[UPDATE] AMD Announces Polaris Architecture - 4th Generation GCN Arriving in Mid 2016

[Fetzie]

Interesting name from a company hell bent on poralis-ing the community.

Don't quit the day job

[Majestic]

Don't quit the day job

 

Give it a rest,

[Exotath]

Well I hope to see a flagship card with this soon then.

[samcool55]

I'm confused... Is it 16nm or 14nm?

 

TL:DR says 16nm but amd's video says 14nm, so i have no idea what it is now...

I'm thinking 14? Help me D: someone...

[fire219]

I'm kinda disappointed in everyone convinced AMD's test is rigged.

 

Yes, the power draw looks low on the Polaris system, but doesn't it also look low on the 950 powered system? I think they did some trickery with the meters to just measure the GPU power consumption. Maybe put the cards on a separate PSU to the rest of their systems?

[ivan134]

should i feel bad for buying a 390 last year on black friday?

Depends on what you had before that.

[Tomsen]

I'm confused... Is it 16nm or 14nm?

 

TL:DR says 16nm but amd's video says 14nm, so i have no idea what it is now...

I'm thinking 14? Help me D: someone...

It's both 16nm and 14nm.

Ryan explains it in the anandtech article.

[suicidalfranco]

Depends on what you had before that.

 

a laptop... with a 650M

[ivan134]

a laptop... with a 650M

I wouldn't feel bad then, lol. Your 390 won't stop working because new cards came out. It's still a solid 1080p/1440p card.

[WereCat]

let's get back to AMD's bullcrap testing shall we?

 

a non-reference GTX950 has a peak power draw (just the video card) of 110W http://media.bestofmicro.com/O/G/520576/gallery/Power_w_600.png

system 2, as shown, drew at the wall 152W - that leaves the rest of the system to use 42W

 

now, system 1 draws at the wall 85W, minus 42W would make the Polaris based card to draw 43W on it's own ?!

 

 

note: reference GTX950 is rated for 90W power draw (not TDP !!!) - http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-950/specifications

By the way both cards were at Power Limit set to 80% so 950 was not using 110W but around 90W max.

[Fetzie]

By the way both cards were at Power Limit set to 80% so 950 was not using 110W but around 90W max.

The CPUs were also both set to 80% power limit.

[Tomsen]

The polaris GPU is running at 850Mhz / 0.8375V.

I dont see their numbers as been completely out of reach..

[Sakkura]

So I can assume this new architecture won't be good because they've been alternating between good and bad every generation-

tahiti- good

hawaii- bad

fiji- good

polaris- bad?

 

What? Hawaii was great. At the time, it nearly matched the MUCH bigger GK110, and nowadays it actually tends to beat GK110. Hawaii still competes well against the much newer GM204.

[patrickjp93]

By the way both cards were at Power Limit set to 80% so 950 was not using 110W but around 90W max.

 

 

The CPUs were also both set to 80% power limit.

 

 

The polaris GPU is running at 850Mhz / 0.8375V.

I dont see their numbers as been completely out of reach..

 

You three do realize no actual proof was presented, right? When Anandtech digs into this and busts AMD for lying their tails off I'm gonna laugh all the way to the bank.

[don_svetlio]

You three do realize no actual proof was presented, right? When Anandtech digs into this and busts AMD for lying their tails off I'm gonna laugh all the way to the bank.

14nm could very much yield low power GPUs. Beating a 950 isn't exactly hard.

[Sakkura]

let's get back to AMD's bullcrap testing shall we?

 

a non-reference GTX950 has a peak power draw (just the video card) of 110W http://media.bestofmicro.com/O/G/520576/gallery/Power_w_600.png

system 2, as shown, drew at the wall 152W - that leaves the rest of the system to use 42W

 

now, system 1 draws at the wall 85W, minus 42W would make the Polaris based card to draw 43W on it's own ?!

 

note: reference GTX950 is rated for 90W power draw (not TDP !!!) - http://www.geforce.com/hardware/desktop-gpus/geforce-gtx-950/specifications

 

The GTX 750 Ti draws about 50W when gaming. This is basically a 14nm equivalent of a GTX 750 Ti. The power consumption is not unrealistic, though I'm sure AMD has taken a bit of a best-case reading. So where you're calculating a 43W draw, maybe it'll come in closer to 50W in reviews.

 

Edit: This particular GPU seems to be 14nm, not 16.

[Tomsen]

You three do realize no actual proof was presented, right? When Anandtech digs into this and busts AMD for lying their tails off I'm gonna laugh all the way to the bank.

