AMD Polaris architecture – GCN 4.0

[TheRealDeal]

As i said above, it was on here somewhere, as i too was expecting it to be FX/Ti perfroamnce, but now everyones saying its only 390 performance, because it was maxed.

There are rumors that AMD will still be using Figi as there top end which i think would be pointless with what Nvidia just unveiled. I think this chip has to be pushing towards Fury X levels of performance. The good thing with Polaris is it's not being hyped like Fury X so it may speak for itself unlike Fury X which when released was under what everyone's expectations. Good chips speak for themselves so i am hoping that's the case here.

I think AMD would have to be trying really hard to screw this up to be around 390 performance. They could just shrink hawaii and have a faster chip due to core clock increases.

 

[queamin]

This above, they could have just shrink a 390 and get over a 980ti performance so they need to screw up really bad if they not close to a 980ti

 

[LoadsaMoney]

They aren't competing with Nvidias 1070/80, they are going for the market below the 970/290s, with cheap cards around their 390s performance, so all those on cards below those, who want to join the massive VR market, can do so, cheaply, and with better performance.

That is where AMD are going to take market share back from Nvidia.

 

[queamin]

They aren't competing with Nvidias 1070/80, they are going for the market below the 970/290s, with cheap cards around their 390s performance, so all those on cards below those, who want to join the massive VR market, can do so, cheaply, and with better performance.

That is where AMD are going to take market share back from Nvidia.

Yes but how high will they go i cannot see vega going form nano price point all the way up to titan X price point they need 4 cards for that, i can see them trying to complete with the 1070 and hopefully getting close to 1080 maybe in between the 1070 and 1080, then have vega go after the 1080 and pascal titan.

 

[andybird123]

They aren't competing with Nvidias 1070/80, they are going for the market below the 970/290s, with cheap cards around their 390s performance, so all those on cards below those, who want to join the massive VR market, can do so, cheaply, and with better performance.

That is where AMD are going to take market share back from Nvidia.

Dont AMD already have a cheap card around 390 performance?

 

[muziqaz]

Yes but how high will they go i cannot see vega going form nano price point all the way up to titan X price point they need 4 cards for that, i can see them trying to complete with the 1070 and hopefully getting close to 1080 maybe in between the 1070 and 1080, then have vega go after the 1080 and pascal titan.

If AMD goes after 1080 with Vega, that will be a major disappointment. Period.

 

[LoadsaMoney]

Im just completely bemused, as to why they wouldn't go for the 970/290 and above market, as the 970 was supposedly a phenomenal seller, never mind the amount of people who must still be on 980s/2/390s, but what do i know.

If AMD goes after 1080 with Vega, that will be a major disappointment. Period.

Well the top cards have never totally blown away the cards below them, so Vega will have to be an absolutely massive, massive increase in performance over the P10 to get above the 1080, never mind match it.

 

[queamin]

It depends how high polaris goes I think it will come close to 1080 but it depends on die size I do think if we had the same size dies polaris would be faster as it looks like pascal is a maxwell+on 16nm where polaris has a lot of changes from Gcn1.2 which had a few bottleneck, I am hoping now the gpu team are independent of the cpu team as such we see good things from them and they make better decisions about their gpu's.

 

[LoadsaMoney]

I wouldn't pin my hopes on it, as look what happened with the FuryX, that was hyped as a TX killer, but when it turned up, it didn't even outgun the 980.

 

[humbug]

I wouldn't pin my hopes on it, as look what happened with the FuryX, that was hyped as TX killer, yet it didn't outgun the old 980.

The Fury-X is a solid performer at 4K. often beating a 980TI

It doesn't do so well at 1080P because it has a DX11 batching bottleneck, that bottleneck is being addressed with Polaris, this is what Queamin is talking about.

 

[queamin]

Fury was a rush job on 28mn it has a few problems but it still does well most of the time, polaris is build for dx12 ground up, now was pascal not so sure really dx12 caught Nvidia by surprise I think.

 

[LoadsaMoney]

We'll see, i hope P10 does turn out to be a cracker, as i got caught up in the Fury hype only to be disappointed, never again.

 

[queamin]

Loads if you look at the 390 and what it can do in dx12 on 28nm what do you think polaris will do on 14nm with all it improvements if they mess this up they deserve to go bankrupt as dx12 and 16/14nm is what Amd have been waiting for for their cpu's and gpu's.

 

[queamin]

We'll see, i hope P10 does turn out to be a cracker, as i got caught up in the Fury hype only to be disappointed, never again.

Yes and I do see what you coming form and to be honest if you had 1 card I do think you would have been a lot happier.

 

[queamin]

Amd will support their older cards into dx12 for a long time, not so sure about Nvidia really.

 

[faceman123]

Will polaris have hbm?

 

[Seanspeed]

Yes but how high will they go i cannot see vega going form nano price point all the way up to titan X price point they need 4 cards for that, i can see them trying to complete with the 1070 and hopefully getting close to 1080 maybe in between the 1070 and 1080, then have vega go after the 1080 and pascal titan.

Difference between 1070 and 1080 isn't likely to be *that* big. They're both the same chip. GP100 is likely going to be another BIG jump from the 1080, much like the 680->Titan. Vega will have to compete with GP100 on its own. It's also possible they produce two Vega cards, a 350-400mm^ Vega and a 500-600mm^ Vega.

In the meantime, they use discounted Fiji to 'compete' with the 1070/1080.

 

[queamin]

Yes and that is why Polaris need to get close to 1080 or they could need 3 or 4 cards from Vega and that is a lot of cut down dies, do we know what the largest die Polaris is going to be?.

 

[queamin]

Will polaris have hbm?

No don't think so but until we get close launch you never know if a high end Polaris will have it but I would say no.

As you never know what happen until close to launch most of the time.

Amd normal keep quiet about their gpu's look at 4870 and 5870 and eye infinity no one knew about that until launch not even the name was not known.

 

[humbug]

Even the 290/390 has a DX11 CPU batching problem, i know that for a fact as i went from a 290 to a GTX 970 and given that i develop my own games i can see whats going on in the background.

The truth of the matter is Maxwell is far more efficient at batching, the result is Maxwell can push about 70% more drawcalls in DX11 than AMD can.

The reason why its far more pronounced with Fiji than it is with Hawaii-XT at 1080P is because Fiji is a massively more powerful architecture.
This is why you often see a Fiji GPU only a little faster than a Hawaii-XT, at 1080P, at higher resolution like 4K where the power of the GPU is more important than Drawcalls Fiji comes into its own.

As a GPU its actually more powerful than a GTX 980TI, Fiji's problem is the 980TI is able to use more of its rendering power than Fiji can, at least at 1080P in DX11.

Now, the head of the AMD Radeon Group, Raja Koduri, has actually admitted this problem with thier GPU's, and has said they have completely rebuilt the front end of their GPU's for Polaris to fix this problem.