• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Polaris architecture – GCN 4.0

I can see where the cautious optimism comes from, and I share it.

Thing is, in theory AMD could just stick the 390X design on 14nm and do a 3072 core version running 1300+ MHz in the same size as Polaris 10. Something like that, requiring no arch improvements would match/beat a stock 980 Ti.

So including improvements, I expect the Polaris 10XT should be able to match the 1070.


EDIT: Also who's already forgetting all the hurr durr there's no way a ~320mm2 GTX 1080 could beat the 980Ti, it's just too powerful, etc.?

For the number of shaders it has, the 1080 is quite big, makes me wonder what is taking up all that extra room. Unless it is to do with them doubling the amount of cache since they halved the SM size.

Also the 2560 Ellesmere XT should beat a 390X at similar clocks, by how much i couldn't say. But consider a 60% uplift in clocks it should give it performance around or above an overclocked 980ti.

Especially with the improvements GCN4 has for DX11 based titles, it should easily beat a 390x at similar clocks.
 
Its seems stock of the 300 series and fury lineup is about to disappear here in Denmark.. also prices have been dropped a little bit. Wonder if this means 400 series replacements are about to arrive
 
Would mean it would have to almost match the GTX 980 ti at stock clocks, which I somehow highly doubt it will be able to do. Don't think they want to maximize performance with the Polaris lineup and instead go for a more balanced approach, so I'm guessing 390x performance with much lower power consumption.


390X is 438mm^2
P10 is 232mm^2

Samsung 14nm LPP is 2x min and up to 2,3x density of 28nm LPP

Best and worst case
P10 @ 2x density = 464mm^2 (+7%) 2816 Shaders + 7% = 3013 Shaders
P10 @ 2.3x density = 533mm^2 (+25%) 2816 Shaders + 25% = 3520 Shaders

Even by worst case P10 at the same clock rate is faster than a 390X. even without any architectural improvements.

The truth about how much GPU they can pack in 14nm LPP is probably somewhere in the middle, 2.15x.

From what we do know about P10 conservatively it should match a 980TI.
 
Last edited:
For £200, i can't see it.

Who says the P10XT is £200?

When they said they wanted to bring down the entry level VR price, I always assumed they were talking about the P10pro.

It still makes a lot of sense if P10XT is 1070 performance for £300, and P10pro is 390X performance for £200.
 
Who says the P10XT is £200?

When they said they wanted to bring down the entry level VR price, I always assumed they were talking about the P10pro.

It still makes a lot of sense if P10XT is 1070 performance for £300, and P10pro is 390X performance for £200.

+1

P10-XT @ 980TI/Fury-X> performance for £300
P10-Pro @ 980/390X> performance for £200

Is my prediction.
 
390X is 438mm^2
P10 is 232mm^2

Samsung 14nm LPP is 2x min and up to 2,3x density of 28nm LPP

Best and worst case
P10 @ 2x density = 464mm^2 (+7%) 2816 Shaders + 7% = 3013 Shaders
P10 @ 2.3x density = 533mm^2 (+25%) 2816 Shaders + 25% = 3520 Shaders

Even by worst case P10 at the same clock rate is faster than a 390X. even without any architectural improvements.

The truth about how much GPU they can pack in 14nm LPP is probably somewhere in the middle, 2.15x.

From what we do know about P10 conservatively it should match a 980TI.

Exactly this.

With no clockspeed improvements and no arch improvements, P10XT should already mildly beat a 390X.

Considering both clock&arch improvements are definite on top of that, people are being too pessimistic.

And like I mentioned earlier, remember we all went through this with the 1070/1080 couldn't possibly be as fast as they've turned out to be.
 
+1

P10-XT @ 980TI/Fury-X> performance for £300
P10-Pro @ 980/390X> performance for £200

Is my prediction.

Same what i calculated. I just added the difference between the two in TDP.
P10 rumored TDP is 175W, but they said the Pro will only consume 110-130W. So plenty of room to raise clocks too on the XT
 
Originally Posted by Mauller
Pretty much. We know as much about the 1080 as we do about Polaris..


True though. We have no third party benchmarks, everything is just PR spin atm and information on features. Even the doom vulkan demo is not a great thing to judge right now as we have no idea what the FPS is on other hardware.
 
