AMD: Speculation, Rumors, and Discussion (Archive)

[lanek]

So your saying that this card is aimed for the hardcore OC that will use LN2 on her?

No LN2 ( and again, i have allready seen it )... But the cheap overclockers who run on team.. you know we even rebench continously low cheap gpu's,... competition mean entry, and got points within category. .. OC is tweaking and have fun doing WR, whatever is the category of the gpu's. There's no absolute score, only category scores. (HWbot )

I could bench only with High end gpu's, and got 1 entry for this card, and got only point for it, but i can too bench every category of GPU's and CPU, and got 10 entries who give point for me and my team for each category.

This said, i was not speaking "overclockers" competition, but gamers who overclock their gpu's.

Along this you have all the guys who dont have 700$ to spend on a gpu's, but who will love to have fun with a 200-300$ by OC it for bench or for gaming.

You know, forums are full of peoples who like to bench and show their scores, or just been able to put xgpu@xxxxxmhz in their profiles. Never ask in a gaming,hardware forum what is their perf on this game, as you will got hundred of response of guys who put you a +400mhz fps result.

 

[Love_In_Rio]

110 watts power load at max by the leakers card manual.Also states 150 watts allows overclocking boost clock to near 1,4 Ghz.The 110 watts figure is strangely similar to the TDP of the RX 470.

 

[Deleted member 13524]

The 110 watts figure is strangely similar to the TDP of the RX 470.

Not that strange considering it's essentially the same card.

 

[Alexko]

110 watts power load at max by the leakers card manual.Also states 150 watts allows overclocking boost clock to near 1,4 Ghz.The 110 watts figure is strangely similar to the TDP of the RX 470.

Wait, what, where? I feel like the thread's title keeps changing the power figure and there's rarely any solid reason for it. Did I miss something again?

 

[CarstenS]

That is a disappointingly - nay, may I say ludicrously - small heatsink. And what on earth is its basic design? An extruded alu base with a copper slug set into it, connected to what on the other side, solid alu fins? No, that couldn't work with a 150W heat load... Must at least be some pipes tacked onto there somewhere.

This was on Barts Pro (HD 6850) with a 127 watt Board Power - and it looks like it's quite a bit smaller:

I don't have any concerns for the cooler shown for RX480 being capable of dissipating 150 watts. Larger coolers will be more quiet of course.

 

[lanek]

(surely not the reference clock, but with a small increase on this XFX )

 

[CSI PC]

110 watts power load at max by the leakers card manual.Also states 150 watts allows overclocking boost clock to near 1,4 Ghz.The 110 watts figure is strangely similar to the TDP of the RX 470.

I assume though that is the 4GB model with the 8GB being more, and this is further compounded I expect when OC memory.

Just to add we also saw the 110W figure for the 470, which is a cut down 480. They should be notably different just look at 1070/1080 from a power consumption and also TDP perspective.
Not sure what figures to believe anymore.
Cheers

 

[Putas]

Depends how the capacity is realized, by using more chips or same amount of chips with higher density? More memory can lead to lower consumption if it helps to avoid bus transfers.

 

[CSI PC]

Depends how the capacity is realized, by using more chips or same amount of chips with higher density? More memory can lead to lower consumption if it helps to avoid bus transfers.

How does that work when the 4GB and 8GB 480 models are both meant to be using the top 8GHz memory modules?
The 8GB will use more power and probably around 20W-25W excluding OCing compounding this.
The 470 from a memory perspective should be marginally more efficient from a memory consumption perspective as it is using the 7GHz memory modules, let alone the fact it is a cut down 480.
Importantly all 3 models I think are meant to be 256-Bit bus.
CHeers

 

[Grall]

The 8GB will use more power and probably around 20W-25W excluding OCing compounding this.

The DRAM array of a GDDR device uses a negligible amount of power compared to the host interface, so 8GB RAM version will use only marginally more power. Perhaps unmeasurably more even.

 

[Gipsel]

How does that work when the 4GB and 8GB 480 models are both meant to be using the top 8GHz memory modules?
The 8GB will use more power and probably around 20W-25W excluding OCing compounding this.

