AMD Polaris architecture – GCN 4.0

CAT-THE-FIFTH · 11 May 2016 at 09:14

Ayahuasca said:
You know you're struggling when flopper is already writing Polaris off

Nah,he is even more awestruck with Vega!!

CAT-THE-FIFTH · 11 May 2016 at 09:15

JediFragger said:
If Polaris is brighter than a thousand Suns, how many is Vega floppy??

Mtom · 11 May 2016 at 10:36

So it seems the AIB Polaris cards will be shown at Computex, which is logical after the end of May event. Also it seems AMD will show something at E3 as well.

Overlag · 11 May 2016 at 10:39

Gerard said:
Not a single fury x in stock, interesting.

Hard launch of some cards soon hopefully.

Overlag · 11 May 2016 at 10:42

Gerard said:
Not a single fury x in stock, interesting.

D.P. said:
That would be incredibly wasteful.

Same for cpu's though although not so much these days. Speed binning parts. Some will hit magical numbers some will almost be duds.

It's kinda like the fury, cores disabled to make a cheaper card yet it's core costs the same to make as the one in the nano and fury X.

maonayze · 11 May 2016 at 11:59

CAT-THE-FIFTH said:

That is so awesome. I love Rhod Gilbert....and not just because he is Welsh.

LoadsaMoney · 11 May 2016 at 12:32

Mtom said:
AMD allegedly pushes VEGA GPU forward to October

Doesn't make sense to me that tbh, as they are in a different market, so why would they panic about the 1070/80, and bring Vega forward.

CAT-THE-FIFTH · 11 May 2016 at 12:40

LoadsaMoney said:
Doesn't make sense to me that tbh, as they are in a different market, so why would they panic about the 1070/80, and bring Vega forward.

Maybe so they can launch it with Zen??

queamin · 11 May 2016 at 12:51

The lower Vega might be against the 1080

Mei · 11 May 2016 at 12:56

AMD should release some slides of vega performance
that would be pretty great

andybird123 · 11 May 2016 at 13:12

AllBodies said:
Also this means if AMD were to use the maximum 2.3x increase possible with 14LPP, they could fit 7.1 Bn transistors in their 232mm2 Polaris 10. So only 0.1 Bn less than the 1080. So I won't count out Polaris 10 till we see the full details

where did you get this figure of 2.3 from?
Samsung's own documentation says 28nm to 14nm LPP is a 0.55 reduction, 1.9 times, a bit more than the 1.7 or so that nvidia have gone with (apparently to go for the clock speed advantages), but not to the extent you seem to be suggesting

14nm LPP and 16ff+ are not drastically different and are based on 20nm tech, rather than Intels 14nm which is more accurately called 14nm (TSMC's 10nm is about the same as Intel's 14nm)

N19h7m4r3 · 11 May 2016 at 13:13

andybird123 said:
where did you get this figure of 2.3 from?
Samsung's own documentation says 28nm to 14nm LPP is a 0.55 reduction, 1.9 times, a bit more than the 1.7 or so that nvidia have gone with (apparently to go for the clock speed advantages), but not to the extent you seem to be suggesting

the 2.3x is AMD saying upt to 2.5x the performance per watt compared to their current offerings.

andybird123 · 11 May 2016 at 13:15

N19h7m4r3 said:
the 2.3x is AMD saying upt to 2.5x the performance per watt compared to their current offerings.

but hes trying to apply it to transistors per mm :/

N19h7m4r3 · 11 May 2016 at 13:20

andybird123 said:
but hes trying to apply it to transistors per mm :/

Yeah that doesn't really fit into what AMD were saying at all, they specifically only talk about it in regards to performance to watt.

In the end though, we just need to wait. I'll be selling my remaining 980Ti and getting a Polaris card to hold me over until Vega, but with Vega apparently coming so soon, we'll have to see what they really do.

Hoping they have something substantial at the end of the month.

Muz333 · 11 May 2016 at 13:27

Think I'll go 1080 this time around but I'm holding off to see what Polaris may bring. I think it's going to be a 290 equivalent but the slightly smaller node may surprise with performance compared to Nvidias offerings?

nomadd · 11 May 2016 at 13:38

JediFragger said:
If Polaris is brighter than a thousand Suns, how many is Vega

It's champagne supernova.
In the sky.

LoadsaMoney · 11 May 2016 at 15:47

AllBodies · 11 May 2016 at 16:01

andybird123 said:
where did you get this figure of 2.3 from?
Samsung's own documentation says 28nm to 14nm LPP is a 0.55 reduction, 1.9 times, a bit more than the 1.7 or so that nvidia have gone with (apparently to go for the clock speed advantages), but not to the extent you seem to be suggesting

14nm LPP and 16ff+ are not drastically different and are based on 20nm tech, rather than Intels 14nm which is more accurately called 14nm (TSMC's 10nm is about the same as Intel's 14nm)

20nm Planar is supposed to be 2x (or 1.82x depending on which 28nm process is compared against)

Samsung/GloFo released this slide at a presentation saying their 14nm process was 'up to 15%' denser than 'others' (so TSMC).

TSMC say 16nm is 2x the density of 28nm, so that means Samsung are claiming 2.3x.

So the 14LPP process is somewhere between 2.1-2.3x the density of 28nm, depending on who's 28nm you use. But it is denser than TSMC's 16nm, that's for sure.

flopper · 11 May 2016 at 16:03

Muz333 said:
Think I'll go 1080 this time around but I'm holding off to see what Polaris may bring. I think it's going to be a 290 equivalent but the slightly smaller node may surprise with performance compared to Nvidias offerings?

price/performance ratio might be something special.

JediFragger · 11 May 2016 at 16:10

flopper said:
price/performance ratio might be something special.

What about performace per inch???