AMD Polaris architecture – GCN 4.0

[humbug]

A 232mm would have the equivalent Stream Processors of around 3,400, this vs 2,816 on the 390X, so like for like we are looking at a Fury

Having said that there is more to it, AMD are going to be using the same Tessellation reduction technology as Nvidia, and they are also updating the front end to make them more efficient in CPU heavy situations.

How fast they will end up is all just speculation, but even in DX11 they do have the potential to be as fast as a 980TI.

It all depends on price, Nvidia should have a GTX 1080 to match it, but they may want £400 to £500 for it.
If AMD can do it for £300 it may force Nvidia to climb down from their huge margins.
But then again i think there are too many willing to pay whatever it takes to remain an Nvidia club member, and Nvidia know it.

 

[Mauller]

The improvements in GCN4 should allow Polaris parts to reach their theoretical max in single threaded situations and more so in multi threaded ones.

Just look at the 390X, it has a theoretical performance a few hundered gigaflops greater than the 980TI at stock. This only really shows up when a moderate amount of Async is used. And then we even have the 390X catching up to a 980TI with a 38% overclock.

It will be very interesting to see how much they have improved GCN to enable it to use all of its available power. Before any of the extras that help reduce the processing burden and any advantages in Node change.

 

[humbug]

The improvements in GCN4 should allow Polaris parts to reach their theoretical max in single threaded situations and more so in multi threaded ones.

Just look at the 390X, it has a theoretical performance a few hundered gigaflops greater than the 980TI at stock. This only really shows up when a moderate amount of Async is used. And then we even have the 390X catching up to a 980TI with a 38% overclock.

It will be very interesting to see how much they have improved GCN to enable it to use all of its available power. Before any of the extras that help reduce the processing burden and any advantages in Node change.

AMD GPU's have a tenancy to mature.

little over a year ago Hawaii XT was struggling to keep up with a GTX 970, now its a match for the GTX 980, in some cases catching up with the GTX 980TI.

Two years ago Hawaii XT was a GPU considered nearly as fast as a GTX 780TI, a cheaper alternative.
Now the 780TI can't get anywhere near it.

Since introduction Hawaii XT performance looks like its up at least 20%.

Strange....

 

[bru]

It is just as likely that AMD will push the price up with their new 'We are not a budget brand attitude', but I expect that both sides second tier cards will be around 980ti/FuryX in performance and around the £300 bracket. With the top tier card being a bit faster, this is of course till very late in the year when the true big chips from both will probably hit, even though we may have to wait till early next year for them.

 

[Overlag]

To be fair we only really got those mega big chips because we were stuck on 28nm for so long. Whos to say we might never get one that big again?

 

[Mauller]

To be fair we only really got those mega big chips because we were stuck on 28nm for so long. Whos to say we might never get one that big again?

We will, The shrinking of transistors is becoming harder. They might be able to make test SoC's on 10nm atm but it will be harder to transition large monolithic dies. we will more than likely have a good 2 - 3 years on 14nm. Maybe with some lower end parts working into 10nm during that time.

Things get far harder to do pure silicon transistors at 10nm.

 

[r7slayer]

AMD GPU's have a tenancy to mature.

little over a year ago Hawaii XT was struggling to keep up with a GTX 970, now its a match for the GTX 980, in some cases catching up with the GTX 980TI.

Two years ago Hawaii XT was a GPU considered nearly as fast as a GTX 780TI, a cheaper alternative.
Now the 780TI can't get anywhere near it.

Since introduction Hawaii XT performance looks like its up at least 20%.

Strange....

yep i said this long ago, around the time when titan x came out and got my head took off by certain nVidia peeps. 290x was slower than a 780Ti now its faster. The fury x could end up having a similar fate or more so with Dx12. Who knows.

 

[Orangey]

we will more than likely have a good 2 - 3 years on 14nm.

That's normal.

28nm lasted 5 years which is almost 3 times the normal cycle.

 

[Jono8]

AMD GPU's have a tenancy to mature.

little over a year ago Hawaii XT was struggling to keep up with a GTX 970, now its a match for the GTX 980, in some cases catching up with the GTX 980TI.

Two years ago Hawaii XT was a GPU considered nearly as fast as a GTX 780TI, a cheaper alternative.
Now the 780TI can't get anywhere near it.

Since introduction Hawaii XT performance looks like its up at least 20%.

Strange....

Not really, at least not to the extent you have made out.

There are ALWAYS certain games where one brand/card shines over the other.

By your logic, I could say the 970 was just as fast /snapping at the heels of the Fury / Fury X if I looked at these games:

https://www.guru3d.com/articles_pages/fallout_4_pc_graphics_performance_benchmark_review,7.html

https://www.guru3d.com/articles_pag..._graphics_performance_benchmark_review,7.html

https://www.guru3d.com/articles_pages/anno_2205_pc_graphics_performance_benchmark_review,7.html

Don't get me wrong. I am not for one minute saying the 970 is as fast as the Fury X if you looked at the whole picture, but just because the 390/390x comes a bit close to a 980Ti in the just released Hitman DX12 bench, it doesn't mean that Hawaii XT has matured THAT much.

 

[humbug]

Not really, at least not to the extent you have made out.