For the number of shaders it has, the 1080 is quite big, makes me wonder what is taking up all that extra room. Unless it is to do with them doubling the amount of cache since they halved the SM size.

Also the 2560 Ellesmere XT should beat a 390X at similar clocks, by how much i couldn't say. But consider a 60% uplift in clocks it should give it performance around or above an overclocked 980ti.

Especially with the improvements GCN4 has for DX11 based titles, it should easily beat a 390x at similar clocks.

you need to understand that maxwell is a special architecture, where Nvidia striped the chip with a lot of hardware related to compute and scheduling they deemed not vital for gaming graphics, thats the main reason nvidia was able to have that efficiency and performance, but a lot of new things changed, VR, async, APIs, Nvidia just couldn't keep the same architecture as maxwell going forward, and puting back the hardware on pascal, diminished their efficiency and performance jump when clock to clock against maxwell.
well thats my guess.
 
390X is 438mm^2
P10 is 232mm^2

Samsung 14nm LPP is 2x min and up to 2,3x density of 28nm LPP

Best and worst case
P10 @ 2x density = 464mm^2 (+7%) 2816 Shaders + 7% = 3013 Shaders
P10 @ 2.3x density = 533mm^2 (+25%) 2816 Shaders + 25% = 3520 Shaders

Even by worst case P10 at the same clock rate is faster than a 390X. even without any architectural improvements.

The truth about how much GPU they can pack in 14nm LPP is probably somewhere in the middle, 2.15x.

From what we do know about P10 conservatively it should match a 980TI.

Had come to the same conclussion myself, roughly. I so hope this is true. 980ti performance for a bit less than the 1070 MSRB. It would be an instant buy for me.
 
you need to understand that maxwell is a special architecture, where Nvidia striped the chip with a lot of hardware related to compute and scheduling they deemed not vital for gaming graphics, thats the main reason nvidia was able to have that efficiency and performance, but a lot of new things changed, VR, async, APIs, Nvidia just couldn't keep the same architecture as maxwell going forward, and puting back the hardware on pascal, diminished their efficiency and performance jump when clock to clock against maxwell.
well thats my guess.

AMD did the same with Fiji, they stripped out the 1:2 FP64 compute of Hawaii down to 1:16.
 
Originally Posted by Mauller
Pretty much. We know as much about the 1080 as we do about Polaris..



True though. We have no third party benchmarks, everything is just PR spin atm and information on features. Even the doom vulkan demo is not a great thing to judge right now as we have no idea what the FPS is on other hardware.

We have the full specs and Nvidia publicly stating the cards relative gaming performance against a 980 and 980 SLI.

I no everyone likes to say "HURR DURR NVIDIA LIES", but the real performance is hardly going to be majorly different to what they promised or it would be a PR disaster for them.

Not to mention, the specs pretty much correlate with the performance they have stated.

What do we have that is absolutely confirmed about Polaris (and by that I mean confirmed press releases from AMD themselves)?

To say we know as much about the 1080 as we do about Polaris is utter bobbins.
 
We have the full specs and Nvidia publicly stating the cards relative gaming performance against a 980 and 980 SLI.

Oh no everyone likes to say "HURR DURR NVIDIA LIES", but the real performance is hardly going to be majorly different to what they promised or it would be a PR disaster for them.

Not to mention, the specs correlate with the performance they have stated.

No, what Nvidia stated was the relative gaming performance in VR scenarios.

You can even go back and check the vertical axis naming yourself on their relative performance chart. And the only reason for that is due to using single pass rendering and Multi projection.

Under normal gaming situations the performance won't be anywhere near as high as they have stated.

A recent video that was pulled from youtube had the 1080 ahead of a stock titan x by up to 17% in the division.
 
No, what Nvidia stated was the relative gaming performance in VR scenarios.

I think you need to watch and read the Nvidia press conference and material again.

A recent video that was pulled from youtube had the 1080 ahead of a stock titan x by up to 17% in the division

A recent video on youtube had a an overclocked 1080 at double the frames of a stock 980Ti. So what?
 
Last edited:
Back
Top Bottom