AMD: Speculation, Rumors, and Discussion (Archive)

[pharma]

July 4,2016

Now we come to the most important - the load distribution. AMD uses - that affect the allocation of the individual loads as logical conclusion - four of the six phases of the PEG, two phases for the PCIe connector and a separate extra-phase for memory and Others (also on PCIe). Add to that the extra phase for the 3.3 volt statements on PEG, the AMD but not obviously counts.

Even if one were to reduce the power consumption of the memory and the rest of the electronics boards drastically to below realistic values, no real balance of three phases is also possible!

It is actually not possible with two or three phases for the PEG to achieve the 75-80 watts, which could be read as a measurement result in the last few days. With two stages to be at 40 to 41 watts, with three to 60 to 62 watts.

Sure, the IR 3567B of International Rectifier could theoretically also utilize the phase asymmetrical, but what sense should this arise (including the then emerging uncertainties and the overload on the 6-pin connector)?

But to draw attention back to the beginning and to AMD's announcement: We are very curious how a correction of abuses without BIOS updates of firmware and solely by means of a new driver can be realized, as AMD has announced.

http://www.tomshardware.de/amd-rade...ungsaufnahme-pci-sig,testberichte-242143.html

Edit: Would welcome a better translation then provided Google translate ...

 

[xEx]

Still no stock of 480 in the USA since AMD announced.

 

[CSI PC]

You'd be surprised what happens to a loaded-down ROP partition. Filling the ALUs won't always get you max power consumption.

Sure, which is why I think it needs to be analysed to know for sure how it behaves in those situations as currently one can only speculate, although as I mention it also needs to be not just AoTS async compute but other game designs with post processing\engine effects that work very well on GCN such as Quantum Break.

Cheers

 

[Gipsel]

July 4,2016


http://www.tomshardware.de/amd-rade...ungsaufnahme-pci-sig,testberichte-242143.html

Edit: Would welcome a better translation then provided Google translate ...

He is basically writing roughly the same as I did here in the thread. Obviously 4 phases of the GPU VRM source their supply from the slot, the other 2 phases (and the memory VRM and some other small stuff like the fan) are supplied by the 6pin plug. The exact numbering of the phases is still unclear (so don't take the image in the article too serious), it could also be numbered in reverse order. It really depends to which pins of the VRM Controller the MOSFET drivers are connected (one would need to measure this, what he obviously didn't do). This numbering would decide if one could indeed switch off one or two phases connected to the slot at normal load (without OC) or if the only possibility would be the load biasing between the phases (if one wants to operate the VRM with a lower amount of phases, one has to start the deactivation from the last phase, one can't deactivate lower numbered ones, with the higher ones active; the order is fixed).

 

[sir doris]

Still no stock of 480 in the USA since AMD announced.

Still plenty of stock in the UK.

 

[xEx]

Still plenty of stock in the UK.

could be that te demand in the UK is really low? in the US there have been no stock since july 2.

 

[Gubbi]

could be that te demand in the UK is really low?

Or they shipped everything to the UK to maximize sales there before the sterling collapses

Cheers

 

[CarstenS]

Plenty of stock here in germany as well. 4-GiB-variants go for as low as 220 EUR (that's incl. VAT/sales tax whatever as required per law here).
That said, GTX 1070/1080 is also readily available from multiple renowned shops, despite apparenty shortage in other parts of the world. Prices are high, still.

 

[CSI PC]

He is basically writing roughly the same as I did here in the thread. Obviously 4 phases of the GPU VRM source their supply from the slot, the other 2 phases (and the memory VRM and some other small stuff like the fan) are supplied by the 6pin plug. The exact numbering of the phases is still unclear (so don't take the image in the article too serious), it could also be numbered in reverse order. It really depends to which pins of the VRM Controller the MOSFET drivers are connected (one would need to measure this, what he obviously didn't do). This numbering would decide if one could indeed switch off one or two phases connected to the slot at normal load (without OC) or if the only possibility would be the load biasing between the phases (if one wants to operate the VRM with a lower amount of phases, one has to start the deactivation from the last phase, one can't deactivate lower numbered ones, with the higher ones active; the order is fixed).

Yeah like you still not assuming any information beyond the Tom's Hardware/PCPer tests, nor assuming what options (key also being ease of implmentation) are available for AMD to resolve this characteristic.
I would think the most ideal situation in this (specifically 480 design issue as not replicated with earlier models) for AMD would be if 4 true phases to the GPU are on the auxiliary 6-pin and linked in ideal way for duty cycle and interleaved phase control, with the other 2 GPU phases on PCIe slot so these would be in a permanent low-side state, but nothing can be taken for granted on how they built it as it is strange.

Cheers

 

[CSI PC]

On a separate issue but related to power.
What was the reason for the idle power characteristic also being way too high?
Not an issue-consideration like the maximum current draw/distribution, but curious what is the cause.

Thanks

 

[Alexko]

Is there any indication about the RX 470's release date?

 

[Gipsel]

On a separate issue but related to power.
What was the reason for the idle power characteristic also being way too high?
Not an issue-consideration like the maximum current draw/distribution, but curious what is the cause.

According to the launch reviews, AMD notified the testers that there is a driver issue which prohibits the RX480 to enter the lowest power state in idle. They announced a fix to that (allegedly reducing the idle power by ~6W, i.e. from 16W to 10W or something in that range).

 

[xEx]

Mid july(about 14th) but we will see

 

[ninelven]

950 is the current gen, AFAIK there's no sucessor for it yet, so unless AMD has a psychic this accusation is unwarranted.

It is not an accusation. It is a factual observation. People incapable (or unwilling) of understanding that are also part of the problem.

 

[CarstenS]

Plenty of stock here in germany as well. 4-GiB-variants go for as low as 220 EUR (that's incl. VAT/sales tax whatever as required per law here).
That said, GTX 1070/1080 is also readily available from multiple renowned shops, despite apparenty shortage in other parts of the world. Prices are high, still.

Interestingly, mere hours later, all 4 GiB variants have vanished from the price search engines or are only availble for prices as high as the 8 GiB variants. Must be in high demand.

 

[xEx]

According to the launch reviews, AMD notified the testers that there is a driver issue which prohibits the RX480 to enter the lowest power state in idle. They announced a fix to that.

Why AMD lunches a card with so many driver issues? It could be better to just delay the lunch until they could get a proper driver.

 

[Gipsel]

Interestingly, mere hours later, all 4 GiB variants have vanished from the price search engines or are only availble for prices as high as the 8 GiB variants. Must be in high demand.

Probably because some of the first buyers of the 4GB started to report, that there are indeed 8GB of Samsungs 8Gbps GDDR5 on them (only a BIOS flash needed to activate the full 8GB?).

 

[xEx]

Probably because some of the first buyers of the 4GB started to report, that there are indeed 8GB of Samsungs 8Gbps GDDR5 on them (only a BIOS flash needed to activate the full 8GB?).

I doubt after the phenom phenomenon AMD would let something like that happen again.

 

[Gipsel]

I doubt after the phenom phenomenon AMD would let something like that happen again.

Just look it up in some fora. They reports of that are trickling in. I'm not kidding you.

Edit:
http://www.overclock.net/t/1604567/polaris-bios-mod-rx480/10#post_25314500
or in German: http://www.forum-3dcenter.org/vbulletin/showthread.php?p=11090628#post11090628

 

[CarstenS]

Probably because some of the first buyers of the 4GB started to report, that there are indeed 8GB of Samsungs 8Gbps GDDR5 on them (only a BIOS flash needed to activate the full 8GB?).

Quite possible.