The 480: power consumption, PCI-E powerdraw

[William Gaatjes]

Well, in this case you're actually more interested in the input capacitors than the output capacitors of the VRM. They're they ones that really help suppress input ripple, along with an input inductor if the designer chose to include one.The input capacitors aren't as sexy spec-wise and they often don't get the attention they deserve in VRM design, but they have a massive impact on input ringing and stability. Often the layout of the input capacitor plays second fiddle to optimizing the output; sometimes to the detriment of overall noise. Josh Mandlecorn from TI wrote an interesting app note on the effects of input capacitance positioning and sizing and its effect on input current ripple; I should see if I can dig it up.

This is totally true.
The input capacitor and dampening network is widely ignored but helps a lot with reducing switching noise.

For example, Linear has a buck smps that has a controllable slewrate on one of their smps controllers. The result is that together with a magnetically shielded inductor and proper input capacitors, there is almost no radiated emi.

http://www.linear.com/product/LT8614

http://www.linear.com/product/LT8641

http://www.linear.com/parametric/Ultralow_Noise_Regulators

 

[Doom2pro]

Well, in this case you're actually more interested in the input capacitors than the output capacitors of the VRM. They're they ones that really help suppress input ripple, along with an input inductor if the designer chose to include one.The input capacitors aren't as sexy spec-wise and they often don't get the attention they deserve in VRM design, but they have a massive impact on input ringing and stability. Often the layout of the input capacitor plays second fiddle to optimizing the output; sometimes to the detriment of overall noise. Josh Mandlecorn from TI wrote an interesting app note on the effects of input capacitance positioning and sizing and its effect on input current ripple; I should see if I can dig it up.

Yes, the input capacitors to the VRMs are for voltage regulation stability... A VRM can only stabilize output voltage relative to a known voltage, and having a noisy input voltage will cause the regulation loop to go a little crazy.

Output capacitors on VRM are just to smooth out the very well known ripple caused by the switching circuit that drives the VRM.

You are right about the input capacitors being more important here, because the input ripple is more of an unknown and it needs to be filtered adequately or output voltage will suffer.

 

[William Gaatjes]

They will just cost a little more by a few cts for the whole...

I mean there are physical limits. No way that i believe that these inductors can handle 60A each. Maybe if they are two turns of 4mm wire and an inductance of a several nanoHenry. I doubt that is the case. And then i ignore the core material as well.

 

[Erenhardt]

There is no cost associated with trace size, so unless the board is super space constrained you'll probably find that the traces are oversized if they are delivering power. Or you might find the connector/component is directly connected to a power plane.

There is additional cost to every part.

As a rule of the thumb in designing electric circuits is to have connectors overbuild to accommodate for human error/pin damage/debris decreasing contact area.

In order to burn out the atx power connector, the connector needs to damaged or not plugged correctly. Otherwise overcurrent protection will kick in, or traces/wires will smoke.

 

[MrTeal]

What is also the issue, is that these are smps. Switching power supplies. That means they radiate a lot emi. What MrTeal means is that your oscilloscope probe can also pick these emi signals up and make them look larger than they really are. Even professional measurement tools can fall victim to this.
An easy and cheap way to measure the voltage on smps power supplies is to use a twisted cable with a small capacitor to filter the really hf and connect that to the signal you want to measure to an oscilloscope probe, far away from the smps. It really works. the emi cancels out more or less because of the twisted cable.

Yeah, the very first VRM I layed out had test points at the various interesting voltage points, and some ground pins around to attach the probe lead to. It was terrible, measurements were useless I squeezed the loop of the ground clip together to try to minimize loop area. Now everything just has properly spaced voltage and ground pads/vias beside each other so I can stab them with the scope probe with a grounding spring.

 

[William Gaatjes]

Yes, the input capacitors to the VRMs are for voltage regulation stability... A VRM can only stabilize output voltage relative to a known voltage, and having a noisy input voltage will cause the regulation loop to go a little crazy.

Output capacitors on VRM are just to smooth out the very well known ripple caused by the switching circuit that drives the VRM.

You are right about the input capacitors being more important here, because the input ripple is more of an unknown and it needs to be filtered adequately or output voltage will suffer.