There are ALWAYS certain games where one brand/card shines over the other.

By your logic, I could say the 970 was just as fast /snapping at the heels of the Fury / Fury X if I looked at these games:

https://www.guru3d.com/articles_pages/fallout_4_pc_graphics_performance_benchmark_review,7.html

https://www.guru3d.com/articles_pag..._graphics_performance_benchmark_review,7.html

https://www.guru3d.com/articles_pages/anno_2205_pc_graphics_performance_benchmark_review,7.html

Don't get me wrong. I am not for one minute saying the 970 is as fast as the Fury X if you looked at the whole picture, but just because the 390/390x comes a bit close to a 980Ti in the just released Hitman DX12 bench, it doesn't mean that Hawaii XT has matured THAT much.

Yet your not looking at the whole picture, you cherry picked 3 games.

This is the whole picture....

https://www.techpowerup.com/reviews/ASUS/R9_380X_Strix/23.html

 

[bru]

Yet your not looking at the whole picture, you cherry picked 3 games.

This is the whole picture....

https://www.techpowerup.com/reviews/ASUS/R9_380X_Strix/23.html

Well to be fair a good as TPU reviews are for a nice broad selection of games it is not exactly the whole picture either. The whole picture as far as game reviews and benchmarks is not a practical thing to show as there are just too many games and benches out there.

 

[humbug]

Well to be fair a good as TPU reviews are for a nice broad selection of games it is not exactly the whole picture either. The whole picture as far as game reviews and benchmarks is not a practical thing to show as there are just too many games and benches out there.

What then is the whole picture?

A hundred games? A thousand games? Ten thousand games?

As you said TPU have a broad selection, we can be endlessly anal about what constitutes the whole picture, the truth is 'to be really anal' there is no such thing.
But we can settle on something reasonable.

 

[tommybhoy]

Got a better source you can recommend bru?

 

[bru]

Got a better source you can recommend bru?

Not really, TPU does have a good selection and the way they do their graphs is good in my opinion, but there will always be other benchmarks that could be included. It just doesn't seem fair to criticize someone for cherry picking benches and then say that this is the whole picture and provide another set of cherry picked benches, albeit by TPU this time.

 

[humbug]

Not really, TPU does have a good selection and the way they do their graphs is good in my opinion, but there will always be other benchmarks that could be included. It just doesn't seem fair to criticize someone for cherry picking benches and then say that this is the whole picture and provide another set of cherry picked benches, albeit by TPU this time.

Cherry picking is when you pick out a couple of reviews from a site and link them individually.

Not posting links to a round up of 15+ games...
There is a difference.

 

[tommybhoy]

Under the impression that 'cherry picking' is looking for results that benefit a gpu just like in the examples he used, where as TPU benches a broad section of set titles across the board giving an indication of all round performance.

Not saying he shouldn't have, but agree with bug as TPU is my first stop for a good indication of all round performance especially when a new round of gpu's hit usually skipping right to the Performance Summary to compare the performance gains and work my way back.

 

[bru]

Yup I have used them my self on many occasion, but still it is not right to say that is the whole picture, when they have only picked 15 or so benchmarks out of the countless ones available.

This point came up when humbug said that the Hawaii GPU's performance has matured a lot over its life, now it consistently beats the 970 even the 980 and sometimes gets close to the 980ti, all of which is quite true, but when Jono supplies some benchies to show that it could be said that a 970 gets close to a FuryX. that is not looking at the whole picture, fair enough, but by the same standard the TPU benchmarks even though they are a much broader selection of benches they are also not the whole picture either. That is my point.

If humbug had of said this is a fairer selection of benchies or something that that then it wouldn't of been an issue in the first place.

 

[humbug]

Yup I have used them my self on many occasion, but still it is not right to say that is the whole picture, when they have only picked 15 or so benchmarks out of the countless ones available.

This point came up when humbug said that the Hawaii GPU's performance has matured a lot over its life, now it consistently beats the 970 even the 980 and sometimes gets close to the 980ti, all of which is quite true, but when Jono supplies some benchies to show that it could be said that a 970 gets close to a FuryX. that is not looking at the whole picture, fair enough, but by the same standard the TPU benchmarks even though they are a much broader selection of benches they are also not the whole picture either. That is my point.

If humbug had of said this is a fairer selection of benchies or something that that then it wouldn't of been an issue in the first place.

Semantics.

 

[Jono8]

Yet your not looking at the whole picture, you cherry picked 3 games.

This is the whole picture....

https://www.techpowerup.com/reviews/ASUS/R9_380X_Strix/23.html

What!? I know! That was my whole point!

You rather misleadingly claimed it was starting to catch up to the 980ti in some cases just because of hitman's dx12 bench. I then made a point of cherry picking a few benchmarks where i could just as easily claim that the 970 is catching up with the Fury cards.

 

[Davedree]

What!? I know! That was my whole point!

You rather misleadingly claimed it was starting to catch up to the 980ti in some cases just because of hitman's dx12 bench. I then made a point of cherry picking a few benchmarks where i could just as easily claim that the 970 is catching up with the Fury cards.

Or you could even use that hitman bench and state how poorly fury performs in dx12 compared to hawaii considering the shader spec and hbm that fury has.