The 480: power consumption, PCI-E powerdraw

[Bacon1]

But then why didn't they test their new software before releasing it?

One would assume that when you make a new wattage tool, the first thing you test is the actual bloody wattage!

There can be lots of testing done and bugs still get through. It was probably setting the default voltage level too high for some reason. They've fixed it now by updating and fixing the software.

 

[PhonakV30]

Looking at results from new drivers this is AMD release fail all over again. Why not release it with compatibility mode at default and loose the 3% but be in spec, avoid bad rep and also show much lower power usage. Then offer a "Gaming mode" which they now call standard as a feature.

3% performance for 20% more power is a stupid trade-off. What were they thinking? Or are other games affected more?

20% more ? where did you get those numbers?

 

[antihelten]

There can be lots of testing done and bugs still get through. It was probably setting the default voltage level too high for some reason. They've fixed it now by updating and fixing the software.

Sure, but this would have been one of the most blatantly obvious things, so something like this shouldn't go through unless a) you do zero testing whatsoever of power usage (after just having created a new tool that is all about power usage) or b) you choose to ignore the issue.

20% more ? where did you get those numbers?

He might be referring to this article:

http://semiaccurate.com/2016/07/01/investigating-thermal-throttling-undervolting-amds-rx-480/

They tested with the power limit at 80%, which resulted in a 5% performance loss and an 18-19% reduction in power usage (assuming the card was using 160-165W at 100% power limit)

 

[MrTeal]

Well the power max depends a lot on switching frequency... The higher the frequency the smaller the inductor can be for a given power requirement.

Does anyone know what the switching frequency is for these VRMs on the RX 480? I'd bet it's > 1 Mhz.

Do you mean the effective frequency or the per phase frequency? I would be surprised if the phase was that high, but something more standard like 300kHz or 500kHz will give an effective ripple frequency that's higher than 1MHz of course. Even with integrated driver devices with extremely low inductance between the high and low side fets going over the 1MHz will negatively affect efficiency. With a discrete SO8 design it wouldn't be common.

I was basing my estimate on the size of inductors and output caps... It has to be AT LEAST 500kHz... For the power it's pumping out.

Possibly this one, wich get up to 1MHz.

http://www.infineon.com/dgdl/ir3550.pdf?fileId=5546d462533600a4015355cd7c831761

I got my card, BTW. The switching frequency is 300kHz.

 

[Yakk]

Because there is a group of users on these forums that have 0 experience with mining or running 900W-1000W GPUs on a 1000W Platinum PSU. They have no real world experience running mining rigs for years to know what load a modern mobo and PSU can actually handle. A 1000W EVGA or Corsair or SeaSonic PSU should be able to handle 5 or even 6 RX 480s mining @ 100% load 24/7 without a sweat (assuming it has enough PCIe cables/connectors).

Also, the older generation of PC gamers created so many myths that have stayed around. For instance, how a "GPU degrades over time", which we know is absolutely false. The amount of electro-migration is so immaterial that by the time a GPU fails from natural causes due to "being warn out", it'll be surpassed by a $50 GPU.
https://www.youtube.com/watch?v=iZUshOSWQRo

The reason some CPUs cannot hold their overclocks after 5-7 years long-term is because people have been running them overvolted beyond spec, which exponentially accelerated electromigration. Of course we also have the blower myth. Another myth from the old days is that you shouldn't run your PSU loaded by more than 75-80% of its max load or it'll "die a premature death." Fact of the matter is high quality PSUs are actually rated to run at 50C 24/7 max spec loaded. That's why there is warranty too. Most good PSUs now have 7-10 year warranties, mobos have 3 years.

Gigabyte RX 480 = 3 year warranty
Asus RX 480 = 3 year warranty

The fear mongering is out of control.

I had SeaSonic PCIe cables burn out due to early batches failing to meet spec, and SeaSonic sent me all new PCIe and 24-pin cables for free within a week.

