The 480: power consumption, PCI-E powerdraw

[sm625]

Good article today over at Tomshardware http://www.tomshardware.com/reviews/amd-radeon-rx-480-power-measurements,4622.html

From the summary, page 2,

5. Load peaks might result in audible interference because of nonlinearities in the analog audio section.

Hoo boy... so it wont blow up your motherboard, but it might blow up your speakers!

 

[zinfamous]

From the summary, page 2,



Hoo boy... so it wont blow up your motherboard, but it might blow up your speakers!

"audible interference" means scratchy noise or buzzing and is a far cry from "blow up your speakers"

 

[MrTeal]

I agree, but we don't have that direct comparison.

Maybe it could be extrapolated from this:

So it's still a substantial gain in performance per watt. Considering the more VRAM.. I still think it gets close to 2x.

Using those numbers, the 380X has 25% higher performance per watt than the 285.
480 - 0.337 FPS/W
380X - 0.193 FPS/W
285 - 0.154

If you want to compare, I wouldn't use the single score FO4 benchmark to compare Polaris to Tonga, as it's a single data point. I would use the ratio of 380X to 285 (single it's the same chip), and then scale the TPU data by that amount. IE, scale 55 on the TPU chart by 1.25 to get 69 for the 380X, and then compare that to Polaris across the cross section of games.

 

[William Gaatjes]

From the summary, page 2,



Hoo boy... so it wont blow up your motherboard, but it might blow up your speakers!

5. Load peaks might result in audible interference because of nonlinearities in the analog audio section.

What you quote is what i wrote before. The large current spikes can induce currents in nearby high impedance circuits. Analog circuits usually are high impedance. That is a possibility, but not the only one.

It is also normal that when the gpu gets active, the main psu has to deliver more power. This can be apparent as more switching noise on the other voltage rails. A MB with the analog audio section powered directly from such a rail could start to produce noise at the moment a lot of power is delivered by the main psu. When the analog part of the audio section has its own linear voltage regulators, removing any switching noise and creating a clean voltage rail for the audio, the chance of noise is a lot less to zero when the audio circuit is designed with proper filtering and shielding. Now this is quite a feature on a populated Motherboard.

Also, when the cpu is idle and starts up to do some work, it draws more current in spikes. It is not uncommon that when you move your mouse or move a window, that you can hear that when turning the volume level up.

And price says nothing about how good it is.
My cheap mobo with onboard audio, the Gigabyte GA-F2A75M-HD2 (rev. 1.0) (with A10-6700 APU) is very silent.
I have to turn my amplifier to 100% volume before i can hear on my headphones in the backgound any "zzzzz" noise when i move a windowed program.

Forgot my key note :
This is nothing special and not something that will only occur with an RX 480.
It can happen with any cpu and gpu.

 

[SolMiester]

Why don't they fix it for the people who own O/C'd monitors who bought their cards? Would it kill them? Would likely cost them a minimal amount to sort that.

Apparently they have, but their missed driver release coming, so it will have to be a patch for now until it comes out in the driver.

So yes, they have written a fix for those out of spec.

 

[sirmo]

Using those numbers, the 380X has 25% higher performance per watt than the 285.
480 - 0.337 FPS/W
380X - 0.193 FPS/W
285 - 0.154

If you want to compare, I wouldn't use the single score FO4 benchmark to compare Polaris to Tonga, as it's a single data point. I would use the ratio of 380X to 285 (single it's the same chip), and then scale the TPU data by that amount. IE, scale 55 on the TPU chart by 1.25 to get 69 for the 380X, and then compare that to Polaris across the cross section of games.

I think you're being way too generous to the 380X It's less than 10% faster than a 380, so I have a hard time believing it has a 25% better perf/watt. Even with disabled cores they should be consuming the same amount of power (worst case).

Something is not making sense, I suspect is the more efficient Samsung VRAM later versions of 380X use, that could be causing different results.

 

[Yakk]

"audible interference" means scratchy noise or buzzing and is a far cry from "blow up your speakers"

Correct, and most cheap integrated audio chipsets already have a lot of noise so I would guess very few people would even notice anything more than they already have.

 

[garagisti]

Why don't they fix it for the people who own O/C'd monitors who bought their cards? Would it kill them? Would likely cost them a minimal amount to sort that.

