The 480: power consumption, PCI-E powerdraw

[William Gaatjes]

Yes but measurable with an oscilloscope... Just probing the gates of the Mosfets should give you a good indication.

It is very possible that it is a fixed frequency controller. But there is a limit to achieve regulation over a wide range of power output from say for example 5watt to 100w. I wonder how they would do that. And with such small inductance inductors. Amazing.

I know of several solutions, but i wonder if that is applicable for a gpu core smps. I wander if they use pulse skipping when the load is too low when the gpu is not doing much and low clocked. That would cause for more output ripple but since the gpu is not doing anything much , it is not an issue when it comes to leakage current. Very interesting to wonder about.

 

[Doom2pro]

Possibly this one, wich get up to 1MHz.

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

That controller has integrated mosfets... Reference RX 480s are using discrete SO8 mosfets.

 

[MrTeal]

Possibly this one, wich get up to 1MHz.

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

That's not a controller, it's an output device. A very excellent one at that, I've used them extensively and love them. Even then though, you don't want to run at 1MHz. Take a look at Figure 12, Normalized Power Loss vs Switching Frequency.

That's also an integrated device which are much easier to clock high due to significantly reduce switch node inductance. The RX480 uses discrete high and low side devices separated a ways with the driver between. Clocking those at 1MHz without excessive ringing would be a challenge.

 

[William Gaatjes]

Possibly this one, wich get up to 1MHz.

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

That is not the controller i mean ((IR3567) but thank you.
I was looking for integrated power stages.

Maybe i am wrong, when i looke at the picture and zoom , instead the mosfets look like discrete mosfet parts in a son-8 package. But i could be wrong. What do you think ?

 

[Doom2pro]

Clocking those at 1MHz without excessive ringing would be a challenge.

Yes especially since you cannot put a ferrite bead on one of the legs (because it's SO8)... I don't think anyone uses SMD inductors with SO8 except maybe for the gate?

 

[Doom2pro]

Maybe i am wrong, when i looke at the picture and zoom , instead the mosfets look like discrete mosfet parts in a son-8 package. But i could be wrong. What do you think ?

They look like SO8 to me...

 

[Abwx]

That is not the controller i mean ((IR3567) but thank you.
I was looking for integrated power stages.

Maybe i am wrong, when i looke at the picture and zoom , instead the mosfets look like discrete mosfet parts in a son-8 package. But i could be wrong. What do you think ?

I know but what matters are the switching devices, speed is not limited by controlers.

According to the pic it looks like mosfets and separate controlers to drive them.

 

[MrTeal]

That is not the controller i mean ((IR3567) but thank you.
I was looking for integrated power stages.

Maybe i am wrong, when i looke at the picture and zoom , instead the mosfets look like discrete mosfet parts in a son-8 package. But i could be wrong. What do you think ?

I believe so. If they're using Infineon fets to go with the IR controller they call their package TDSON-8 for most of the optiMOS family, but the exact package isn't really as important as to whether they're using discrete or integrated devices. Looking at the images, it's pretty obviously the former.

 

[Erenhardt]

according tho him:
https://www.youtube.com/watch?v=qG2e-v94L4M

it is this one:
http://www.infineon.com/dgdl/pb-ir3567b.pdf?fileId=5546d462533600a4015356803a7228ef

 

[Doom2pro]

according tho him:
https://www.youtube.com/watch?v=qG2e-v94L4M

it is this one:
http://www.infineon.com/dgdl/pb-ir3567b.pdf?fileId=5546d462533600a4015356803a7228ef

Switching frequency from 200kHz to 2MHzper phase

Wow that's pretty high... 2Mhz...

 

[Erenhardt]

He livestreams overclocking and modding 480:
https://www.twitch.tv/buildzoid

Will we see motherboard fireworks live?

 

[William Gaatjes]

Yes especially since you cannot put a ferrite bead on one of the legs (because it's SO8)... I don't think anyone uses SMD inductors with SO8 except maybe for the gate?

Well, if you want to control ringing, there is a way to do it even when you have smt mosfets. Wurth Elektronik has a nice application note about it :

"Ringing” is a common term referring to the undesired oscillation that occurs when a power semiconductor switch turns on or off in the presence of parasitic inductance and capacitance. Energy stored in the parasitic junction capacitance of the switch is released during the switching transition and rings with parasitic inductance coming from the stray fields of discrete power inductors and the wiring inductance of the PCB traces, component leads, connectors, etc. In real circuits on real circuit boards parasitics are always present, and hence all switching converters produce at least some ringing.


Summary
Ferrite beads placed in series with the bootstrap pin of buck regulators can make excellent ringing control components, and when applied properly they can be used to treat the source of the high frequency noise without taking up too much PCB area and without large sacrifices in power efficiency. Their benefits over tuned resistors are in ease of selection and reduction in the lab testing time needed to design them in.

PDF :

http://www.we-online.com/web/en/ind...rung/Ferrite_Beads_for_Ringing_Control_EN.pdf

 

[William Gaatjes]

Wow that's pretty high... 2Mhz...

Yep, that is the datasheet. But it does not say much. It does however mention something extremely interesting. Just like a 8 cylinder engine can turn of 2 or more cylinders (by uses of the ECU disabling fuel injection) while cruising to increase fuel economy, this six phase pwm controller can turn off phases one by one when there is only a light output load. That is very interesting.

