AMD: Speculation, Rumors, and Discussion (Archive)

[silent_guy]

Esp if they can use cheaper ram and cheaper cooling (if the chip uses less power / generates less heat) the savings could translate to pcb layers , power hardware and so on

Didn't those videocardz numbers list GDDR5 at 8Gbps? That'd be the same RAMs as the 1070. With, historically, AMD needing more BW to achieve the same performance, they'll need all the BW they can get. And we (well, I) don't expect any significant power efficiency over Pascal.

So almost equal performance would require very similar power.

 

[eastmen]

we will have to see. The fury isn't much difference than the chips in the 290. Nvidia made massive gains in performance per watt when they moved from keplar to Maxwell .

I don't see why AMD also can't improve.

 

[Razor1]

Didn't those videocardz numbers list GDDR5 at 8Gbps? That'd be the same RAMs as the 1070. With, historically, AMD needing more BW to achieve the same performance, they'll need all the BW they can get. And we (well, I) don't expect any significant power efficiency over Pascal.

So almost equal performance would require very similar power.

I don't expect the same perf/watt as Pascal, I think it will be ~25% less in that category. Which will be a hell of a lot better then the 290x/390 vs their competitors.......

 

[Esrever]

I don't expect the same perf/watt as Pascal, I think it will be ~25% less in that category. Which will be a hell of a lot better then the 290x/390 vs their competitors.......

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.

 

[Razor1]

At same performance levels, hell no, they haven't shown that and they have been avidly avoiding to show that too by using frame rate locks, not to mention they are going out of their way to say they are targeting midrange for now.

AMD has been stating they are getting 2x the perf/watt of THEIR current midrange which is Tonga. Not nV's cards. Maxwell, gm 204, or the gtx 980, is considerably higher in perf/watt of the 380x.\

https://www.techpowerup.com/reviews/ASUS/R9_380X_Strix/24.html

At same performance levels for Titan X and 1070, you get up to what AMD has been talking about. x2 perf/watt of their current midrange. If nV can do this with their performance chips, what is stopping them from doing it for their mid range, and then their low end?

nV with the gtx 980 has a 1.6 perf/watt advantage over then 380x, so added on top of that Pascal gives another ~1.6-1.7x more compared to the gtx 980, AMD needs to make up much more ground to even out the scales in perf/watt

 

[silent_guy]

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.

I'm not convinced. Perf/W is usually optimal when your hardware is operating at full load. IMO, there's a good reason why AMD made the weird choice to use a frame cap of 60fps for that perf/W demo: comparing a Polaris 11 at full load against a GTX 950 that was not, is likely to be skewed in favor of the Polaris chip. A better comparison is to simply look at Fury X at full load against a GTX 1080 at full load. The difference there is ~2.3x.

 

[Deleted member 13524]

AMD has been stating they are getting 2x the perf/watt of THEIR current midrange which is Tonga.

The claim back in January was over 2x the per/watt of Hawaii and Bonaire.

http://www.amd.com/en-us/press-releases/Pages/amd-demonstrates-2016jan04.aspx

The world’s first public showcase of 14nm FinFET PC GPU technology promises more than double the performance per watt over previous generations1; now sampling to OEMs
(...)
Testing conducted by AMD internal labs as of Dec 15, 2015 with AMD’s previous “Hawaii” and “Bonaire” architecture based platforms and preliminary “Polaris” architecture based engineering sample. Systems tested with Intel i7-4770K with 8GB DDR3-1600 RAM, Driver 15.30 beta, Windows 10 64bit. POL-2

As it says above, this was based on a preliminary engineering sample, based on tests made over half a year ago.
Since March, the claims have been 2.5x, not 2x:

 

[SimBy]

And yet framerate cap is probably much more realistic real world scenario when it comes to gaming. Probably something reviewers should also consider. A lot of people limit the framerate. Either because they play with vsync on or to stay in adaptive sync range or to simply save a lot of power (less heat -> less noise).

 

[Deleted member 13524]

And yet framerate cap is probably much more realistic real world scenario when it comes to gaming. Probably something reviewers should also consider. A lot of people limit the framerate. Either because they play with vsync on or to stay in adaptive sync range or to simply save a lot of power (less heat -> less noise).

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

 

[pjbliverpool]

Any chance any of the polaris chips exceed fury x?

I'm just having trouble seeing even this new architecture not be competitive with the 1070. What could they've been thinking to even drop it on the midrange?

