Pascal consumer = Maxwell on speed confirmed

[AlTech]

28 minutes ago, raphidy said:

Ah yes, AdoredTV. Everything he says is the truth now. I'm still waiting for the efficiency to come from AMD. I was pretty disappointed with polaris though.

Each AMD Polaris Compute Unit is 15-20% faster and more efficient than AMD's Hawaii Compute Unit.

 

So 36 Polaris CUs is roughly equal to 41.4-43.2 Hawaii CUs

 

Or 2304 Polaris Stream processors are equal to 2650-2765 Hawaii Stream Processors. Hence why it beats the 290 and 390 easily.

 

EDIT: But this is just per CU performance. So let's take into account Polaris 10 higher clock speed of 1200Mhz instead of Hawaii's 1000Mhz clock speed.

 

And then you have another 20% boost. So then 36 Polaris CUs would be equal to 50.4 Hawaii CUs.

 

OR roughly 2304 Polaris SPs are equal to 3226 Hawaii SPs

 

I'd give AMD a gold star. Good job AMD. @don_svetlio Can tell you this is probably correct.

[Sintezza]

19 minutes ago, Dabombinable said:

The main thing with Pascal is the significantly lower power consumption. Yeah they perform the same clock for clock, but when Pascal consumes a shitload less power for the same performance...I don't see why people have a problem.

 

Yeah i dont realy see the point either.

I mean who cares, because the Pascal cards have higher clock speeds, they also perform signicantly better.

And consume less power.

I personaly dont realy see the whole point of this.

Their claims hit nor land nor sea.

The only fact is that Pascal is a new architecture no matter how people look at it.

[Sintezza]

7 minutes ago, AluminiumTech said:

Each AMD Polaris Compute Unit is 15-20% faster and more efficient than AMD's Hawaii Compute Unit.

 

So 36 Polaris CUs is roughly equal to 41.4-43.2 Hawaii CUs

 

Or 2304 Polaris Stream processors are equal to 2650-2765 Hawaii Stream Processors. Hence why it beats the 290 and 390 easily.

 

EDIT: But this is just per CU performance. So let's take into account Polaris 10 higher clock speed of 1200Mhz instead of Hawaii's 1000Mhz clock speed.

 

And then you have another 20% boost. So then 36 Polaris CUs would be equal to 50.4 Hawaii CUs.

 

OR roughly 2304 Polaris SPs are equal to 3226 Hawaii SPs

 

Is the RX480 that much better in terms of raw performance then the 390X?

I cannot remember that i have realy seen that.

It kinda trade blows with a R9-390 depending on the game which comes ontop.

[The Benjamins]

3 minutes ago, Sintezza said:

 

Is the RX480 that much better in terms of raw performance then the 390X?

I cannot remember that i have realy seen that.

 

AFAIK it is right in between a 290 to 290x

[AlTech]

5 minutes ago, Sintezza said:

 

Is the RX480 that much better in terms of raw performance then the 390X?

I cannot remember that i have realy seen that.

It kinda trade blows with a R9-390 depending on the game which comes ontop.

390X and 390 are special. they're not Hawaii 1.0 which is what I was referring to. I was more talking about compared to 290 and 290X.

[Sintezza]

Just now, AluminiumTech said:

390X and 390 are special. there's not Hawaii 1.0 which is what I was referring to. I was more talking about compared to 290 and 290X.

Yeah but Grenada is Hawai just with higher clockspeeds.

[AlTech]

Just now, Sintezza said:

Yeah but Grenada is Hawai just with higher clockspeeds.

And higher VRAM speed (5GHz vs 6Ghz).

 

And some efficiency changes. Hawaii != Grenada.

 

[Sintezza]

3 minutes ago, AluminiumTech said:

And higher VRAM speed (5GHz vs 6Ghz).

 

And some efficiency changes. Hawaii != Grenada.

 

Yeah they did some minor tweaks to the cards so that they allowed higher clocks.

But thats pretty much it, nothing special.

Still the same architecture, still the same sp´s.

 

Anyway, i think that we should stay at the Pascal topic

 

[AlTech]

4 minutes ago, Sintezza said:

Yeah but Grenada is Hawai just with higher clockspeeds.

They put in some efficiency changes they put into the Fury series. Which is why it was quieter and ran cooler.

[DocSwag]

Can I please say something?

 

It is 100% true that the cores themselves-ALUs n stuff, are completely the same as Maxwell. HOWEVER, Nvidia has changed quite a bit on the front end of Pascal (and in the process improved Async compute as well as added new features like SMT) as well as improved Delta Color Compression, which IS NOT software. Need proof? Read this entire article then come back: http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review

[Briggsy]

AdoredTV is a classic example of the Dunning-Kruger effect (are the stupid too stupid to realize they're stupid), where his lack of understanding causes him to build a house of cards out of every topic and find conclusions wherever his bias is confirmed. A stopped clock is right twice a day, A blind squirrel finds his nuts every so often, etc..

 

His conclusions about Pascal answer a question that only he asked, in a pre-emptive response to the notion that Polaris is actually Hawaii on speed with color compression, hardware culling, half the ACE's and 2/3 actual power draw to match performance (but that is being nit-picky, am I right?).