In a smps, the output capacitor is an integral part of the system. I would not dare to say that it is just for filtering ripple. It is not a 50Hz Greatz circuit.
Of course, the ESR and ESL of the capacitor determine some ripple. But there is more to it than that.

 

[Abwx]

I mean there are physical limits. No way that i believe that these inductors can handle 60A each. Maybe if they are two turns of 4mm wire and an inductance of a several nanoHenry. I doubt that is the case. And then i ignore the core material as well.

https://www.coilws.com/Surface Mount Products/High_Amps_SM.htm

 

[Doom2pro]

In order to burn out the atx power connector, the connector needs to damaged or not plugged correctly. Otherwise overcurrent protection will kick in, or traces/wires will smoke.

Yes this seems to be the problem with the misinformation going around about the RX 480's PCI-E power draw... People don't realize that components and specifications have inherent derating applied to their maximum specifications.

You NEVER want to set your Max to the Failure Point, you always want a decent amount of headroom in your spec otherwise your MTBF will be very short!

 

[Doom2pro]

In a smps, the output capacitor is an integral part of the system. I would not dare to say that it is just for filtering ripple. It is not a 50Hz Greatz circuit.
Of course, the ESR and ESL of the capacitor determine some ripple. But there is more to it than that.

Wait, I'm talking about filtering voltage ripple cause by the DC-DC not noise ripple from somewhere else...

You do know that DC-DC supplies have a "chopper" circuit, they drive high frequency currents into an inductor (with or without secondary) and these AC currents need to be rectified and then filtered with capacitors.

This chopper induced ripple is what I was referring to.

 

[Abwx]

You do know that DC-DC supplies have a "chopper" circuit, they drive high frequency currents into an inductor (with or without secondary) and these AC currents need to be rectified and then filtered with capacitors.

What is filtered are the voltages varations, not the currents..

 

[Doom2pro]

What is filtered are the voltages varations, not the currents..

Where did I say currents? I said VOLTAGE RIPPLE... The AC Currents I mentioned was only for the Rectification to DC part, I assumed the rest was obvious.

LOL what do I have to do? Write a DC-DC tutorial to avoid everyone getting confused?

 

[William Gaatjes]

https://www.coilws.com/Surface Mount Products/High_Amps_SM.htm

I do not know what you are trying to prove. Of course there are high current inductors, i am well aware of them. If i take a random pdf from the ones you link they are ~14.4mm by ~13.2mm by 5.3mm. They are different than the ones used on the rx 480.

If you can get the datasheet of those inductors used on the rx480,. That is interesting. Because then you can say something about the theoretical maximum output current. The inductor must never saturate.

 

[railven]

Looks like AMD issued a statement and will be issuing a driver fix :




Much ado about nothing IMHO.

Was just the most plausible answer.

Haha this topic has been dubbed PowerGate, at least over on r/AMD.

People are digging up specifications and documentations on this one. Haha.

One guy made a comment, it spread like wildfire, he updated his comment sort of redacting his prior comment, crickets to his update.

I'm sure AMD is safe, I doubt they'd release a product knowing it was out of spec. Now, them releasing a buggy driver - that I believe.

This could all be Wattmans fault for all we know.

AMD introduces new power management feature with new cards, new cards go ape on Power draw. It makes sense to me.

For all AMD does right, rolling out new software features is hit-or-miss. And I say this from previous software bugs that broke functionality of hardware features in the past, not as a "AMD drivers sucks LOLZ!!1!!"

Ultimately it was the break of ZeroCore when they trotted out Frame Pacing that made me sell off my 7970s.
[EDIT: To elaborate, there was a bug with frame pacing drivers that stuck your second GPU at max clocks even if in idle and never activating ZeroCore (which should disable the card). The only solution was to disable ULPS which negated ZeroCore and left you with two idling cards and with the two 7970s, that is a good amount of heat just for browsing the internet..)



The fallout is the unexpected focus on the power issues. But I guess that comes on the heels of all the "are most efficient technology ever" and if you ask me that AMD rep in the reddit didn't help by stating "oh we were talking about RX 470" because it felt like a bait and switch.

Here is to this ugly mess being behind them. Hopefully the fix improves performance as some other cards might have been throttling due to needless overvolting.