Oh most certainly. I'm very aware that this is an example to put polaris efficiency gains in the best light as possible.

I'm however, going to be doubtful that anandtech will bust AMD for anything, as nothing was really presented..

[LokiFire]

Can't wait to see what FinFet gives us.

[patrickjp93]

14nm could very much yield low power GPUs. Beating a 950 isn't exactly hard.

 

No, Dennard Scaling died back at 90nm. This expectation is insane.

 

Oh most certainly. I'm very aware that this is an example to put polaris efficiency gains in the best light as possible.

I'm however, going to be doubtful that anandtech will bust AMD for anything, as nothing was really presented..

They presented that, at equal performance levels and (presumed but not proven) equal system setups, AMD's GPU architecture is 65% more energy efficient. It's utter BS by the above statement. It's the same reason why ARM stagnated at the A57 core and had to pursue clock speeds to get better performance while sacrificing lower power to get the A72. Power scaling with node shrinkage slowed down to a crawl after 90nm. Further, it's Global Foundries. Even with Samsung's 14nm tech (designed for low power SOCs, not performance GPUs), do you honestly believe this given its track record of consistent disappointment against TSMC?

[LokiFire]

No, Dennard Scaling died back at 90nm. This expectation is insane.

 

They presented that, at equal performance levels and (presumed but not proven) equal system setups, AMD's GPU architecture is 65% more energy efficient. It's utter BS by the above statement. It's the same reason why ARM stagnated at the A57 core and had to pursue clock speeds to get better performance while sacrificing lower power to get the A72. Power scaling with node shrinkage slowed down to a crawl after 90nm. Further, it's Global Foundries. Even with Samsung's 14nm tech (designed for low power SOCs, not performance GPUs), do you honestly believe this given its track record of consistent disappointment against TSMC?

If so explain Fermi to Kepler for us please as that was 40NM to a lower process whilst making big gains in performance, NVIDIA pulled efficiency from nowhere?

[Belgarathian]

I'm cheering for your, RTG. Please don't let me down, I have fond memories of ATI since the x1900xtx days. 

[patrickjp93]

If so explain Fermi to Kepler for us please as that was 40NM to a lower process whilst making big gains in performance, NVIDIA pulled efficiency from nowhere?

 

No, Fermi was just bad design. GCN isn't bad design. It's just dense (HDL-High Density Libraries and all that). While FinFET voltage control is very different, did we all forget what happened to Intel upon moving to FinFET? Heat went up, and overclocking headroom came down. That was true for both the mainstream and HEDT platforms, where HEDT chips were still soldered to their heat spreaders.

 

Nvidia didn't get it from nowhere. They underwent a paradigm shift in design philosophy. While Koduri hasn't been with AMD for a while, GCN is still based on his previous work. GCN 2.0/4.0 won't be a drastic architectural change other than allowing more TMUs and ROPs as Fiji should have.

[Tomsen]

They presented that, at equal performance levels and (presumed but not proven) equal system setups, AMD's GPU architecture is 65% more energy efficient. It's utter BS by the above statement. It's the same reason why ARM stagnated at the A57 core and had to pursue clock speeds to get better performance while sacrificing lower power to get the A72. Power scaling with node shrinkage slowed down to a crawl after 90nm. Further, it's Global Foundries. Even with Samsung's 14nm tech (designed for low power SOCs, not performance GPUs), do you honestly believe this given its track record of consistent disappointment against TSMC?

Have you taken the Nvidias node disadvantage into account?

It really don't seems so unlikely. Systems was locked in med setting with a 60 fps frame cap.

ARM is stagnated because it is in the period of transformation, and these things take time. There are a few ARM archictures coming targeting higher performance. All the got in the meantime are low power designs to tweak.

Also, this seems to be a low-end unit. Who is to say big dies aren't to be fabbed at TSMC?

[WereCat]

You three do realize no actual proof was presented, right? When Anandtech digs into this and busts AMD for lying their tails off I'm gonna laugh all the way to the bank.

I am simply discussing about information that were in OP. If they were lying or not doesn't matter.. Neither of us have any proof about that. My opinion is the results they got are plausible.

[Sakkura]

No, Dennard Scaling died back at 90nm. This expectation is insane.

 

Expecting a low-end 14nm GPU to draw a bit less power than a low-end 28nm GPU (GM107) is not insane.

 

 

Power scaling with node shrinkage slowed down to a crawl after 90nm.

 

Yeah that's why the GTX 660 couldn't deliver over 90% the performance of a GTX 580 with half the power consumption.

 

Oh wait. It could.