the core can always run, just at a very low clock to warn the user that there is no pcie power present.


All in all , it seems to be like a storm in a glass of water.

 

[JDG1980]

Your argument reads like the same ones I've heard for a decade. Looking to the future where AMD products will suddenly perform better then they are now. In the last 15 years I've heard this, it's happened exactly once. And that's with the 7950/7970 vs 670/680.

How about 290/290X vs 780/780Ti/Titan? The Hawaii cards held up a lot better.

And no one even bothers to do testing on the GK106 cards (GTX 660, GTX 650 Ti, GTX 650 Ti Boost) any more, even though they were big sellers at the time. Back then they often edged out Pitcairn (7850/7870), but now...

 

[Erithan13]

LegitReviews article on undervolting the 480

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

I've seen concerns that the upcoming power fix by AMD may be a simple case of lowering the clocks and reducing performance, which would basically be turning one controversy into another. However this very much suggests the fix could lower temperatures/power consumption while also increasing performance which is a win all round. The reference cooler will also look considerably better and less overworked as a result. We'll know more tomorrow.

I've said it before but people have really been piling on AMD for failing at perf/w with Polaris. Getting the 480 decisively below 150W and resolving the pcie overdraw will be one big step in the right direction. If the power draw can be brought closer to 100W with clever driver optimisation etc that paints a much better picture for Polaris and will also give stiffer competition to the 1060 in that area.

We also have to wait and see how the 470 and 460 stack up in perf/w. It's been pretty ridiculous to see declarations that the whole 14nm process is a bust and Zen/Vega are doomed based only on the 480.

 

[MrTeal]

I agree it is a bit on the edge when it comes to the maximum available electrical power. But now more and more positive signals come from the field that undervolting does not harm stability but it does lower the amount of power needed to run the card.
It seems to me that the BTC is not used or there is a software bug in the calibration or the core voltage is not set appropriate.


The 6+1 phases smps controller IR3567 has some great features. It can turn of phases when there is not much power needed. That makes me think, to make it work i guess it is possible to put one phase on the pcie card connector so that there is always 12V to power the core. And put the other 5 phases on the pcie power connector (8 pin version).
That way, the core can always run, just at a very low clock to warn the user that there is no pcie power present.


All in all , it seems to be like a storm in a glass of water.

Yeah, you can use phase shedding to increase efficiency at low loads. That's actually a pretty common feature in modern multiphase controllers. You wouldn't use that for the application you're describing most likely, as it's a dynamic feature you'd program into the VRM. You'd use the Loop_x_Max_no_phases register to drop to top phases to 1 or 2 when you're running with no 12V on the higher phases.

Edit: I definitely agree though, this is a lot of controversy over not very large of an issue. This isn't going to be a problem for the slot connector itself, and won't be an issue for the motherboard unless you have multiple cards installed. I wouldn't want to put four cards into a board without using powered risers, for example. That's definitely a specialized application though.
The one other thing I'd say is that I wouldn't really draw a big distinction between lower cost and high cost motherboards in this regards. I have my doubts that the quality of connectors is that much higher if at all on the better motherboards, and while there might be more board copper on some of the better boards the primary dissipation mechanism on the connectors is through the wires anyway, which is the reason the Minifit connectors have a higher current rating for wire to wire than wire to board.

 

[zinfamous]

Your argument reads like the same ones I've heard for a decade. Looking to the future where AMD products will suddenly perform better then they are now. In the last 15 years I've heard this, it's happened exactly once. And that's with the 7950/7970 vs 670/680.

Here are the facts, nvidia outperforms AMD in every game, maybe you can find an outlier somewhere but that's pretty much the reality. Until that actually changes, you're living in a world that doesn't actually exist if you think AMD is better die to a feature that still has them being 2nd best.

So you can taunt async compute all you want... You're still slower.

Me? I'm not a team, lol.

outperforms AMD in every game.

that's rather bold, especially with DX12 for the next couple of years. Looks bad for your team, though.

 

[namx01]

I have not seen this mentioned anywhere (maybe because it is stupidly impossible) but what would happen if you used a 6pin to 8pin adaptor on the 480?