AMD Polaris architecture – GCN 4.0

Gregster · 2 Jun 2016 at 18:16

AlamoX said:
you ran the bench at stock clock ? or OC ?

Stock.

humbug · 2 Jun 2016 at 18:17

Jono8 said:
Gregster has already done it with the settings used in the slide.

Where is his link?

bru · 2 Jun 2016 at 18:19

Well I for one am so pleased that the absolutely non confusing benchmark they showed at the announcement, which has been dissected by loads of forum users, then analysed by reviewers, clarified by AMD on redit and linked too by our very own Matt, has given us a nice clear and concise picture of the performance of the RX480.

On a separate note, personally I reckon the RX480 is the X variety and the lesser counterpart will just be called the R480. (they have just moved the position of the X ).

humbug · 2 Jun 2016 at 18:25

bru said:
Well I for one am so pleased that the absolutely non confusing benchmark they showed at the announcement, which has been dissected by loads of forum users, then analysed by reviewers, clarified by AMD on redit and linked too by our very own Matt, has given us a nice clear and concise picture of the performance of the RX480.

On a separate note, personally I reckon the RX480 is the X variety and the lesser counterpart will just be called the R480. (they have just moved the position of the X ).

Over and over again AMD do this, i do wonder if the have an Nvidia guy advising them on marketing, you couldn't make it up, they constantly do things that are complicated, confusing and bring up a ton of questions with no answers which they later have to explain and defend unsuccessfully.

Its a GPU, then you have another GPU, how does one compare them to eachother straight up? its not hard.
Why all of this enigma crap?

AlamoX · 2 Jun 2016 at 18:25

bru said:
On a separate note, personally I reckon the RX480 is the X variety and the lesser counterpart will just be called the R480. (they have just moved the position of the X ).

i think they rid of the X variant altogether, the cutdown version will probably be RX470 instead of R480, polaris 11 get RX250/RX260

Seanspeed · 2 Jun 2016 at 18:25

bru said:
Well I for one am so pleased that the absolutely non confusing benchmark they showed at the announcement, which has been dissected by loads of forum users, then analysed by reviewers, clarified by AMD on redit and linked too by our very own Matt, has given us a nice clear and concise picture of the performance of the RX480.

On a separate note, personally I reckon the RX480 is the X variety and the lesser counterpart will just be called the R480. (they have just moved the position of the X ).

That would be super confusing.

Gregster · 2 Jun 2016 at 18:37

humbug said:
Where is his link?

I am at work and did tell you I ran the bench at the same settings as that 480

humbug · 2 Jun 2016 at 18:37

Gregster said:
Look back through my posting today and you will see I ran the test at the same settings as used by the 480. It ended up as 37% slower than my Titan X.

Ok so lets take that, @ 40 FPS your stock TX was +37% so = 55 FPS.

So your stock TX was 90% the performance of the 1080? what is it overclocked? 65/70 FPS?

That doesn't bode well for the 1080.

fs123 · 2 Jun 2016 at 18:40

The demo was 1080P running at Crazy, 8xAA as confirmed by Robert Hallock.
https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/

Hallock also says:

As a parting note, I will mention we ran this test 10x prior to going on-stage to confirm the performance delta was accurate. Moving up to 1440p at the same settings maintains the same performance delta within +/-1%.

If the crossfire 480 demo was running at 151% scaling as claimed then a single card would be getting around 40 fps.
If it also can achieve around 38-40 fps @1440P Crazy, 8xAA then it puts it at around Fury Nano level :eek:

With the Doom demo running 1440P VSR @ around 70fps average I think we can safely say it's Fury Nano performance for less than £200.

h4rm0ny · 2 Jun 2016 at 18:42

fs123 said:
The demo was 1080P running at Crazy, 8xAA as confirmed by Robert Hallock.
https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/

Hallock also says:

If the crossfire 480 demo was running at 151% scaling as claimed then a single card would be getting around 40 fps.
If it also can achieve around 38-40 fps @1440P Crazy, 8xAA then it puts it at around Fury Nano level

With the Doom demo running 1440P VSR @ around 70fps average I think we can safely say it's Fury Nano performance for less than £200.

Or to put it in the simplest terms: AWESOME!

CAT-THE-FIFTH · 2 Jun 2016 at 18:44

humbug said:
Over and over again AMD do this, i do wonder if the have an Nvidia guy advising them on marketing, you couldn't make it up, they constantly do things that are complicated, confusing and bring up a ton of questions with no answers which they later have to explain and defend unsuccessfully.

Its a GPU, then you have another GPU, how does one compare them to eachother straight up? its not hard.
Why all of this enigma crap?

Maybe because AMD are constantly in fear that Nvidia might do something to screw up the launch?? Hence they are being vague about things so Nvidia does not know exactly what to expect.

After all we saw the GTX980TI kind of screw the Fury X launch.

