Wednesday, August 17th 2016

High PCIe Slot Power Draw Costs RX 480 PCI-SIG Integrator Listing

AMD's design of the Radeon RX 480 graphics card, which draws over 75W of power from the PCI-Express x16 slot, has cost it a product listing on the PCI-SIG Integrators List. The list is compiled for hardware devices implementing the various PCI-Express specifications to the letter. The RX 480 is off-spec, in that it overdraws power from the slot, as the card needs more power than what the slot and the 6-pin PCIe power connector can provide while staying within specs. According to these specs, the slot can provide up to 75W of power, and the 6-pin connector another 75W. The RX 480 was tested to draw more than this 150W power budget.

What this means for AMD is that it cannot display the PCI-Express certification logo on the product or its marketing materials. This, however, may not affect AMD's add-in board (AIB) partners that are PCI-SIG members in their own right, and make graphics cards with their own sub-vendor IDs, provided their power-supply designs comply with PCIe specs. Custom-design cards with an 8-pin PCIe power connector, instead of 6-pin, may qualify as the combination of the 8-pin connector and the slot yields a power budget of 225W. AMD released a software fix to the issue of the cards overdrawing power from the slot, with the Radeon Software Crimson Edition 16.7.1 Beta.
Source: Heise.de
Add your own comment

63 Comments on High PCIe Slot Power Draw Costs RX 480 PCI-SIG Integrator Listing

#26
TheinsanegamerN
JalleR, post: 3507480, member: 50355"
And it is stamped OK as a standard by PCI-SIG ? last time i was reading about it the 2x8 wasnt a standard in their papers, but it is maybe ½ a year ago so it could have changed.

The reason for not putting the 8pin is 1. bad design not knowing the total power draw of a full card. or 2. it will look bad that their "medium" graphic card is needing one 8pin power connector when other brands are only using one 6 pin.

And yes with the fix it is still not compliant it is 75 Watt from the slot and 75watt from 6 pin, not 84Watt from the 6 pin.
It may not be a "standard" configuration, but the complaint is that AMD is going above the max rated power draw for the connectors. The 2x8 pin+slot setup doesnt do that.
Posted on Reply
#27
JalleR
Thats right... so there is no problem with it, it is just missing the PCI Express Catification logo :)

This is from wiki so no guarantees:
"Some cards are using two 8-pin connectors, but this has not been standardized yet, therefore such cards must not carry the official PCI Express logo. This configuration would allow 375 W total (1×75 W + 2×150 W) and will likely be standardized by PCI-SIG with the PCI Express 4.0 standard."
Posted on Reply
#28
KainXS
[QUOTE=]
Im Rahmen des derzeit stattfindenden IDF (Intel Developer Forum) erklärte Al Yanes, der Präsident der PCI-SIG, nun im Gespräch mit c't, dass die beiden augenscheinlich konträren Standpunkte sehr wohl zusammenpassen: Innerhalb der festgelegten Konformitätstests gibt es schlicht keinen Subtest, der die Einhaltung der vorgeschriebenen Stromaufnahme überprüft. Dieser Sachverhalt erklärt auch, warum manch andere Grafikkarten mitunter mehr Saft aus ihren zusätzlichen Stromsteckern ziehen als eigentlich zulässig.[/quote]This clears up some things though, the president of PCI-SIG said that they don't test slot power draw in their own testing even though the standard/requirements they put out says that its supposed to be enforced.


. . . . . . .I don't even know what to say about that.
Posted on Reply
#29
Caring1
KainXS, post: 3507513, member: 42957"
This clears up some things though, the president of PCI-SIG said that they don't test slot power draw in their own testing even though the standard/requirements they put out says that its supposed to be enforced.


. . . . . . .I don't even know what to say about that.
That makes sense to me, they draw up the standard and certify that components comply.
It should be up to manufacturers to ensure they comply to qualify.
It's no different than automobile manufacturers conducting their own crash tests for compliance in that standard.
Posted on Reply
#30
qubit
Overclocked quantum bit
This power problem was fixed with a driver update a little while ago so why is it still an issue?
Posted on Reply
#31
m1dg3t
And yet, nVidia is still selling 970s as '4Gb' RAM and 64?? ROPs...