That's why one of the most important components in a high quality PC is the PSU. If you are building a $700+ PC and buying a $20-30 PSU for it, it's your fault if things burn out.

The silly pcie issue was way overblown, but to their credit, AMD addressed it in record time, +1 to them for it.

I've mined, and been around... enough to know the amount of BS in forums in general.

It all boils down to:

1) Those who can, do.

2) Those who can't, repost random information they read from Google and misc. forums.

3) Those with a predetermined agenda misrepresent facts so they can be repeated by #2 above.

Rinse & repeat.

 

[MrTeal]

No, it's not.

The 75W rating for 6-pin and 150W rating for the 8-pin are not true ratings.

The six-pin connector uses two +12 V wires to carry up to 75 W, whereas the eight-pin connector uses three +12 V wires to carry up to 150 W. Although these figures are what the specifications allow, the wires and terminals of each connector are technically capable of handling much more power. Each pin in the PCI Express auxiliary power connectors is rated to handle up to 8 amps of current using standard terminals—more if using HCS or Plus HCS terminals. By counting the number of terminals, you can calculate the power-handling capability of the connector.

If you actually looked up the detailed specs for 6-pin graphics connector, you'd see that most modern PSUs made in the last 10 years can provide 192-288W of power for it:

http://www.tomshardware.co.uk/power-supply-specifications-atx-reference,review-32338-12.html

My R9 295X2 has been mining for a long time now and it draws way more power than its 375W rating suggests (75W PCIe + 150Wx 2 for each 8-pin). The 75W 6-pin and 150W 8-pin "paper" spec maximum are misleading and incorrect. What's more, modern PSUs that provide the 6-pin connector are usually made from a 6+2 split (i.e., an 8-pin cable) to allow for compatibility. That means the actual cable which is being used to provide power to the RX480 is rated to withstand at least 288W. (Using Std. Terminals (W))

In conclusion, unless you are buying a $10-20 PSU, there should be no issues.



Because there is a group of users on these forums that have 0 experience with mining or running 900W-1000W GPUs on a 1000W Platinum PSU. They have no real world experience running mining rigs for years to know what load a modern mobo and PSU can actually handle. A 1000W EVGA or Corsair or SeaSonic PSU should be able to handle 5 or even 6 RX 480s mining @ 100% load 24/7 without a sweat (assuming it has enough PCIe cables/connectors).

Also, the older generation of PC gamers created so many myths that have stayed around. For instance, how a "GPU degrades over time", which we know is absolutely false. The amount of electro-migration is so immaterial that by the time a GPU fails from natural causes due to "being warn out", it'll be surpassed by a $50 GPU.
https://www.youtube.com/watch?v=iZUshOSWQRo

The reason some CPUs cannot hold their overclocks after 5-7 years long-term is because people have been running them overvolted beyond spec, which exponentially accelerated electromigration. Of course we also have the blower myth. Another myth from the old days is that you shouldn't run your PSU loaded by more than 75-80% of its max load or it'll "die a premature death." Fact of the matter is high quality PSUs are actually rated to run at 50C 24/7 max spec loaded. That's why there is warranty too. Most good PSUs now have 7-10 year warranties, mobos have 3 years.

Gigabyte RX 480 = 3 year warranty
Asus RX 480 = 3 year warranty

The fear mongering is out of control.

I had SeaSonic PCIe cables burn out due to early batches failing to meet spec, and SeaSonic sent me all new PCIe and 24-pin cables for free within a week.

That's why one of the most important components in a high quality PC is the PSU. If you are building a $700+ PC and buying a $20-30 PSU for it, it's your fault if things burn out.