 

[Doom2pro]

I do not know what you are trying to prove. Of course there are high current inductors, i am well aware of them. If i take a random pdf from the ones you link they are ~14.4mm by ~13.2mm by 5.3mm. They are different than the ones used on the rx 480.

If you can get the datasheet of those inductors used on the rx480,. That is interesting. Because then you can say something about the theoretical maximum output current. The inductor must never saturate.

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.

 

[William Gaatjes]

Wait, I'm talking about filtering voltage ripple cause by the DC-DC not noise ripple from somewhere else...

You do know that DC-DC supplies have a "chopper" circuit, they drive high frequency currents into an inductor (with or without secondary) and these AC currents need to be rectified and then filtered with capacitors.

This chopper induced ripple is what I was referring to.

I also talk about the output ripple. That is the same as what you call chopper induced ripple. This ripple is dependent from (the top of my head) the inductor(R and L), the output capacitor (ESR and ESL). The switching frequency and the delay that the pwm controller has before it initiates another switching cycle.

 

[MrTeal]

I mean there are physical limits. No way that i believe that these inductors can handle 60A each. Maybe if they are two turns of 4mm wire and an inductance of a several nanoHenry. I doubt that is the case. And then i ignore the core material as well.

The Cooper Bussman FP1007 I usually use have an RMS current rating of 61A and an Isat of 60A@125C for 170 nanohenries. They're actually very similar to the ones used in the 290X design, though they use the 150nH version.
Edit: datasheet

There is additional cost to every part.

As a rule of the thumb in designing electric circuits is to have connectors overbuild to accommodate for human error/pin damage/debris decreasing contact area.

In order to burn out the atx power connector, the connector needs to damaged or not plugged correctly. Otherwise overcurrent protection will kick in, or traces/wires will smoke.

That's probably the case, but keep in mind that if the RX480 is averaging ~7A on the 12V rail through the motherboard, three of them is already 31% over the 16A (8Ax2) rating of even the top tier Minifit Plus HCS contacts.
At 31% or more over your contact rating, it doesn't necessarily take much to cause an issue especially in a mining environment where it's 24/7 and often in quite elevated ambients.

 

[William Gaatjes]

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.

That is true. Unfortunately, the datasheet for the controller is a bit of a secret.
The ones available is very global and gives no explanation at all. And since it is a digitally programmable controller, the switching frequency is not fixed but programmable.

 

[Doom2pro]

I also talk about the output ripple. That is the same as what you call chopper induced ripple. This ripple is dependent from (the top of my head) the inductor(R and L), the output capacitor (ESR and ESL). The switching frequency and the delay that the pwm controller has before it initiates another switching cycle.

Yes the output stage of any DC-DC is basically a high power low-pass filter.

 

[William Gaatjes]

The Cooper Bussman FP1007 I usually use have an RMS current rating of 61A and an Isat of 60A@125C for 170 nanohenries. They're actually very similar to the ones used in the 290X design, though they use the 150nH version.
Edit: datasheet

Aha, that makes it more clear. Specifications are an undeniable truth.
I often use Wurth Elektronik inductors.

 

[William Gaatjes]

Yes the output stage of any DC-DC is basically a high power low-pass filter.

Very simply put yes.

 

[Doom2pro]

That is true. Unfortunately, the datasheet for the controller is a bit of a secret.
The ones available is very global and gives no explanation at all. And since it is a digitally programmable controller, the switching frequency is not fixed but programmable.

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

 

[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.

 

[MrTeal]

That is true. Unfortunately, the datasheet for the controller is a bit of a secret.
The ones available is very global and gives no explanation at all. And since it is a digitally programmable controller, the switching frequency is not fixed but programmable.

I was actually planning on poking around a little with the VRM. My IR DPDC subscription has lapsed, but it stops yelling at me if I set my computer's clock back to a time when it was valid. Hopefully my installed version is new enough to have that chip in the list.

 

[Doom2pro]

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.

 

[Abwx]

That is true. Unfortunately, the datasheet for the controller is a bit of a secret.
The ones available is very global and gives no explanation at all. And since it is a digitally programmable controller, the switching frequency is not fixed but programmable.

Possibly this one, wich get up to 1MHz.

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