Funny how that works, eh.
Posted on Reply
#32
JalleR
qubit, post: 3507527, member: 46003"
This power problem was fixed with a driver update a little while ago so why is it still an issue?
As i wrote earlier...

"And yes with the fix it is still not compliant it is 75 Watt from the slot and 75watt from 6 pin, not 84Watt from the 6 pin."


Regarding the 970:
It has 4GB memory the issue is that the last 0.5gb is not full speed (can't figure out if it is the total 4GB that is slowed down or only the 0.5gb??)
Posted on Reply
#33
HD64G
:roll::roll::roll::roll::roll::roll::roll::roll::roll::roll::roll::roll::roll::roll:

Too little too late!

They make a topic out of something non-existant atm and didn't notice once it was at least existant.

Ridiculous announcement by them (not TPU) to sum it up...
Posted on Reply
#34
qubit
Overclocked quantum bit
JalleR, post: 3507530, member: 50355"
"And yes with the fix it is still not compliant it is 75 Watt from the slot and 75watt from 6 pin, not 84Watt from the 6 pin."
I remember the fix being tested by various tech sites which made the card compliant.
Posted on Reply
#35
Everton Souza
JalleR, post: 3507530, member: 50355"
Regarding the 970:
It has 4GB memory the issue is that the last 0.5gb is not full speed (can't figure out if it is the total 4GB that is slowed down or only the 0.5gb??)
Gonna try to help on this one.

3,5 gb - 224 bit bus
0,5 gb - 32 bit bus

So compared to a 980 all memory access is slower. You only uses/gets the advertised 256 bit memory bus when accessing data from faster "partition" and slower one at the same time.
Posted on Reply
#37
JalleR
qubit, post: 3507535, member: 46003"
I remember the fix being tested by various tech sites which made the card compliant.
TPU test dosn't show that. there is still to high draw from the 6pin (84Watt).
https://www.techpowerup.com/img/16-07-08/rx-480-pcie-fix-power.jpg


OK
Again from Wiki so.....
For accessing its memory, the GTX 970 stripes data across 7 of its 8 32-bit physical memory lanes, at 196 GB/s. The last 1/8 of its memory (0.5 GiB on a 4 GiB card) is accessed on a non-interleaved solitary 32-bit connection at 28 GB/s, one seventh the speed of the rest of the memory space. Because this smaller memory pool uses the same connection as the 7th lane to the larger main pool, it contends with accesses to the larger block reducing the effective memory bandwidth not adding to it as an independent connection could
Posted on Reply
#38
KainXS
JalleR, post: 3507553, member: 50355"
TPU test dosn't show that. there is still to high draw from the 6pin (84Watt).
https://www.techpowerup.com/img/16-07-08/rx-480-pcie-fix-power.jpg


OK
Again from Wiki so.....
For accessing its memory, the GTX 970 stripes data across 7 of its 8 32-bit physical memory lanes, at 196 GB/s. The last 1/8 of its memory (0.5 GiB on a 4 GiB card) is accessed on a non-interleaved solitary 32-bit connection at 28 GB/s, one seventh the speed of the rest of the memory space. Because this smaller memory pool uses the same connection as the 7th lane to the larger main pool, it contends with accesses to the larger block reducing the effective memory bandwidth not adding to it as an independent connection could
its still ok the 6 pin is overly robust and the spec should be changed to accommodate that 84W is perfectly fine.

and what does this have to do with the 970, . . . . I miss something?
Posted on Reply
#39
ty_ger
$ReaPeR$, post: 3507398, member: 56172"
"The list is compiled for hardware devices implementing the various PCI-Express specifications to the letter."
right.. spikes in the power draw don't count I imagine..
You mean RMS? Yes, they are using RMS rating like all other power companies, utilities, and applications do.

