Polaris: 2x Perf/Watt of Current AMD mainstream GPUs

[Arachnotronic]

What do you think AMD means by "mainstream" GPUs in this case? And didn't AMD say previously Polaris was 2.5x perf/watt?

 

[Glo.]

Mainstream is always 199-249$ bracket.

So R9 380 and 380X.

We are looking at 50 GFLOPs/Watt for Polaris.

 

[Sweepr]

And didn't AMD say previously Polaris was 2.5x perf/watt?

They did.

I'm afraid 2x Tonga might not be enough.

 

[Glo.]

I'm afraid 2x Tonga might not be enough.

Are you sure? 3584 GCN chip with 1266 MHz of core clock and 180W TDP has 9 TFLOPs of compute power.

And 50 GFLOPs/watt.

Impossible on new node?

Secondly, I always thought that Vega 10 is high-end part from AMD, and Polaris, mainstream chip somehow has to compete with high-end part from Nvidia?

It would be hilarious if they would be able to compete with it, however, don't you think?

 

[el etro]

What do you think AMD means by "mainstream" GPUs in this case? And didn't AMD say previously Polaris was 2.5x perf/watt?

Or 2x versus Fiji or 2.5x versus Tonga, otherwise it will not be able to conpete with Pascal on perf/w.

 

[Arachnotronic]

Or 2x versus Fiji or 2.5x versus Tonga, otherwise it will not be able to conpete with Pascal on perf/w.

Says 2x of current "mainstream AMD GPUs." Fiji is not a mainstream GPU.

 

[Glo.]

Or 2x versus Fiji or 2.5x versus Tonga, otherwise it will not be able to conpete with Pascal on perf/w.

If Polaris is GP104 competitor, then which GPU competitor is Vega 10? GP102?

So now we get into situation, where Mainstream has to compete with High end.
High-end with enthusiast.

Pretty bonkers if you would ask me. Hilarious if AMD would be able to do this, however...

 

[Flapdrol1337]

I'm afraid 2x Tonga might not be enough.

Good enough for me.

It used to be 2.5x right? Maybe they increased the stock clocks a bit in response to the silly clocks of pascal.

 

[brandonmatic]

I wonder what made them lower that number from 2.5x to 2x in such a short time.

But if 2x perf/watt compared to a 380x results in 2x performance vs a 380x, that would be pretty competitive in terms of overall performance. A 200W Polaris card would give you Titan X performance if it was twice as fast as a 380x. Although it would be less efficient than the 1080.

 

[Arachnotronic]

I wonder what made them lower that number from 2.5x to 2x in such a short time.

Actual silicon results perhaps?

But if 2x perf/watt compared to a 380x results in 2x performance vs a 380x, that would be pretty competitive in terms of overall performance. A 200W Polaris card would give you Titan X performance if it was twice as fast as a 380x.

http://tpucdn.com/reviews/ASUS/R9_380X_Strix/images/perfrel_2560_1440.png

Unlikely that we will see Polaris at 200W. Rumors point to substantially lower TDPs.

 

[Glo.]

I wonder what made them lower that number from 2.5x to 2x in such a short time.

But if 2x perf/watt compared to a 380x results in 2x performance vs a 380x, that would be pretty competitive in terms of overall performance. A 200W Polaris card would give you Titan X performance if it was twice as fast as a 380x. Although it would be less efficient than the 1080.

At 200W you would get 10 TFLOPs of compute power from Polaris GPU. It would be faster than GTX 1080...

Old benchmarks for performance/watt are meaningless, because you do not know how new architecture affects performance. Only way to count overall performance is from GLOPs/watt.

 

[brandonmatic]

At 200W you would get 10 TFLOPs of compute power from Polaris GPU. It would be faster than GTX 1080...

Old benchmarks for performance/watt are meaningless, because you do not know how new architecture affects performance. Only way to count overall performance is from GLOPs/watt.

That's a good point. And "efficiency" varies by resolution for different cards when talking about benchmarks.

 

[kondziowy]

All Maxwell cards are loosing today against their price competitors from AMD + Fury X does it at 1440p and up. http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/26.html

Any improvement from here is great, especially 2x

 

[swilli89]

I don't really read a whole lot into if this is either 2x or 2.5x. I'd imagine the variables when combined are near infinite for our purposes ie, driver version + card frequency + which actual card SKUs being compared + which game suite are they using + game patch version + many others.

