AMD Polaris architecture – GCN 4.0

humbug · 13 May 2016 at 15:55

flopper said:
seems Polaris will totally own Pascal.

Doubt that but i'm trying to work out how much perforamce, what speed was Polaris 11 running at in that demo matching a GTX 950.

Is it just me or did someone say 850Mhz? be that as it may @ 1400Mhz its running 65% faster then in that demo, thats gotta be a 970, this for the baby Polaris.

Mauller · 13 May 2016 at 16:06

humbug said:
Doubt that but i'm trying to work out how much perforamce, what speed was Polaris 11 running at in that demo matching a GTX 950.

Is it just me or did someone say 850Mhz? be that as it may @ 1400Mhz its running 65% faster then in that demo, thats gotta be a 970, this for the baby Polaris.

No one knows the speed it was running, for all we know it could have been running at full clocks with a frame cap.

But even at 1ghz, if it does end up having pitcairn numbers of shaders, then its already beating the 950 and on its way to beating the 960

humbug · 13 May 2016 at 16:10

Mauller said:
No one knows the speed it was running, for all we know it could have been running at full clocks with a frame cap.

But even at 1ghz, if it does end up having pitcairn numbers of shaders, then its already beating the 950 and on its way to beating the 960

How do you work that out?

Mauller · 13 May 2016 at 16:12

humbug said:
How do you work that out?

Becasue the 270x which is pretty much pitcairn with better ram is already above the 950 and on average a few fps behind the 960 in benchmarks.

Lokken86 · 13 May 2016 at 16:13

Pinch of salt on board, if that is the specs they are really dull and I wouldn't call the faster one anything but 480x. Saying that they will probably sell if priced correctly, bringing 1440p to the masses.

humbug · 13 May 2016 at 16:14

Mauller said:
Becasue the 270x which is pretty much pitcairn with better ram is already above the 950 and on average a few fps behind the 960 in benchmarks.

Ah.... the 270X has 1280 Shaders, Pitcairn.
898 is Bonaire, 260X

So... erm, wow.

Rroff · 13 May 2016 at 16:26

humbug said:
Doubt that but i'm trying to work out how much perforamce, what speed was Polaris 11 running at in that demo matching a GTX 950.

Is it just me or did someone say 850Mhz? be that as it may @ 1400Mhz its running 65% faster then in that demo, thats gotta be a 970, this for the baby Polaris.

At 1400MHz if those specs are true that places it around 970-980 performance :S assuming architecture differences mean the TF is closer to actually what it can bring to in game performance then a few handful of percent faster than a 980.

Was hoping the rumours of around 1600MHz were true which would make it more competitive albeit a little behind the 1070.

fs123 · 13 May 2016 at 16:26

humbug said:
Can we leave Nvidia posts for Nvidia threads now please?

The Nvidia salesmen can't resist though. Every bit of Polaris info has to be countered with what Nvidia can do.

AllBodies · 13 May 2016 at 16:43

LoadsaMoney said:
http://videocardz.com/59903/possible-polaris-10-and-polaris-11-specifications-emerge

I'm extremely sceptical about these figures.

Firstly, as others mentioned, it makes no sense for a 232mm2 14nm chip to only have 2048 cores, unless the cores are much larger than previous GCN.

But also the R7 260X is only 160mm2, so that should make Polaris 11 less than 80mm2, so again that doesn't add up.

humbug · 13 May 2016 at 16:53

Lets do some maths....

P11 898 Shader
P10 2048 Shader (+130%)

by that ^^^ chart GTX 950 and P11 is at 46%

GTX 950: 46%
GTX 970: 93%
R9 390: 97%
Fury-X: 120%
980TI: 128%

Typical shader scaling is 0.7 with GDDR5X 0.8?
80% of 130 = 109.

P11 @ 46% + 109% = 96% = R9 390 = 470X = GTX 1070 for me.