On the other hand if it was even a bit above 10~% faster than fury x it'd be exceeding the 1080 in some dx12 apps.

This puts it slightly slower than the Fury (non-X) and so a decent way behind the 1070. AMD have already stated they aren't competing with NV at that level with Polaris so I'd say it's pretty much case closed on that possibility. Until Vega, AMD will effectively have the 4 fastest performing single GPU's on the market.

http://videocardz.com/60253/amd-radeon-r9-480-3dmark11-benchmarks

 

[CSI PC]

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

But would the power be even less if they had also downclocked the 950, not sure how flexible Boost2 is in that regard with framecap.
Cheers

 

[CSI PC]

And yet framerate cap is probably much more realistic real world scenario when it comes to gaming. Probably something reviewers should also consider. A lot of people limit the framerate. Either because they play with vsync on or to stay in adaptive sync range or to simply save a lot of power (less heat -> less noise).

I would hate any reviews where they artificially limit the framerate, who would arbitrarily decide what the framerate should be and how it applies to product tiers/resolution/settings/etc.
It would hide too much regarding behaviour of not just fps but frame-time/latency when cards are pushed IMO.
Also it hides TDP and cooling systems performance.
Cheers

 

[Razor1]

The claim back in January was over 2x the per/watt of Hawaii and Bonaire.

http://www.amd.com/en-us/press-releases/Pages/amd-demonstrates-2016jan04.aspx



As it says above, this was based on a preliminary engineering sample, based on tests made over half a year ago.
Since March, the claims have been 2.5x, not 2x:

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! (You have to keep in mind nV and AMD calculate their power values differently. While r390x is stated to use 275 watts, it goes beyond that in typical gaming many times and ends up past 300 watts. So what numbers are being used?)

Using the 275 watts for r390x you get the same as the 2.0 perf/watt as Tonga.

Coincidence, no, they just used two different starting points to confuse everyone and fool people at times too.

Now I would take what they stated at the last financial conference call as more credible then anything else they have stated, its also the most recent as well.Which both the CEO and CFO stated 2.0 perf per watt over the current midrange. Does that mean the 280x or the 380x? If its the 280x AMD is way way behind, if its 380x, they are around ~20-25% behind Pascal.

Added to this, the test board, preliminary samples stuff yeah that doesn't fly, unless they had some serious issues in the silicon, power consumption, frequency and things like that aren't going to change much by minor changes. How many times has things like this been talked about here at B3D? Many times, you might get a minor % change, but nothing significant.

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

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.

 

[Orion]

This puts it slightly slower than the Fury (non-X) and so a decent way behind the 1070. AMD have already stated they aren't competing with NV at that level with Polaris so I'd say it's pretty much case closed on that possibility. Until Vega, AMD will effectively have the 4 fastest performing single GPU's on the market.

http://videocardz.com/60253/amd-radeon-r9-480-3dmark11-benchmarks

It's one thing to not compete at the high end another to not compete at mid range. I've not heard of amd abandoning mid and going full low end, that would be madness.

 

[Razor1]

But would the power be even less if they had also downclocked the 950, not sure how flexible Boost2 is in that regard with framecap.
Cheers

There is no flexibility, it doesn't down clock due to framelock.

 

[pjbliverpool]

It's one thing to not compete at the high end another to not compete at mid range. I've not heard of amd abandoning mid and going full low end, that would be madness.

But the 1070 and 1080 aren't mid range. They are high end. Even when the Ti releases they'll still be pretty high end, but by that time AMD will have something to compete in that space in the form of Vega. I expect Vega in different forms to be positioned against both GP104 and GP102.

 

[CarstenS]

Have had that discusion in our own forums. Depends on what you view as a card's dominant characteristic.

From a price and performance perspective, they are high end.
From an energy consumption perspective, they are mid-range.
From a chip-size and nomenclature perspective, they are upper midrange/performance class

 

[silent_guy]

And yet framerate cap is probably much more realistic real world scenario when it comes to gaming.

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

 

[silent_guy]

As it says above, this was based on a preliminary engineering sample, based on tests made over half a year ago.
Since March, the claims have been 2.5x, not 2x

During her last financial conference call end of April, Lisa Su said 2x ...

 

[silent_guy]

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

Yes. But if one is running at 90% and the other at 50%, you could still have a significant difference. Pure hypothetical numbers, BTW.