AMD's New Radeon RX 480 Driver Fixes Power Issues

[calken]

The interesting question is: Can they extract more?

Is it possible that they could shave another 10W of power off and increase frame rates by 3%?

The one week turnaround is more like 3-4 weeks as Toms notified AMD about their findings prior to publication, still, 4 weeks for the release of a new driver is seriously impressive.

 

[ohim]

Yet another Radeon driver issue. This has been going on for SO many years. It's one of the reasons I gave up on those AMD/ATI cards years ago.
nVidia for the win!!!

Dude, the issue is with the RX 480 reference cards on old motherboards, the driver actually fixes this issue, there were no driver issue, it was a power delivering issue in the board being fixed by a driver release. Use your brain first.

 

[nitrooo]

How about XFX RX 480 over clocked editions? Did anyone test them? Any links?

 

[gpelamino]

I don't agree with the conclusions of the article at all.

If you want to sell a product that is PCI-Express compliant you should not rely on "compatibility modes" that are not enabled by default to stay within PCI-E limits.

Also the frequency on the technical product sheet should be lowered, because either you define your product to be PCI-E compliant and thus you declare lower frequencies, or you don't and then you're allowed to state the current, higher, frequencies.

You can't declare the product to respect PCI-E requirements and then state the higher frequencies (the one that make the card fail to respect those requirements), that is misinformation

 

[FormatC]

How about XFX RX 480 over clocked editions? Did anyone test them? Any links?

No samples available 🙁

 

[tsnor]

Now that there is a driver fix does that mean benchmarks need to be corrected?

Benchmark results will wiggle a lot over the next few driver releases as bug are fixed (maybe slow things down) and performance improvements are made (definitely speed things up). "corrected" results are likely the ones you'll get in 6 months to a year.

From the current benchmarks it's pretty clear where this card lives performance-wise. +/- 5% would change nothing for me.

I was interested in the REF card OC performance just to see what happened with power consumption and throttling with the new driver modes. I don't think OC performance runs on the ref card will/should change buying decisions.

 

[Onus]

If the issue is how VRM phases are fed, it seems that rerouting the traces that feed them from the PCIe slot to the PCIe connector is really the only long-term way to fix this. Software isn't going to cut it, especially once overclocking is introduced. This is something the board partners will need to address.

 

[JakeWearingKhakis]

*Starts a slow clap

 

[MeanMachine41]

Interesting 3DMark Firestrike results. compared to the GTX-1080. OOPs

http://www.guru3d.com/news-story/radeon-rx-480-performance-benchmarks-with-16-7-1-driver.html

 

[FormatC]

Interesting 3DMark Firestrike results. compared to the GTX-1080. OOPs
Exactly]http://www.guru3d.com/news-story/radeon-rx-480-performance-benchmarks-with-16-7-1-driver.html

Exactly 1.39% difference. 😉

Try five runs of Firestrike and you will get five different results. I don't use it because the failure rate is above one percent. All this results on Guru3D are more or less is within the tolerance range. But it shows, that nothing has been decreased. This is the better conclusion 😉

 

[alextheblue]

I don't agree with the conclusions of the article at all.

If you want to sell a product that is PCI-Express compliant you should not rely on "compatibility modes" that are not enabled by default to stay within PCI-E limits.

Also the frequency on the technical product sheet should be lowered, because either you define your product to be PCI-E compliant and thus you declare lower frequencies, or you don't and then you're allowed to state the current, higher, frequencies.

You can't declare the product to respect PCI-E requirements and then state the higher frequencies (the one that make the card fail to respect those requirements), that is misinformation

I don't mean to sound like a jerk but people need to slow down and comprehend what they're reading. This is the second comment along these lines. Pay careful attention: The driver fixes the PCI-E slot power draw issue in DEFAULT mode at the SAME CLOCKS. It shifts the power over to the extremely capable 6-pin PCI-E cable attached to the PSU. I repeat, the new driver in default mode fixes the PCI-E slot power draw issue TH, without reducing speed.

Now for compatibility mode: Compatibility mode is an OPTIONAL setting that reduces power consumption on BOTH the PCI-E slot AND the 6-pin PCI-E PSU cable, reducing overall power, and it STILL barely budges clocks. I don't think anyone will ever need or want to turn this on, outside of lab testing. But even if you turned it on just to be contrary, the difference in performance is so slight you wouldn't notice it outside of benchmarking.