Again, the issue isn't really running the cards off a PSU, it's when they draw so much power from the slot as all that current is running through a couple pins on the 24 pin ATX connector. Even with the absolute best PSU setup (Minifit Plus HCS and 16 gauge wire) the current rating on two pins of the 24pin ATX is 8A.
http://www.molex.com/pdm_docs/ps/PS-45750-001.pdf
With a more common setup (standard Minifit Jr and 18 gauge wire) it's only rated at 5A per pin.
http://www.molex.com/pdm_docs/ps/PS-5556-001.pdf


Now, there's a lot of slop built in there. All circuits on a 24 pin are energized but not all are carrying full current, so that will give a good bit of wiggle room. The max ratings are also the current that causes a connector temperature of 30°C over ambient, which for most people isn't a temperature that will cause degradation.

Takeaway from that is that you can probably safely run 10A-12A per pin 24/7 long term safely from an ATX connector. More than that and bad stuff has a very good chance of happening. With 6 cards hanging off unpowered risers Yakk could run fine though it is really pushing it as most cards only draw 2-3A off the 12V rail. Running 6 RX 480s off unpowered risers is a pretty big ask though, when each is is pulling 6A. Even with healthy, pristine hardware you're likely going to see damage.

 

[MrTeal]

44-66 ratio

AMD, giving it that 110%

Hehe, they do try hard. 44-56, sorry.

 

[LurchFrinky]

Again, the issue isn't really running the cards off a PSU, it's when they draw so much power from the slot as all that current is running through a couple pins on the 24 pin ATX connector. Even with the absolute best PSU setup (Minifit Plus HCS and 16 gauge wire) the current rating on two pins of the 24pin ATX is 8A.
http://www.molex.com/pdm_docs/ps/PS-45750-001.pdf
With a more common setup (standard Minifit Jr and 18 gauge wire) it's only rated at 5A per pin.
http://www.molex.com/pdm_docs/ps/PS-5556-001.pdf


Now, there's a lot of slop built in there. All circuits on a 24 pin are energized but not all are carrying full current, so that will give a good bit of wiggle room. The max ratings are also the current that causes a connector temperature of 30°C over ambient, which for most people isn't a temperature that will cause degradation.

Takeaway from that is that you can probably safely run 10A-12A per pin 24/7 long term safely from an ATX connector. More than that and bad stuff has a very good chance of happening. With 6 cards hanging off unpowered risers Yakk could run fine though it is really pushing it as most cards only draw 2-3A off the 12V rail. Running 6 RX 480s off unpowered risers is a pretty big ask though, when each is is pulling 6A. Even with healthy, pristine hardware you're likely going to see damage.

Don't most motherboards have an additional 4pin, 12V connector? That's 2 extra 12V wires coming in.

 

[Flapdrol1337]

I saw an analysis of the pcb, and it said the 6 pin has 3 phases and the memory phase, and the pci-e has the 3 other phases.

How did the card originally draw more current from the pci-e pin? All I can think of is a big voltage drop on the pci-e slot.

2 pins on atx->mobo traces->pci-e slot->special measurement riser->card obviously has more resistance than the 6pin plug with some measurement inbetween. And having to draw more current results in an even bigger drop. But it's still strange.

Or are there other things connected to the pci-e too?

 

[MrTeal]

Don't most motherboards have an additional 4pin, 12V connector? That's 2 extra 12V wires coming in.

No, the 4 or 8 pin connector up by the CPU just feeds the CPU. It doesn't supply other motherboard peripherals.

 

[Doom2pro]

I got my card, BTW. The switching frequency is 300kHz.

Wow that's lower than I thought... But then again I guess since there are so many phases they can get away with it.

 

[Yakk]

Takeaway from that is that you can probably safely run 10A-12A per pin 24/7 long term safely from an ATX connector. More than that and bad stuff has a very good chance of happening. With 6 cards hanging off unpowered risers Yakk could run fine though it is really pushing it as most cards only draw 2-3A off the 12V rail. Running 6 RX 480s off unpowered risers is a pretty big ask though, when each is is pulling 6A. Even with healthy, pristine hardware you're likely going to see damage.

