AMD: Our Next Generation (Arctic Islands) GPUs will offer Double the Performance Per Watt of the Current Generation

[Mr_Troll]

It looks like the GPUs in 2016 are going to be a major step up from current gen, because we keep on hearing about performance basically doubling, thanks to the transition to the smaller node. There have been reports in the past about Nvidia’s Pascal having 17 Billion transistors (more than twice as many as the Maxwell flasghip) and even now at AMD’s earnings call, Lisa Su stated that the next generation AMD graphic cards will double the performance per watt delivered by the current generation (Fiji) lineup.
 

AMD’s Arctic Islands lineup will offer atleast ‘double the performance per watt’ of Fiji processors

 

While it has been implied in the past that the performance jump is going to be enormous, this is one of the few times where the company has explicitly mentioned just how big the jump is going to be. Lisa Su made the following statement on the recently held earnings call for the results of the third quarter, year ended 2015:The alleged codename of the Arctic Islands’ flagship is ‘Greenland’ and if what AMD says is true (and there is no reason for it to not be just that) we are looking at a 100% increase in performance over Fiji. But just what exactly does that translate to in terms of numbers? Using what we know about the FinFET process, we can arrive at a tentative answer.

 

 

“We are also focused on delivering our next generation GPUs in 2016 which is going to improve performance per watt by two times compared to our current offerings, based on design and architectural enhancements as well as advanced FinFET products process technology.” – Lisa Su, AMD

 

 

While Nvidia is confirmed to be on the 16nm FinFET+ node from TSMC, the same cannot be said about AMD. We do however, have more authentic reports about AMD being on the 16FF+ node as opposed to 14nm FinFET. So if we are to assume that the same is true for AMD, then it is going to adhere to the same rules as Nvidia’s next process. Here is an official summary of the 16FF+ node from TSMC official:
 

TSMC’s 16FF+ (FinFET Plus) technology can provide above 65 percent higher speed, around 2 times the density, or 70 percent less power than its 28HPM technology. Comparing with 20SoC technology, 16FF+ provides extra 40% higher speed and 60% power saving. By leveraging the experience of 20SoC technology, TSMC 16FF+ shares the same metal backend process in order to quickly improve yield and demonstrate process maturity for time-to-market value.

 

 

Fiji GPU has a total of 4096 stream processors and 8.9 Billion transistors and the die size is 596mm^2. So for double the performance, we are looking around 18 Billion transistors for the Greenland (Arctic Islands) chip. Ofcourse, if the die size is ~500mm^2 initially then anything in the range of 15 Billion transistors goes. For the ~500mm assumption, after accounting for performance increases via architectural improvements, a gain of 66%+ (anything between 70-80% realistically) can be expected. While that is not a full 100%, it is close enough. Ofcourse we have no idea what the stream processor size will be, so attempting to reverse engineer the exact core count would be an exercise in futility. It is a safe bet however that the number would be somewhere around the 8192 mark for a ~600mm die or around the 6784 mark for the ~500mm die (plus minus a few multiples of 128).

 

The GM200 on the other hand, has 8 Billion transistors, so double the density would yield around 16 Billion transistors for a ~600mm chip according to TSMC specs assuming the transistor specifications remain the same. If we are to assume a ~500mm^2 chip for the initial 16nm FinFET+ run then the transistor count would be roughly 14 Billion transistors. If the rumor about Nvidia bringing 17 Billion transistors to the table is true, then either TSMC can manage a ~600mm^2 chip right off the bat, or, Nvidia’s next generation Pascal GPUs will have a new transistor-to-core ratio and design (improbable).

 

While the fact that TSMC is using the 20nm back end process helps in the claims of a ~600mm^2 die, I myself,  have difficulty believing that these companies will launch with such huge dies. It is possible that competition might force them to do so, but in my opinion, the probability of an initial ~500mm^2 die is much greater. At any rate, we are going to have an absolutely amazing showdown in 2016, with the first full fledged, 4K @60fps monsters landing in the DirectX 12 arena. It is most definitely a good time to be a PC gamer.

 

 

So double the performance of fiji. Pure marketing BS. Dont trust this until i see benchmarks

 

 

Source: http://wccftech.com/amd-next-generation-arctic-islands-graphic-cards-double-performance-watt-current-generation/

AMD Q3 earning call

[ThatOneRussian]

Pretty sure Ive heard this before

[don_svetlio]

Performance per Watt - that means lower power consumption and lower thermals from what I gather. I guess 5000 series is repeating

They already proved it possible with the Nano

[Sauron]

So double the performance of fiji. Pure marketing BS. Dont trust this until i see benchmarks

 

that is NOT what they said. they said 2x performance per watt. It's VERY different.

[Necrodead]

"per watt" is key here I think, they will only be abit more powerful overall, but require much less power than current gen. IMO.

[patrickjp93]

Maybe if AMD can push clock speeds they can take a momentary lead.

 

 

that is NOT what they said. they said 2x performance per watt. It's VERY different.