Was the article seriously not clear enough?

 

[TJ Hooker]

I don't agree with the conclusions of the article at all.

If you want to sell a product that is PCI-Express compliant you should not rely on "compatibility modes" that are not enabled by default to stay within PCI-E limits.

Also the frequency on the technical product sheet should be lowered, because either you define your product to be PCI-E compliant and thus you declare lower frequencies, or you don't and then you're allowed to state the current, higher, frequencies.

You can't declare the product to respect PCI-E requirements and then state the higher frequencies (the one that make the card fail to respect those requirements), that is misinformation

I don't mean to sound like a jerk but people need to slow down and comprehend what they're reading. This is the second comment along these lines. Pay careful attention: The driver fixes the PCI-E slot power draw issue in DEFAULT mode at the SAME CLOCKS. It shifts the power over to the extremely capable 6-pin PCI-E cable attached to the PSU. I repeat, the new driver in default mode fixes the PCI-E slot power draw issue TH, without reducing speed.

Well, according to the PCIe spec, the 6 pin is rated for 75W (even if it's capable of much more), so in that regard you could say the 480 is still exceeding the spec.

 

[InvalidError]

Well, according to the PCIe spec, the 6 pin is rated for 75W (even if it's capable of much more), so in that regard you could say the 480 is still exceeding the spec.

The difference here being that the PCIe slot is only rated up to 1.1A per pin at 90C and are located in an area of the board that usually gets no cooling whatsoever to prevent it from going into thermal runaway while the Molex MiniFitJr pins used for auxiliary PCIe power are rated at 9A each for the standard version and 13A each for the high-current variant despite the PCIe spec only calling for 6.25A.

Why is the PCIe spec so grossly under-rating auxiliary power connectors? Mainly to eliminate the need to worry about thermal derating in typical applications - the MiniFitJr connector is never going to get hot enough to worry about derating when pins are carrying less than 5A RMS each.

 

[bit_user]

Unfortunately, a 3% performance hit would knock the RX 480 down to a VR performance index of 6.4 (below the old R9 290's score of 6.5).

 

[filippi]

Does anyone remember that the reference clocked, PEG powered GTX 750 Ti regularly goes over 75 watts of power usage in Metro Last Light at 1080? The RX 480 is hardly a 4k card, so running a benchmark scenario that would clearly be unplayable seems a bit out of touch and designed to do one thing, show the absolute worst power usage possible. I would love to see the worst case power readings of a PEG powered 750 Ti under the same conditions, if it would even cooperate.

Tom's GTX 750 Ti Power Consumption Charts

The purpose of 4K is to stress the card and give a highest possible situation. Thats why they did 1080P with the 750 Ti even though it is not a 1080P GPU.

750ti was not a 1080p gpu? Now you really hurt my feelings 😀

 

[FormatC]

The 750 Ti has a power consumption of 62 Watts on PEG, less than 5.3 A in average and full within the specs. The spikes were a lot higher, especially the Asus card was ugly, but this is a another story. If the ppl are not able to read charts.... Not my fault

 

[RedJaron]

It seems there were a few hand-made cards and someone shared info, pics and hardware.

So some of the 480 cards sent out for reviews were indeed one-offs and significantly different than what you will see in a retail product? Does this mean the mass produced cards will see smaller problems than the 480 you reviewed ( driver fix or not )?

 

[FormatC]

I compared my card with retail reference - no difference. I would suggest to buy a custom design with an 8-pin connector. 😉

 

[littleleo]

The interesting question is: Can they extract more?

Is it possible that they could shave another 10W of power off and increase frame rates by 3%?

The one week turnaround is more like 3-4 weeks as Toms notified AMD about their findings prior to publication, still, 4 weeks for the release of a new driver is seriously impressive.

Yes for AMD 4 weeks is awesome I believe it had to be in the works all a long. I still have customers waiting for a Xpress 200 driver update, shhh I still have nightmares about those.

 

[jimmysmitty]

The interesting question is: Can they extract more?

Is it possible that they could shave another 10W of power off and increase frame rates by 3%?

The one week turnaround is more like 3-4 weeks as Toms notified AMD about their findings prior to publication, still, 4 weeks for the release of a new driver is seriously impressive.