http://practicalphysics.org/explaining-rms-voltage-and-current.html

RMS:
In statistics and its applications, the root mean square (abbreviated RMSor rms) is defined as the square root of mean square (the arithmetic mean of the squares of a set of numbers).
Just because some site created a video with an oscilloscope and then provided a convoluted method of filtering and talked about spikes doesn't make you all the sudden smarter than the industry.
Posted on Reply
#40
TRWOV
mmm well this means that Dell XPS and Alienware PCs will surely sport an RX 470 instead... maybe Sapphire will have to churn out some "reference" 470s for them.
Posted on Reply
#42
HD64G
Caring1, post: 3507525, member: 153156"
That makes sense to me, they draw up the standard and certify that components comply.
It should be up to manufacturers to ensure they comply to qualify.
It's no different than automobile manufacturers conducting their own crash tests for compliance in that standard.
Wrong. Both in EU and US there are seperate organisations for crash-testing vehicles with their own standards. Car manufacturers try to comply with those tests by simulating them nowadays on computers or in prevous decades by crash-testing pre-production cars to be sure they wouldn't fail when the official test was conducted. :toast:
Posted on Reply
#43
semitope
FYI the 1080 is not there. Neither are the 1070 and 1060. The 470 and 460 are also missing.
nvm they are listed by the chip name. Which is silly because the chip is not the entire gpu+board...

Wondering what this list really is. Because you would expect several versions of each chip to correlate with the board design, but for everything there is only one listed.

Lots of cards are missing.
Posted on Reply
#44
wiak
JalleR, post: 3507331, member: 50355"
Not at all, but still nice to see that there is consequences, and AGAIN it still just show how bad AMD design department is...... why didn't they just put a 8pin from the start.... Swallow that less than 150W pride suckers.... :)
lol didnt the geforce 750ti or 950 or something do the exact same thing
Posted on Reply
#45
JalleR
hmm I'm a RED kinda guy, so I don't know OLD Nvidia Tech, but I know for sure that my living on the RED is long gone, so my new GFX is going to be a 1080... I had hops for the RED team but let me just say i'm really disappointed..... they could have done much better just rebranding the 290 again and make it in 14nm FinFET.
But that is just my opinion.
Posted on Reply
#46
semitope
JalleR, post: 3507682, member: 50355"
hmm I'm a RED kinda guy, so I don't know OLD Nvidia Tech, but I know for sure that my living on the RED is long gone, so my new GFX is going to be a 1080... I had hops for the RED team but let me just say i'm really disappointed..... they could have done much better just rebranding the 290 again and make it in 14nm FinFET.
But that is just my opinion.
You can't rebrand a GPu and make it on a smaller process.

and if you are looking at a 1080 why on earth do you care about a 290 rebrand instead of a 480? At best you should be saying vega is taking forever.
Posted on Reply
#47
$ReaPeR$
ty_ger, post: 3507567, member: 54252"
You mean RMS? Yes, they are using RMS rating like all other power companies, utilities, and applications do.

http://practicalphysics.org/explaining-rms-voltage-and-current.html

RMS:


Just because some site created a video with an oscilloscope and then provided a convoluted method of filtering and talked about spikes doesn't make you all the sudden smarter than the industry.
couldn't find anything confirming what you said. furthermore, i find this whole story a bit shady, i mean, first they have the rating then they dont.. its a load of bs if you ask me.
Posted on Reply
#48
wiak
GTX 1060 isnt even on the list...
Posted on Reply
#49
qubit
Overclocked quantum bit
JalleR, post: 3507553, member: 50355"
TPU test dosn't show that. there is still to high draw from the 6pin (84Watt).
https://www.techpowerup.com/img/16-07-08/rx-480-pcie-fix-power.jpg
You're taking that graph out of context.

Here's the last paragraph of the conclusion which says that it's fixed (heck the title of the article says it's fixed) especially with compatibility mode which maxes it out at 73W. Without compatibility mode it averages 76W with peaks of 79W which is what the graph you linked to shows and isn't a problem for all but the crappiest of motherboards.
Here you can see that the new driver adds about 2.3% performance, which is a pretty decent improvement. Once you enable Compatibility Mode though, performance goes down slightly below the original result (0.8% lower), which means Compatibility Mode costs you around 3%, in case you really want to use it. I do not recommend using Compatibility Mode, personally I don't think anyone with a somewhat modern computer would have run into any issues due to the increased power draw in the first place, neither did AMD. It is good to see that AMD still chose to address the problem, and solved it fully, in a good way, and quick.
And here's the link if you wanna read the whole thing in context: https://www.techpowerup.com/223981/amd-releases-pci-express-power-draw-fix-we-tested-confirmed-works
Posted on Reply
#50
Pewzor
Was 3.5gb/4gb ok as well?
Posted on Reply
Add your own comment