Even the HD5870,HD4870 and 9700 PRO launches had less noise about performance than this one.

humbug · 2 Jun 2016 at 18:48

CAT-THE-FIFTH said:
Maybe because AMD are constantly in fear that Nvidia might do something to screw up the launch?? Hence they are being vague about things so Nvidia does not know exactly what to expect.

After all we saw the GTX980TI kind of screw the Fury X launch.

Even the HD5870,HD4870 and 9700 PRO launches had less noise about performance than this one.

Be that as it may they need to stop thinking like that.

Its a $200 card, if it sits in the middle between a 980 and 980TI or Fury-Nano / Fury- Pro performance then just make that clear as day.

You show an audience that and tell them its $200 they will be picking their jaws up off the floor.

Nothing to fear. what can Nvidia do about that? they have already set their price vs perf.. they would look the fools if they back peddle on it.

LoadsaMoney · 2 Jun 2016 at 18:48

CAT-THE-FIFTH said:
After all we saw the GTX980TI kind of screw the Fury X launch.

The only thing they screwed up, was getting the extra money from their TX sales.

D.P. · 2 Jun 2016 at 18:58

fs123 said:
The demo was 1080P running at Crazy, 8xAA as confirmed by Robert Hallock.
https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/

Hallock also says:

If the crossfire 480 demo was running at 151% scaling as claimed then a single card would be getting around 40 fps.
If it also can achieve around 38-40 fps @1440P Crazy, 8xAA then it puts it at around Fury Nano level

With the Doom demo running 1440P VSR @ around 70fps average I think we can safely say it's Fury Nano performance for less than £200.

The scaling was confirmed by AMD to be 1.83x.
So a single 480 gets 34FPS.

https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/d3sw31g

humbug · 2 Jun 2016 at 19:02

D.P. said:
The scaling was confirmed by AMD to be 1.83x.
So a single 480 gets 34FPS.

Do you have a link to that?

1.83x is not 51% utilisation, as AMD said, its 83%!

I would like to confirm exactly what he said because one is very different to the other.

Sacrifice · 2 Jun 2016 at 19:04

humbug said:
Do you have a link to that?

1.83x is not 51% utilisation, as AMD said, its 83%!

I would like to confirm exactly what he said because one is very different to the other.

To clarify this, the scaling from 1->2 GPUs in the dual RX 480 test we assembled is 1.83x. The OP was looking only at the lowest draw call rates when asking about the 51%. The single batch GPU utilization is 51% (CPU-bound), medium is 71.9% utilization (less CPU-bound) and heavy batch utilization is 92.3% (not CPU-bound). All together for the entire test, there is 1.83X the performance of a single GPU in what users saw on YouTube. The mGPU subsystem of AOTS is very robust.

That's what he said: https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/

CAT-THE-FIFTH · 2 Jun 2016 at 19:04

BTW,is AOTS SFR??

IIRC,it does not use AFR.

bru · 2 Jun 2016 at 19:04

humbug said:
Its a $200 card, if it sits in the middle between a 980 and 980TI or Fury-Nano / Fury- Pro performance then just make that clear as day.

You show an audience that and tell them its $200 they will be picking their jaws up off the floor.

+1 they should have shown us just how it is, rather than showing us something that has to be clarified and analysed just to get to the result.

D.P. · 2 Jun 2016 at 19:06

humbug said:
Do you have a link to that?

1.83x is not 51% utilisation, as AMD said, its 83%!

I would like to confirm exactly what he said because one is very different to the other.

https://www.reddit.com/r/Amd/comments/4m692q/concerning_the_aots_image_quality_controversy/d3sw31g

There is no 51% utilization. The 51% means that 51% of the time the GPU was not CPU limited, the utilization is totally unknown. AMD just don't know what the benchmark is even telling them.

humbug · 2 Jun 2016 at 19:07

humbug said:
Do you have a link to that?

1.83x is not 51% utilisation, as AMD said, its 83%!

I would like to confirm exactly what he said because one is very different to the other.

AMD_RobertTechnical Marketing 153 points 8 hours ago*
Scaling is 151% of a single card.
//EDIT: To clarify this, the scaling from 1->2 GPUs in the dual RX 480 test we assembled is 1.83x. The OP was looking only at the lowest draw call rates when asking about the 51%. The single batch GPU utilization is 51% (CPU-bound), medium is 71.9% utilization (less CPU-bound) and heavy batch utilization is 92.3% (not CPU-bound). All together for the entire test, there is 1.83X the performance of a single GPU in what users saw on YouTube. The mGPU subsystem of AOTS is very robust.

That explains a lot, the "51% GPU utilisation" cited at Computex was wrong
It is 83%, 17 percentage points short of full scaling.

So cut the FPS in half 62.6 x0.5 = 31.5 and add the 17% = 36.5 FPS.