At 2 MHz, with all 6 phases enabled, it effectively behaves like a 12MHz smps because i am positive that all the phases are turned on in a sequence with a 60 degrees phase shift.

 

[Bacon1]

Interesting testing @ semiaccurate:

Basically, either go +10% for extra perf reasons, or -20% for best perf/watt

There’s still a performance gain to be had from bumping it up to +10% but it’s rather small. The -10% and -20% settings also have a very limited impact on the performance of the RX 480

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

Tons of undervolting while overclocking results here:

https://www.reddit.com/r/Amd/comments/4qupw4/super_psa_all_rx480_owners_please_attempt_to/

 

[pcmax]

ASUS 8GB RX 480 in hand. Hasn't killed my PCI-e slot yet on a Gigabyte H81 motherboard with a Corsair CX 430 watt PS FYI. Welcome to tests, just let me know. Playing DOOM for an hour (overheats my GTX 770 system to needing my side cover off) so far so good on this system with squirrel fan blowing out the back.

 

[Silverforce11]

The first proven casualty of RX 480s.. 3 of them actually.

While mining ETH on a el-cheapo MB, on non-powered (!!) risers.

https://bitcointalk.org/index.php?topic=1433925.msg15438155#msg15438155

He made the noob mistake of mining multi GPU without powered-risers and paid the price. hehe

To be fair, this could happen for other multi-GPU mining setups as well and it did in the past. This is why miners pay a few $ extra for molex-powered risers to supply additional power to the GPU, rather than going through the MB.

 

[dogen1]

What are you talking about?

They said that it is giving too much power. They made a whole new power management software and it wasn't setting the settings properly. They are fixing the issue in a driver / software update which will tell it to use less power.

Lots of people are undervolting their cards and getting great results:

[/QUOTE] And why weren't they ab... major launch? It's odd for a big company.

 

[Doom2pro]

And why weren't they able to get it right for a major launch?

It's odd for a big company.

No launch is without it's issues... nVidia has issues for it's Pascal cards also...

It happens but thankfully for AMD, they seem to have fixed it in a future driver release.

This whole situation is a classic mountain out of a mole hill... Blown out of proportion. In a few months nobody is going to remember this, well except for some nVidia fanboys

 

[DarkKnightDude]

The first proven casualty of RX 480s.. 3 of them actually.

While mining ETH on a el-cheapo MB, on non-powered (!!) risers.

https://bitcointalk.org/index.php?topic=1433925.msg15438155#msg15438155

He made the noob mistake of mining multi GPU without powered-risers and paid the price. hehe

wtf. If you're going to be getting three cards for mining you should get powered-risers at the very least. Shouldn't be a huge cost issue for that.

 

[Bacon1]

And why weren't they able to get it right for a major launch?

It's odd for a big company.

Because software is very tricky, especially for "Auto" adjusting. I'd know, I'm a programmer .

Nvidia's 1070/1080 won't boot with dvi monitors over 85(?)hz right now.

Both companies have had major driver issues which have locked clock fan speeds, or in Nvidia's case, completely disabled them (multiple times over the years too!).

Both companies have issues with software, hell, I can't think of any software that isn't 100% bullet proof.

 

[mxnerd]

AMD To Issue Software Fix To Address RX 480 Power Consumption Problems

http://www.tomshardware.com/news/amd-power-gpu-rx-480,32200.html

 

[torlen11cc]

I don't understand that issue.
Many GPUs draw more than 75W, like 950 (no power connector):

so... what is the problem with rx 480?

 

[oussama-tn]

I don't understand that issue.
Many GPUs draw more than 75W, like 950 (no power connector):

so... what is the problem with rx 480?

Because it's Amd, because it should be running hot and bad @ p/w
Seeing the results of undervolting the card now just draws more and less 125W. I dont see the meaning of discussing this anyfurther

 

[littleg]

That's the peak power consumption. As I understand it the problem isn't with the peak power, it's a problem if the card is continuously drawing more power than the slot is rated for.

I believe the spec is 75W +/- 8% also which puts the 950 within the margins.

 

[SPBHM]

Because it's Amd, because it should be running hot and bad @ p/w
Seeing the results of undervolting the card now just draws more and less 125W. I dont see the meaning of discussing this anyfurther

that GTX 950 seems like it could be like the 480 if you OC it, but, that's a pretty rare card with no 6 pin connector, reference 950 and the vast majority of the 950s I've seen so far have the extra 6pin and will not have the problem, if they balance the power load like the 960 (which they probably do)

but seriously? you don't see why this discussion is going on?
you mean you think it's fine if you have to manually undervolt you card because at stock settings it's greatly exceeding the PCIE slot specs, really!?

check the PCPerspective article or video.
the specs for the 12v slot connector are around 66W, they tested a 380 and it used around 50W, they tested a 960 and it used around 25W, while the 480 uses easily 10-20W over 66W

this is not about being hot or to power hungry, it's about drawing to much power from the 12v pins from the PCIE slot when it could safely use the 6 pin connector (in which exceeding official PCIE specs is not a big deal) for that.