8 memory chips with 4 Gbits each will consume not that much less than 8 memory chips with 8 Gbits each. The difference will probably be 1-2W at max (so maybe 0.2W per chip). One basically has to refresh more cells every half an eternity (relative to the clock speed). That can't burn that much energy relative to the actual memory transfers, which cost the same energy irrespective of the capacity.

 

[Razor1]

http://videocardz.com/61225/amd-radeon-rx-480-rumors-part-2
3dmark scores with pictures

 

[CSI PC]

The DRAM array of a GDDR device uses a negligible amount of power compared to the host interface, so 8GB RAM version will use only marginally more power. Perhaps unmeasurably more even.

Sorry if this seems pedantic. but by that logic you might as well say there is 0W consumption for the GDDR5 VRAM modules.
If there is zero difference between 4GB and 8GB, there is also negligible difference between none and 4GB (ok some but not much - yeah being too pedantic I agree as deliberately ignoring the interface ).
Basis being they use the same bus width.
I thought I read some time ago AMD or another manufacturer confirmed there is something like 15W with each 4GB on GDDR5 *Shrug*, although would help if I could remember the context as well.
Maybe it was lower *shrug*.

Edit:
I guess I am conflating this with some other facts.
Cheers

 

[CSI PC]

8 memory chips with 4 Gbits each will consume not that much less than 8 memory chips with 8 Gbits each. The difference will probably be 1-2W at max (so maybe 0.2W per chip). One basically has to refresh more cells every half an eternity (relative to the clock speed). That can't burn that much energy relative to the actual memory transfers, which cost the same energy irrespective of the capacity.

Yeah logically that makes sense,
maybe I am conflating this with GDDR5 and HBM comparisons.
CHeers

 

[Deleted member 13524]

http://videocardz.com/61225/amd-radeon-rx-480-rumors-part-2
3dmark scores with pictures

60º operation seems confirmed, though the chip itself should be a bit hotter.

 

[A1xLLcqAgt0qc2RyMz0y]

Wait, what, where? I feel like the thread's title keeps changing the power figure and there's rarely any solid reason for it. Did I miss something again?

Yes the Title has changed at least three time.

First it was 150 watts, then it was 100 watt range and now it is 110 watts.

Who gets to decide title changes and why is their no reason given for the changes?

 

[CSI PC]

60º operation seems confirmed, though the chip itself should be a bit hotter.

Probably right when using a thermal camera/sensor.
Example here of the 1070FE when doing the same (calibrated camera): http://www.guru3d.com/articles_pages/nvidia_geforce_gtx_1070_review,10.html
Cheers

 

[Gipsel]

Sorry if this seems pedantic. but by that logic you might as well say there is 0W consumption for the GDDR5 VRAM modules.
If there is zero difference between 4GB and 8GB, there is also negligible difference between none and 4GB (ok some but not much - yeah being too pedantic I agree as deliberately ignoring the interface ).
Basis being they use the same bus width.
I thought I read some time ago AMD or another manufacturer confirmed there is something like 15W with each 4GB on GDDR5 *Shrug*, although would help if I could remember the context as well.
Maybe it was lower *shrug*.

If the number of memory chips as well as the bus width stay constant, not much changes irrespective of the capacity (the changes are tiny). If you add more chips to increase capacity, yes in that case the power consumption rise. But even then, if you keep the bus width constant, it is actually much less then linear. If you put two GDDR5 chips on a single 32bit Controller port (i.e. run them in clamshell mode where each chip provides 16 bits for each transfer), they won't consume twice as much as a single chip operating in 32bit mode. Last time I checked the power consumption in clamshell mode was something like 60-70% of the maximum when operating in 32bit mode.

 

[xEx]

Yes the Title has changed at least three time.

First it was 150 watts, then it was 100 watt range and now it is 110 watts.

Who gets to decide title changes and why is their no reason given for the changes?

The mod who created it using my name...

 

[Deleted member 13524]

As far as I can see, this is mostly a rumor/leak thread and it will continue to be one for another 9 days.
The latest leak comes from someone who took pictures of the card and claimed its power consumption was 110W.

I know this seems to be coming to bite in the ass of the users who were claiming Polaris was a lot less power efficient than Pascal and that Polaris was already a failure, but fret not: if the card turns out to be consuming 130, 140, or 150 or more watts, the thread title will surely be updated.