Yes for AMD 4 weeks is awesome I believe it had to be in the works all a long. I still have customers waiting for a Xpress 200 driver update, shhh I still have nightmares about those.

I actually remember being excited when ATI announced they were going to start making chipsets for intel. Was hoping for something new and faster.

Then they got bought out by AMD and that killed that idea.

 

[P1nky]

"Hopefully board partners skirt the issue entirely by using an eight-pin PCIe connector."

RX 480 is wired to use the 6-pin connector as an 8-pin (250W max) !!!!!!!!

How can you write this analysis without not knowing this?! I have to take your reporting with a tub of salt because of this.

 

[Olle_P]

In my mind this fix is "a little too little".
1. If it needs this much power then a larger part of it could be pulled off the 3.3V line to start with. (Probably impossible to change by software.)
2. The shift of 12V current from motherboard to external connector should be bigger. Seems suitable to have only ~35% of the 12V power drawn from the motherboard to keep the current decidedly below 5.5A at 160W 12V consumption.
3. The compatibility mode, which I consider an effort to contain the total power consumption within the 150W TDP, should also be a little more aggressive to actually achieve the goal. As of now it seems easy to make consumption stay above 150W over an extended period.

At the same time I don't like Tom's choice of Metro @ 4k for evaluating the power consumption. The reason being that this obviously isn't a card suitable for those settings. A remedy would have been if you actually (manually) played (and enjoyed) the game with those settings while the power was measured.
The performance of this card is great for 1080p and good for 1440p. Use a typical game with *playable* settings for "real life" tests, please!

 

[FormatC]

"Hopefully board partners skirt the issue entirely by using an eight-pin PCIe connector."
RX 480 is wired to use the 6-pin connector as an 8-pin (250W max) !!!!!!!!
How can you write this analysis without not knowing this?! I have to take your reporting with a tub of salt because of this.

I can. A lot of SI are using standard 6-pin PCIe on their low-cost PSUs with only two wires connected.A standard is a standard and the PCI SIG was also asked about the logo. AMD has never requiered a logo and never will not get one for this card. And BTW, the 8-pin allows you max. 150 Watts due two additional ground wires.

The performance of this card is great for 1080p and good for 1440p. Use a typical game with *playable* settings for "real life" tests, please!

I tried. Metro Last Light in 1440p with only high settings (90 FPS) takes also over 160 watts, Risen 3 in 1440p up to 170 watts! Other sites measured the same with other games.

 

[InvalidError]

And BTW, the 8-pin allows you max. 150 Watts due two additional ground wires.

The official PCI-SIG spec says that. The Molex spec on the other hand says that the 6-pin connector is fine up to 9A per standard pin and 13A for high-current variants, which are about 300W and 450W respectively.

The PCIe spec is being extremely conservative by specifying a current that's only 1/4 of what the connector's lower-end spec is actually capable of. The two extra grounds on the 8-pins connector don't really add much in terms of useful current-carrying capacity since up to two of them may be wasted on connector identification anyway.

 

[FormatC]

And BTW, the 8-pin allows you max. 150 Watts due two additional ground wires.

The official PCI-SIG spec says that. The Molex spec on the other hand says that the 6-pin connector is fine up to 9A per standard pin and 13A for high-current variants, which are about 300W and 450W respectively.

The PCIe spec is being extremely conservative by specifying a current that's only 1/4 of what the connector's lower-end spec is actually capable of. The two extra grounds on the 8-pins connector don't really add much in terms of useful current-carrying capacity since up to two of them may be wasted on connector identification anyway.

The idea behind the standards of the PCI SIG is, that the connector isn't delivering pure DC, but a kind of HF current with extreme and fast load level changes (up to 100 KHz). We also tested it internaly with a good lab PSU and resistors - no problem. But the temperatures are going up, if you are using pulsing current, especially on older, used plugs and sockets. The second fact is, that nobody knows, which cable quality is used on the PSU side. AWG 16, 18 or 20? In the case of AWG 20 and two wires I wish you a lot of fun and a good weather if you make the same test 😉

The discussion about industrial standards us useless. And AMD has never requested a PCI logo and the PCIE SIG also said, that this card will not pass the certification with this current specs.