In the same wattage, double the performance. Therefore expecting a Fury X-TDP flagship to have double the performance of the Fury X is not unreasonable when this is the statement.

[Sauron]

In the same wattage, double the performance. Therefore expecting a Fury X-TDP flagship to have double the performance of the Fury X is not unreasonable when this is the statement.

 

where did they say it was the same wattage?

[Guest Kloaked]

Good stuff. Now Sapphire cooled cards will be even better should this be true.

[HKZeroFive]

You literally copy the entire article and just input a sentence as your opinion that is actually incorrect. I'm pissed off because I was literally a minute off from posting this.

Now I'm salty as the Payday 2 fans.

[Mr_Troll]

You literally copy the entire article and just input an opinion that is actually incorrect. I'm pissed off because I was literally a minute off from posting this.

so what does  2x performance/watt mean?... i think double the performance while having same wattage.

[patrickjp93]

where did they say it was the same wattage?

facepalm* They didn't. What I'm saying is if they provide a GPU with the same TDP as the Fury X, it should have double the performance since that's a direct consequence of the claim they've made. If they said double the performance per watt below a certain threshold, they could avoid the scrutiny, but no, typical AMD marketing team failures...

 

E(AI) = 2E(F) where E(x) = perf(x)/TDP(x)

 

perf(AI)/TDP(AI) = 2perf(F)/TDP(F)   =>   perf(AI) = 2perf(F)   iff. TDP(F) = TDP(AI).

 

@MageTank, once again, the math is itself a source. Your mind blown yet? 

[Suika]

I remember a certain, much greener company making this same exact claim.

[AresKrieger]

So they're going to decrease power consumption cool, though this means nothing to me as it doesn't actually mean big gains in performance, sigh

 

so what does  2x performance/watt mean?... i think double the performance while having same wattage.

Doubtful, it only means that they are reducing power consumption or they would have worded it differently

[patrickjp93]

so what does  2x performance/watt mean?... i think double the performance while having same wattage.

Which is a correct alternative interpretation, see functional explanation of that above.

[Sauron]

facepalm* They didn't. What I'm saying is if they provide a GPU with the same TDP as the Fury X, it should have double the performance since that's a direct consequence of the claim they've made.

 

E(AI) = 2E(F) where E(x) = perf(x)/TDP(x)

 

perf(AI)/TDP(AI) = 2perf(F)/TDP(F)   =>   perf(AI) = 2perf(F)   iff. TDP(F) = TDP(AI).

 

@MageTank, once again, the math is itself a source. Your mind blown yet? 

 

and what I'm saying is that they did not say they will. your mind blown yet?

[patrickjp93]

and what I'm saying is that they did not say they will. your mind blown yet?

And I'm saying they will since they'll have to so that they may remain competitive. Both Nvidia and AMD have moved into the 300W space for flagship cards. They'll remain there.

[Stefan1024]

There will be an efficient card with 2x performance / watt conpared to Fiji like the Nano.

Also the top of the line card will be pusged to the edge and deliver ~1.5x performance for 250-300 watts. Still impressive by the way.

Of cource the 2x performance / watt will not apply to the entire lineup.

[Guest Kloaked]

and what I'm saying is that they did not say they will. your mind blown yet?

 

They're saying that if AMD is able to deliver double the performance per watt, it'll literally be double the performance of the Fiji cards if they release a card with the same wattage as a Fury X.

 

It's really that simple 

[AlexGoesHigh]

No shit, they should, 28nm lasted forever, it has to the longest time they been stuck using a single process node, 900 and R9 300 series should have been made with 20nm but the fab got them ready to that late that they jumped ship straight to 14/16nm depending on which fab makes it, most of the improvement will be from the shrink rather that architecture improvements.

[The Benjamins]

so what does  2x performance/watt mean?... i think double the performance while having same wattage.

yes or same performance at half the wattage

[Necrodead]

And I'm saying they will since they'll have to so that they may remain competitive. Both Nvidia and AMD have moved into the 300W space for flagship cards. They'll remain there.

If you look at CPU's, the die shrink means that cpu's are using less power to do more stuff. This will undoubtly be true for GPU's as well since silicon is the same weather it be GPU or CPU. Next gen cards will not be the 300W monsters we have today, but will probably be 150W cards with all the power of todays high end cards.

[Necrodead]

They're saying that if AMD is able to deliver double the performance per watt, it'll literally be double the performance of the Fiji cards if they release a card with the same wattage as a Fury X.

 

It's really that simple 

But they won't make a 300W card, I gaurentee it.

[ICantThinkOfAnyGoodName]

This is probably just as true as Nvidia`s claims that Maxwell would have twice the performance/watt of Kepler.

[patrickjp93]

This is probably just as true as Nvidia`s claims that Maxwell would have twice the performance/watt of Kepler.

That's pretty close to true these days after all the driver optimizations.

[gonnog]

Super great, but no one cares about the wattage. Whats the difference between 200W and 250W for standard consumer? We want POWER!!!!!!!

 

2xperfomance per Watt is still kind of good. Everything between 50%-100% is impresive.