[raphidy]

25 minutes ago, AluminiumTech said:

snip

IIRC, where's the 85W 1440p 60fps on Battlefront that I heard back from January 2016 or December 2015? RX480 draws like 175-200W? Also, Polaris does not even support full dx12.1. I mean even pascal has dx12.1 with their lame async compute.

[Thony]

1 hour ago, don_svetlio said:

We need to wait for Vega to say whether AMD failed or not - 480 was never meant to compete against the 1070/1080/980 Ti

How does that help if 480 can't even compete with GTX 1060 ????

 

Vega 1150/2050 (whatever stupid name they go with) will outperform RX480.... The gap only gets bigger with newer technology not the other way round.

 

 

[Centurius]

1 hour ago, Dresta said:

And what does that mean for gamers? oh yea more cost!!!

Someone who bought it for gaming without research deserved it.

[AlTech]

1 minute ago, raphidy said:

IIRC, where's the 85W 1440p 60fps on Battlefront that I heard back from January 2016 or December 2015? RX480 draws like 175-200W? Also, Polaris does not even support full dx12.1. I mean even pascal has dx12.1 with their lame async compute.

nobody claimed 85w anything. I think your mistaking it for the GTX 950 vs RX 460 BF4 comparison on 1080p.

 

Polaris supports DX 12 FL12_0 whereas pascal support DX12 FL12_1.

 

But that's ok since Maxwell and Maxwell 2 Support DX12 FL11_1.

 

And why does 12_1 matter more than 12_0? DX12 games can adapt to support whatever feature level the GPU support.

[Briggsy]

12 minutes ago, raphidy said:

IIRC, where's the 85W 1440p 60fps on Battlefront that I heard back from January 2016 or December 2015? RX480 draws like 175-200W? Also, Polaris does not even support full dx12.1. I mean even pascal has dx12.1 with their lame async compute.

to be fair I believe that was a Star Wars Battlefront demo at 1080p with undisclosed settings, and the game was locked at 60FPS. I wouldn't put too much stock in that test because AMD Powertune lowers clocks when less performance is needed, while Nvidia clocks say at full speed but overall utilization is lowered. We know that AMD power draw plummets with clock decreases (i.e. R9 Nano vs FuryX = 175 vs. 275 watts for an average decrease in shader clocks of 150-200Mhz). That test had red flags all over it, playing to Powertunes strengths and hiding Polaris' power draw weaknesses.

[Dabombinable]

34 minutes ago, AluminiumTech said:

They put in some efficiency changes they put into the Fury series. Which is why it was quieter and ran cooler.

Are you on something? The only difference was higher quality components, and a different BIOS. Grenada is Hawaii, and the Fury series/Fiji is based heavily on Tonga-the replacement for Hawaii/Grenada.

[rattacko123]

pascal is SUPER efficient, when underclocked, the 1060 has 60W power draw and beats the RX 480! (in dx11 games).

AMDs architecture doesnt scale very well, even at low clocks it still uses a lot of power

[skyler_mertz]

Would this mean that future driver updates through GeForce Experience won't nerf my 980Ti like Nvidia did with Keplar?

[Briggsy]

1 minute ago, skyler_mertz said:

Would this mean that future driver updates through GeForce Experience won't nerf my 980Ti like Nvidia did with Keplar?

Nvidia didn't nerf kepler, AMD just took this long to optimize Hawaii drivers.

[AlTech]

6 minutes ago, Briggsy said:

Nvidia didn't nerf kepler, AMD just took this long to optimize Hawaii drivers.

No they did nerf Kepler.You don't own a Kepler GPU so you're on no position to inform others of this.

[AlTech]

9 minutes ago, skyler_mertz said:

Would this mean that future driver updates through GeForce Experience won't nerf my 980Ti like Nvidia did with Keplar?

They will but later on than Kepler was.

[skyler_mertz]

3 minutes ago, Briggsy said:

Nvidia didn't nerf kepler, AMD just took this long to optimize Hawaii drivers.

Ah, based on the little bit of phone research I did, Maxwell is up to 3 times better at tesslation than Keplar. So modern games with old drivers vs new seems that Keplar gets the shaft. 

 

Im hoping Maxwell is still a contender going into the next year seeing that Pascal is (roughly) shrunken Maxwell. 

[AlTech]

19 minutes ago, Dabombinable said:

Are you on something? The only difference was higher quality components, and a different BIOS. Grenada is Hawaii, and the Fury series/Fiji is based heavily on Tonga-the replacement for Hawaii/Grenada.

They put some architectural efficiency changes from Tonga/Fury to Grenada.

 

They were minor efficiency tweaks.

[The Benjamins]

1 minute ago, skyler_mertz said:

Ah, based on the little bit of phone research I did, Maxwell is up to 3 times better at tesslation than Keplar. So modern games with old drivers vs new seems that Keplar gets the shaft. 

 

Im hoping Maxwell is still a contender going into the next year seeing that Pascal is (roughly) shrunken Maxwell. 

Their was test showing in the same game that new drivers saw a performance dip. can find it now though.