I'd be happy if they dropped the proprietary bs and followed industry standards. wishes were horses, riding, something.

 

[sm625]

http://www.legitreviews.com/amd-radeon-rx-480-undervolting-performance_183699

That voltage spike is extremely worrisome. In their article they just kind of dismiss it like nothing happened.

 

[Erenhardt]

error reading probably.

 

[TeknoBug]

75W being the PCI-E limit, how the hell is it still drawing more than that? I guess that's one of the perks of not being an early adapter, hopefully that issue will be addressed by the time custom PCB releases comes around.

 

[Flapdrol1337]

75W being the PCI-E limit, how the hell is it still drawing more than that? I guess that's one of the perks of not being an early adapter, hopefully that issue will be addressed by the time custom PCB releases comes around.

5 of the pins of the pci-e connector have 12V on them, they're connected to the 2 12V pins on the atx connector. So it's pretty much directly connected to the psu.

If something draws enough current it'll go over the 75W as the psu can easily handle that. It just goes over the specced maximum of the pci-e connector.

 

[Bacon1]

Drivers are out that fix the issue: http://support.amd.com/en-us/kb-art...are-Crimson-Edition-16.7.1-Release-Notes.aspx

 

[Stuka87]

So now to wait for some reviews of the new driver.

 

[Glo.]

http://www.tomshardware.de/amd-radeon-rx-480-polaris-grafikkarte-treiber,testberichte-242146.html

 

[Stuka87]

http://www.tomshardware.de/amd-radeon-rx-480-polaris-grafikkarte-treiber,testberichte-242146.html

Have not seen the english version come up. But their graphs are very confusing. They have "PCIe Power" but then also have "Mainboard Power". So I am assuming PCIe power is actually the 6 pin ATX connector?

 

[IEC]

Have not seen the english version come up. But their graphs are very confusing. They have "PCIe Power" but then also have "Mainboard Power". So I am assuming PCIe power is actually the 6 pin ATX connector?

Yes that's correct. If you need a specific portion of the article translated I'd be happy to help.

 

[Erenhardt]

No performance comparison?

 

[SPBHM]

looks a lot better, also the "compatibility mode off" has the same or slightly higher clocks than before it seems, so it should perform the same or a little better

 

[redzo]

AMD driver release quote:

... popular game titles of up to 3%[1] ...

And know we know where the 3%1 comes from.
It was an important public announcement. Some idiot PR dude copy/pasted this without a proper review. How the hell do these people get hired?

 

[poofyhairguy]

Yes that's correct. If you need a specific portion of the article translated I'd be happy to help.

That looks fixed. I would love to see an OC value though to see where the OC wattage come from.

 

[Stuka87]

AMD driver release quote:

And know we know where the 3%1 comes from.
It was an important public announcement. Some idiot PR dude copy/pasted this without a proper review. How the hell do these people get hired?

Its a copy paste issue. The 1 should be subtext. Not sure why people keep pointing this out, its pretty clear it just did not copy in correctly.

 

[Erenhardt]

AMD driver release quote:

And know we know where the 3%1 comes from.
It was an important public announcement. Some idiot PR dude copy/pasted this without a proper review. How the hell do these people get hired?

What are you thread crapping about now?


Personal attacks are not allowed.

-Rvenger

 

[MrTeal]

That looks fixed. I would love to see an OC value though to see where the OC wattage come from.

It will come from each in proportion to the extra needed. They just changed the gains of the three phases attached to the 6 pin so they take more of the load, but the load is still shared.

 

[William Gaatjes]

http://www.legitreviews.com/amd-radeon-rx-480-undervolting-performance_183699

That voltage spike is extremely worrisome. In their article they just kind of dismiss it like nothing happened.

That is a bit strange. Could be a measurement error. It can not be a load dump effect. That happens when a lot of current is drawn and suddenly, the gpu goes idle. But that would be only for microseconds long and not seconds. Assuming they did not do any creative data massage.

The gpu tells the smps vrm IR3567B via a serial vid interface (SVI) what voltage is needed. I have no idea how the gpu does that in detail. I assume there is some programmable dedicated power management controller (maybe like a embedded arm core ) that constantly is monitoring the power requirements. Others on this forum know much more about it then i do. This controller has obviously firmware that is loaded by the gpu driver.

That spike is concerning, though...