AMD: Speculation, Rumors, and Discussion (Archive)

[eastmen]

Yeah makes sense but comes down to interpretation.
Did AMD clarify like Nvidia what can still be done under warranty?
For Nvidia increasing power target does not affect warranty, I assumed maybe AMD was the same.
Thanks

https://community.amd.com/community...your-graphics-performance-with-radeon-wattman

Note: AMD’s product warranty does not cover damages caused by overclocking, even when overclocking is enabled via AMD hardware and/or software

Once you overclock anything your warranty is void basicly. That is why if its staying in spec when not overclocked then its up to the consumer to decide if running the card out of spec and all the issues that might come with it is okay with them.

 

[Mendel]

This whole pci express power draw thing with rx480s could deserve its own thread.
The revision C7 says a lot about process technology hurdles.
I guess since they had that many respins already, they then finally just went ahead and released the gpu even when the board design with single 6 pin wasn´t enough for it?

 

[CarstenS]

Yeah makes sense but comes down to interpretation.
Did AMD clarify like Nvidia what can still be done under warranty?
For Nvidia increasing power target does not affect warranty, I assumed maybe AMD was the same.
Thanks

AFAIK activating anything in Overdrive-slash-Wattman is outside of warranty and rightly so.

 

[CarstenS]

This whole pci express power draw thing with rx480s could deserve its own thread.
The revision C7 says a lot about process technology hurdles.
I guess since they had that many respins already, they then finally just went ahead and released the gpu even when the board design with single 6 pin wasn´t enough for it?

C7 might be due to repurposed register read by GPU-z.

 

[renderstate]

It's unfortunate, but AMD at this moment seems like a pretty shoddy company where image is more important than true technical achievement ....

Let's be realistic, it's not like the 480 is shining in the "true technical achievement" department.

 

[silent_guy]

C7 might be due to repurposed register read by GPU-z.

Yeah, that makes no sense at all.

 

[xEx]

Well the 480 is the first gpu made by glofo and the first 14nm design by amd with a process designed by another company for arm low power chips. So yeah it might be broken. But we can't know with just the avaliable information.

Enviado desde mi HTC One mediante Tapatalk

 

[CSI PC]

AFAIK activating anything in Overdrive-slash-Wattman is outside of warranty and rightly so.

Well that makes it tricky to personally recommend to my friends because now if they want to undervolt/lower power target a little this will also invalidate warranty.
Not being pedantic, but that was an easy solution around the current power characteristics.
Ah well, fingers crossed custom AIB models sort the power distribution out.

Cheers

 

[eastmen]

Well that makes it tricky to personally recommend to my friends because now if they want to undervolt/lower power target a little this will also invalidate warranty.
Not being pedantic, but that was an easy solution around the current power characteristics.
Cheers

It was like that also with over drive and overclocking for quite awhile.

 

[Kaotik]

While everything you do in WattMan / Overdrive / any such similar program (MSI Afterburner, EVGA precision etc) is technically out of warranty, the companies can't actually prove you've ever used them AFAIK

 

[eastmen]

While everything you do in WattMan / Overdrive / any such similar program (MSI Afterburner, EVGA precision etc) is technically out of warranty, the companies can't actually prove you've ever used them AFAIK

they can't have some bits that are turned from 0s to 1s if it gets written too from wattman or over drive or something ? I would think they could just code that in so when you check okay to overclocking it shows up in the bios

 

[hughJ]

It wouldn't matter if they did, OCing with those applications is covered by warranty (at least for MSI and EVGA from some quick googling.) *Not to suggest that AMD or their partners will take that same stance with the product, but obviously there's enough confidence in the Nvidia-based boards for overclocking to not be a costly risk of killing cards or mainboards.

 

[Grall]

they can't have some bits that are turned from 0s to 1s if it gets written too from wattman or over drive or something ?

How would you read the bits if your card is dead? You could unsolder the BIOS and dump it, but who's going to waste a tech's time doing that? It'd cost more to forensically analyse every RMA'd board than to just replace them...

 

[Deleted member 13524]

Does that mean the current GF's 14nm GF implementation is broken?

Depending on the POV, almost all chips can be called "broken".
AFAIK pretty much all ASICs past a certain transistor amount have bugs, from the newest 14FF chip ever made to the latest revision of another chip made on a super mature process like 40nm.
Revisions usually squash some of those bugs and bring some performance, power efficiency, stability or other kind of improvements in the process. For example, the Snapdragon 810 rev 2.1 brought some pretty substantial improvements in power efficiency and heat output, which in turn resulted in a much improved user experience for smartphones using it compared to the ones with the older version.



That said, what I mostly read from that new information is that since the RX 485 is a revision and admittedly not a higher performance tier, we can probably guess that there won't be a higher performance part based on Polaris 10 anytime soon (save for overclocking-driven OEM solutions with 3rd party coolers, of course).

 

[silent_guy]

Depending on the POV, almost all chips can be called "broken".

He means broken as in "unreasonable amount of power consumption or low speed compared to what could have been expected."

 

[Deleted member 2197]

RX 480 Product Award withdrawn by reviewer .for violation of the PCIe specification ...

The reference model of AMD Radeon RX 480 was allocated at the time of the award review "Good product". Due since arisen ambiguities the editors have chosen to suspend the review award. Other text is indefinite unchanged. In short, there are question marks about whether the card exceeds the PCI Express specification for power consumption, where several independent tests indicate substantial violations - even though the model according to AMD successfully passed all certifications. The editorial staff is extremely restrictive to retrospectively change the rating, but when it became evident the problem can affect or damage the other components we decided to withdraw the award. We await further updates from AMD, and at that point will determine whether or not to reinstate the award's status.

http://www.sweclockers.com/test/22310-amd-radeon-rx-480-8-gb

 

[el etro]

Does that mean the current GF's 14nm GF implementation is broken?

Will be great if they can put efficiency on par(or better) with what Pascal offer. Man, Polaris 11 demonstration was mind-blowing.


If they ended to problems beyond 100W TDP situation will be awkward. Don't know much info about it, but i read Sun Microsystems made the Sparc M7 on the 20nm node. Even AMD skipped that node back ago, but Sun fabbed a even bigger xPU on that 20nm node!



The point of my post is that i believe AMD will do a turnaround from this - if they end with the bigger TDP chips having bad perf/watt - again.

 

[Razor1]

Does that mean the current GF's 14nm GF implementation is broken?

Most likely this isn't a process related issue. Process might compound the issue.

 

[Razor1]

Will be great if they can put efficiency on par(or better) with what Pascal offer. Man, Polaris 11 demonstration was mind-blowing.


If they ended to problems beyond 100W TDP situation will be awkward. Don't know much info about it, but i read Sun Microsystems made the Sparc M7 on the 20nm node. Even AMD skipped that node back ago, but Sun fabbed a even bigger xPU on that 20nm node!



The point of my post is that i believe AMD will do a turnaround from this - if they end with the bigger TDP chips having bad perf/watt - again.

After seeing the the rx480 you think that will happen? I would be very skeptical.

 

[SimBy]

We'll see what the perf/W is on RX470.