What's more important is the demo we saw of Polaris 11 vs a GTX 950 (we think) and it showed an excellent improvement. Granted I believe the GTX 960 is slightly better perf/watt as it is an uncut GM206 but still I think Polaris looks very promising on the efficiency side.

What I think we are looking at is an effective end to the endless and super boring conversation about which card is more efficient. Maxwell did have a tangible leg up here when it came to GM204 vs Hawaii but remember Hawaii was designed to battle Kepler, a chip it actually beats decisively in perf/watt in most modern games. The only time I remember this argument as even being close to valid is when GTX 480 launched and it caused a lot of people to be forced into buying a new PSU simply because it was so incredibly inefficient.

Beyond a discussion of which chip you would want in a laptop, its looking like this generation will be very close in perf/watt and its going to be more about perf/$ and outright performance.

 

[brandonmatic]

Unlikely that we will see Polaris at 200W. Rumors point to substantially lower TDPs.

Yes, and 2x perf/watt might not apply at higher total wattage. It will be interesting to see how the final product actually performs.

 

[IEC]

Multiple possible interpretations:
1) Marketing slide changed what perf/Watt was being compared to, thus the # changed
2) AMD is increasing clocks of Polaris GPUs in response to GTX 1080/1070 reported clocks, reducing perf/Watt
3) Actual silicon results from 14nm FinFET are worse than expected

Should have benchmarks within the next month or so to see what the actual perf/Watt is in terms of GFLOPs/W. Until that, any speculation is just that - speculation.

 

[mohit9206]

I hope it doesn't become 1.5x performance/watt by the time Polaris actually launches. Knowing AMD's history, it is actually quite possible.

 

[jpiniero]

2) AMD is increasing clocks of Polaris GPUs in response to GTX 1080/1070 reported clocks, reducing perf/Watt

Definitely happening, although they could simply redefine what metric they were using, ie: mobile.

3) Actual silicon results from 14nm FinFET are worse than expected

GloFo is going to GloFo Polaris, aren't they.

 

[maddie]

What do you think AMD means by "mainstream" GPUs in this case? And didn't AMD say previously Polaris was 2.5x perf/watt?

Can you link the posted slide?

When was it published?

 

[Arachnotronic]

Can you link the posted slide?

When was it published?

It's on AMD's investor website and it was published within the last couple of days.

 

[caswow]

I hope it doesn't become 1.5x performance/watt by the time Polaris actually launches. Knowing AMD's history, it is actually quite possible.

you make it sound like only amd does this :thumbsdown:

 

[linkgoron]

Obviously this depends on what "mainstream" means. 2x perf/watt of the 270x would give them 90% perf/watt of 1080 at 1080p. Is it average perf/watt? max perf/watt? up-to? at-least? At which resolution?

Anyway, marketing is useless anyway. Is Glofo disappointing? Have AMD decided to increase the clocks because of the lower price of the 1070? Maybe a mixture of both?

Users here hold AMD to a much higher standard than nvidia when they over-promise, so I guess that AMD have learned their lesson.

 

[Mopetar]

It would be hilarious if they would be able to compete with it, however, don't you think?

I think the top end Polaris chip will compete with the 1070, but that's more so because of how cut down the 1070 is than AMD being especially great.

Anyone who thinks it could stack up against a 1080 is huffing something though.

 

[boozzer]

If Polaris is GP104 competitor, then which GPU competitor is Vega 10? GP102?

So now we get into situation, where Mainstream has to compete with High end.
High-end with enthusiast.

Pretty bonkers if you would ask me. Hilarious if AMD would be able to do this, however...

heeh if polaris can compete with 1080 u mean? that would be hilarious.

 

[tviceman]

And the disappointing news starts to slowly set it in. Is perf/w the be end all metric? No. Is perf/w telling of the architecture vs. the competition? Yes.

If Polaris is 2x perf/w vs. Tonga, then AMD will be in the same situation that it was in at the start of Kepler and at the start of Maxwell - losing in perf/w, perf/mm2, and perf/transistor. They will once again be relegated to low margin prices just to stay competitive and will still be facing an uphill battle since no new parts released this year are faster than their previous best and they are releasing behind Nvidia.