AMD: Speculation, Rumors, and Discussion (Archive)

[tunafish]

From what amd has shown, it's likely that Polaris will beat Pascal in perf/watt. Amd was getting 2x perf/watt compared to maxwell in the Polaris 11 demo while nvidia seems to be only getting 1.5x right now. I don't know where you are pulling 25% less from.

Furthermore, the framerate cap was there for both graphics cards, and nvidia chips also save power with framecap (e.g. boost not needed).

nVidia doesn't save nearly as much power with frame cap as AMD does. AMD cards can effectively clock gate very quickly, meaning that dynamic power goes to 0 for the part of the frame that the card is not utilized. nVidia cards cannot do this nearly as well, so most of their power savings at partial utilization come from lower clocks, clocking up and down consumes time and power, and when using a frame cap the cards clock up for the active portion of the frame.

I feel like I've said this two dozen times now -- you cannot trust vendor benchmarks, not because they straight up lie, but because they can run two dozen benchmarks and pick the most compelling one to show us. We will know what the power efficiency of Polaris is when cards are in the hands of independent reviewers, not before.

 

[lanek]

Dont they have show the same thing without frame limit sometimes after just for erase some doubt ?

 

[CarstenS]

Nope.

 

[CarstenS]

nVidia doesn't save nearly as much power with frame cap as AMD does. AMD cards can effectively clock gate very quickly, meaning that dynamic power goes to 0 for the part of the frame that the card is not utilized. nVidia cards cannot do this nearly as well, so most of their power savings at partial utilization come from lower clocks, clocking up and down consumes time and power, and when using a frame cap the cards clock up for the active portion of the frame.

This is actually very true. I measured this roughly in terms of overall PC power consumption and AMD was able to lower their power figures much more efficiently with the current Radeon cards - caveat: Coming from a higher power level to begin with.

 

[silent_guy]

nVidia doesn't save nearly as much power with frame cap as AMD does. AMD cards can effectively clock gate very quickly, meaning that dynamic power goes to 0 for the part of the frame that the card is not utilized. nVidia cards cannot do this nearly as well, so most of their power savings at partial utilization come from lower clocks, clocking up and down consumes time and power, and when using a frame cap the cards clock up for the active portion of the frame.

I wonder how you came to this conclusion?

Clock gating is something that happens at the cycle level. One cycle the clock is on, the other it's off. There is no way for us to know who does one better than the other.

 

[Deleted member 13524]

Yes they did but both Bonaire and Hawaii are worse in perf/watt than the latest Tonga, so the end results, 2.5 perf/ watt on either of those chips vs 2.0 perf/watt on Tonga, is all the same!

Dude you really need to work on your math.
From the perf/watt results you linked yourself, Tonga is ~11% more efficient than Hawaii. That does not turn a 2.5x advantage into 2.0x.

 

[Razor1]

the 390x was not in that review was it?

We are talking about best case here, AMD's best case for the 2.5 is to take the actual power draw of their chip which is higher than the rated 275 watts (around 300 watts) and then go from there. I didn't even factor that in...... There is only one way to go from AMD numbers and that is down, not up, unless they are blowing smoke up our collective asses and OEM's asses lol, just to trick nV.

https://www.techpowerup.com/reviews/MSI/R9_390X_Gaming/31.html

here is a 390x

 

[Deleted member 13524]

290 and 290X = Hawaii.



BTW, we can now say with some certainty that the Macau event will in fact be happening:

https://twitter.com/IanCutress/status/735753643652055042

People are taking their flights today, so the "secret" press event will be either later today or tomorrow.


I guess that after all the FUD being spread about Polaris not appearing at Computex, it's a bit ironic that it's appearing before the show.

BTW, new logo?

 

[Razor1]

well lets take 290x as the base line, you will get what with Polaris at 2.5 times perf/watt over 290x?

100 watts at 290x performance level? Is that what it is? yes, at 150 watts you get Fury X level performance then if its perfect scaling frequency to wattage? Yes, where does that leave it when talking about Pascal? Less than Pascal.

I don't really care how its cut, sliced, mashed what ever, you just don't get the best case number provided by AMD to match up with what we know Pascal is doing right now and these are independent tests not best case.

Now you take Tonga, 2.0 times perf/watt. 95 watts for Tonga (380x) performance. Guess what, its damn close to what that 2.5 times pef/watt is to Hawaii.

 

[Silent_Buddha]

Also boost doesn't increase Maxwell 2's power usage beyond their envelope. So using frame rate lock, doesn't do jack to save power on Maxwell 2 cards, very little savings.

I'd say that's a problem with Maxwell 2 then, if it can't save power when not being stressed as much.

Regards,
SB

 

[Silent_Buddha]

Do you really think that there majority of gamers set a frame cap? I find that very hard to believe.

I do. My monitor only goes to 60 hz. It's a waste if it isn't capped at 60. I had assumed that similar to AMD my Nvidia card was saving some power when it was being capped at 60, but if it isn't, then bleh.

Regards,
SB

 

[Razor1]

No cards prior to the r3xx and Fiji cards get any kind of power savings *anything meaningful* when capped. Its a good technology yes, but when you want to compare performance/watt not really...... Because there could be performance that we can't see when capped and power consumption can stay relatively the same on the cards that don't have this feature.

 

[Silent_Buddha]

No cards prior to the r3xx and Fiji cards get any kind of power savings *anything meaningful* when capped. Its a good technology yes, but when you want to compare performance/watt not really...... Because there could be performance that we can't see when capped and power consumption can stay relatively the same on the cards that don't have this feature.

R3xx = R2xx, so it appeared during R2xx which was quite a while before Maxwell 2. I had assumed that at least by Maxwell 2, Nvidia would have caught up to AMD there. Does Pascal at least do better in this regard?

Regards,
SB

 

[Razor1]

hmm there are differences between r3xx and the r2xx GPU's,

http://www.amd.com/en-us/innovations/software-technologies/technologies-gaming/frtc

I might have misspoke, maybe the higher end r2xx does have this feature

No idea about Pascal, it wasn't talked about in any reviews or presentations, so I am assuming they don't have it or its not that good.

Ah yes there was a driver update that gave the gcn 1.1 cards this feature!

 

[CarstenS]

FRTC is just a software feature - which does not make it any less nice.

 

[Deleted member 13524]

yes, at 150 watts you get Fury X level performance then if its perfect scaling frequency to wattage?

There is no such thing.

 

[Razor1]

I know there isn't

 

[Esrever]

My take was 2x perf/watt compared to Fiji and 2.5 compared to others. Fiji is within the range of efficiency of maxwell. To me this makes the most sense. Tonga isn't really more efficient than other gcn cards. The 390x just has poor efficiency due to the clocks.

Spoiler

2.5x 270x would make sense because they are the same class of chips.

 

[pharma]

Is perf/watt influenced by the type of cooling solution?

 

[3dilettante]

In total board terms, whatever power goes through an onboard fan header would count against it. I suppose off-board solutions should have that power dinged against them, but I don't think many comparisons do.
Depending on the temperature, the static component of power consumption can worsen--although FinFETs do a lot for static consumption and might reduce its contribution.

If RAM gets too hot, the refresh rate does climb, which would have a negative impact on performance, although cooling solutions generally haven't neglected memory to that point. I'm not sure if the power consumed by more refreshes outweighs everything else that stalls because of their waiting on refreshes.