Take my dodgy maths and the nothing we know about what Mhz P11 was running at in the demo'- how you like

If it is BS someone has carfully worked out these slides to look like they match the unsubstantiated 390 performance rumours, or they are correct.
I'd be interested to see, in 2 weeks, which it is, if it is BS someone has really got it in for AMD.

Mauller · 13 May 2016 at 16:55

humbug said:
Lets do some maths....

P11 898 Shader
P10 2048 Shader (+130%)

by that ^^^ chart GTX 950 and P11 is at 46%

GTX 950: 46%
GTX 970: 93%
R9 390: 97%
Fury-X: 120%
980TI: 128%

Typical shader scaling is 0.7 with GDDR5X 0.8?
80% of 130 = 109.

P11 @ 46% + 109% = 96% = R9 390 = 470X = GTX 1070 for me.

Take my dodgy maths and the nothing we know about what Mhz P11 was running at in the demo how you like

To be honest that would be very good for that small a die, considering P11 is the smaller of the two Polaris parts.

D.P. · 13 May 2016 at 16:56

fs123 said:
The Nvidia salesmen can't resist though. Every bit of Polaris info has to be countered with what Nvidia can do.

People should not start talking about Pascal and posting nonsense about nvidia if they don't want to see that in this thread, simples.

humbug · 13 May 2016 at 17:04

Mauller said:
To be honest that would be very good for that small a die, considering P11 is the smaller of the two Polaris parts.

Its the third 290/970 level card in a row from AMD, if my numbers bear out....

I hope P10 is sub £200 and its certainly not for me, i don't think i can wait till October/November.

Mauller · 13 May 2016 at 17:05

humbug said:
Its the third 290/970 level card in a row from AMD, if my numbers bear out....

i hope P10 is sub £200 and its certainly not for me, i don't think i can wait till October/November.

It might be, but with that size of die it will be very cheap for that performance. pushing the performance per dollar up as they said. So expect some laptops with really decent performance using P11 parts.

Plus you were talking about P11, not P10.

Plus Polaris parts aren't meant as replacements for the top end of AMD's lineup.

Killer7xx · 13 May 2016 at 17:26

AllBodies said:
I'm extremely sceptical about these figures.

Firstly, as others mentioned, it makes no sense for a 232mm2 14nm chip to only have 2048 cores, unless the cores are much larger than previous GCN.

But also the R7 260X is only 160mm2, so that should make Polaris 11 less than 80mm2, so again that doesn't add up.

At the bottom of the article on videocardz, it says:

Our comment: We received the exact same specifications just few days ago. Although it was noted that those are specifications for mobile GPUs, so you may want to take this into consideration.

So those specs may just be for the Polaris mobile GPUs.

humbug · 13 May 2016 at 17:32

Killer7xx said:
At the bottom of the article on videocardz, it says:

So those specs may just be for the Polaris mobile GPUs.

Ah.... that makes a lot more sense.

Mauller · 13 May 2016 at 17:50

Killer7xx said:
At the bottom of the article on videocardz, it says:

So those specs may just be for the Polaris mobile GPUs.

Makes sense, it might mean that we have three polaris 10 parts.

2560, ~2300 and ~2000 core P10, the higher two being desktop parts.

Then there was stated as being 3 variants of Polaris 11 in the making, so this could be the lowest of the three.

humbug · 13 May 2016 at 17:58

Mauller said:
Makes sense, it might mean that we have three polaris 10 parts.

2560, ~2300 and ~2000 core P10, the higher two being desktop parts.

Then there was stated as being 3 variants of Polaris 11 in the making, so this could be the lowest of the three.

Not just higher Shader counts, the chances are also higher Mhz, possibly 1600Mhz

Mauller · 13 May 2016 at 18:04

humbug said:
Not just higher Shader counts, the chances are also higher Mhz, possibly 1600Mhz

Aye, it is a good strategy from the start to be honest, means they can salvage a far higher number of parts.

Lokken86 · 13 May 2016 at 18:08

If it can be overclocked to 1600mhz they could have had it like that out the box. Maybe they don't want to pi** on the Fury/nano bonfire just yet.