Yeah, 5 cards on unpowered risers was at the top end of what I'd consider safe, I wouldn't suggest 6. I just mentioned 6x for the difference of the 480 is still no big deal. Also I was running an EVGA 1300 G2 PSU which uses a single rail (and cables with built-in capacitors to help smooth ripples on the additional power inputs).

 

[jpiniero]

The silly pcie issue was way overblown, but to their credit, AMD addressed it in record time, +1 to them for it.

You do realize why AMD did this, right? The 480 without the limiter setting really draws 165W which is obviously way too much for 150W power. They basically did this deliberately to make the results as good as possible without reengineering the ref card to use 8-pin power. The difference is only a couple of percent but AMD obviously wants the 480 to be as close to the 1060 as possible in benchmarks.

 

[IEC]

You do realize why AMD did this, right? The 480 without the limiter setting really draws 165W which is obviously way too much for 150W power. They basically did this deliberately to make the results as good as possible without reengineering the ref card to use 8-pin power. The difference is only a couple of percent but AMD obviously wants the 480 to be as close to the 1060 as possible in benchmarks.

A slight undervolt actually improves the average boost clock on the RX 480 due to less throttling. I'd be more inclined to think it's just another case of AMD overvolting their chips yet again. Every GCN card I've owned has had the ability to undervolt at stock clocks, some by as much as 120mV~!

I also don't expect an 8-pin to significantly improve OC results over a 6-pin - it just means an 8-pin card with redesigned power delivery will draw less at the mainboard slot and also still be well within spec on the 8-pin PCI-e power connector. The reference card is more than capable of having even more than 165W pumped into it, if you can cool it (and you bypass AMD's OCP/OV throttling like der8auer did). Of course, this means I'm pessimistic about seeing >1500MHz even on the 8-pin cards... as it's apparent that power delivery isn't the limiting factor on max overclocks. I'd be happy to be wrong, however.

 

[jpiniero]

A slight undervolt actually improves the average boost clock on the RX 480 due to less throttling. I'd be more inclined to think it's just another case of AMD overvolting their chips yet again. Every GCN card I've owned has had the ability to undervolt at stock clocks, some by as much as 120mV~!

Yeah, but AMD's fix doesn't apparently do that - it shifts some power to the power connector and also drops the chip's power from ~110 to ~100. At least according to Tom's review. Presumably not all cards would be able to undervolt and still be 100% stable.

 

[Silverforce11]

Yeah, but AMD's fix doesn't apparently do that - it shifts some power to the power connector and also drops the chip's power from ~110 to ~100. At least according to Tom's review. Presumably not all cards would be able to undervolt and still be 100% stable.

The driver undervolted my cards. By ~25mV or so for stages 4 to 6. The older driver I had 1115mV or stage 6. Now it's 1087mV (~1235mhz to 1250mhz).

That's how they got the extra 3-4% performance because at the new "stock" voltages, it's not hitting the power limit until ~1250mhz or so, whereas before it would throttle and stay around 1190mhz.

There's more headroom left though, it's not close to the edge of stability.

http://www.pcworld.com/article/3093...x-480s-power-draw-and-boosts-performance.html

If they had launch with this driver for tighter voltage control, RX 480 would be closer to 980/390X in performance without the PEG power issue, AMD hurting themselves as usual. :/

 

[el etro]

There's more headroom left though, it's not close to the edge of stability.

Pretty much. It seems that all RX 480 GPUs have some undervolt juice left.

 

[William Gaatjes]

Does anybody know if the third party RX480 cards have been tested on current draw and how the phases are distributed between the pci-e card connector and the pci-e power connector ?

With the reference models there was such an overblown hassle, i wonder where are the tests with the non reference models ? I looked at Tom's hardware but found nothing. PCper.com also seems to lack any follow up about power usage with the non reference models. Have i overlooked something ?

The reason i ask, is that i am interested. I do not worry about a bit more current being drawn through the card slot pins and MB. But when i can be overly careful, why not.