AMD Polaris architecture – GCN 4.0

[Freddie1980]

Resurrecting the Rage branding would be good.

I'm not sure about that. The old ATI's reputation for having bad drivers is largely down to the Rage line of video cards the gag at the time was 'ATI drivers give you Rage!', this is best confined to history especially since AMD have been on a roll in terms of driver support recently.

 

[Rroff]

1070 won't be under £300 anytime soon, the founders edition is likely to be £400 or so (straight USD to GBP conversion gives ~£375), AIB cards *may* dip closer to £310-320 or so, maybe.

I'm also kinda unsure why everyone keeps saying 980/390X performance, given the rumours so far seem to be a 2560 core for the full '480X' part, clocked around ~1.25-1.3GHz (if not higher based on some bits).

Compared to older cards the 390 (non-X) is 2560 cores, so take that as a base, add 25-30% clock speed, and 2 'generations' improvement in core design (GCN 1.1 to GCN '4'). The 390X is 5-10% faster it seems, Fury ~20% faster, Fury X ~25%, 980Ti ~30%, 1070 say 35-40%?

Even by die size if you take into account the disabled parts of the 1070's die (roughly 75% essentially) then it's 'effective' die size is similar to Polaris 10's rumoured 232mm^2.

Unless I'm missing something it'll either be comfortably ahead of the 390X (and hence into Fury/980ti/1070 territory) or AMD have managed to take a big step backwards this generation...

So far none of the leaks affirm they will have something that can compete with the 1070 - its something AMD need right now though - a card that is within +/-5% of the 1070 but can edge it decently on price (rather than compete necessarily on all out performance but on the ability to turn everything up 60fps gaming as the developer intended it to look kind of thing).

I'm not sure about that. The old ATI's reputation for having bad drivers is largely down to the Rage line of video cards the gag at the time was 'ATI drivers give you Rage!', this is best confined to history especially since AMD have been on a roll in terms of driver support recently.

True but a lot of people have a soft spot for the Rage branding despite that - probably due to the promise of what it would bring more than the reality maybe. One of my first video cards was the ATI 3D Rage IIc.

 

[CAT-THE-FIFTH]

I think AMD should worry less about the name of the card,and try and get a perfect launch with realistic pricing. The last three have been subpar launches:

1.)R9 290/290X

Stock cooler was noisy and on lower fan speeds would throttle. Lack of third party cards for a month or two tarnished the image of the cards as being hot and noisy.

2.)R9 285

Priced too highly and AMD stupidly sent out a few pre-overclocked models which had WORSE power consumption than the R9 280 models,making AMD look like they were going a step backwards with performance per watt when compared to Nvidia.

3.)R9 Fury X and Fury Nano

Unrealistic pricing for the performance and cooler QC problems.

Whatever the performance Polaris 10 and 11 have,they need to:
1.)Price them at an acceptable level for their performance
2.)Take into consideration what performance the competing cards BOTH old and new Nvidia will have
3.)Make sure the models sent for review have decent coolers which have passed QC
4.)Don't send bugged aftermarket models with poor power consumption for review
5.)Make sure they have some decent review drivers

 

[Kaapstad]

I think AMD should worry less about the name of the card,and try and get a perfect launch with realistic pricing. The last three have been subpar launches:

1.)R9 290/290X

Stock cooler was noisy and on lower fan speeds would throttle. Lack of third party cards for a month or two tarnished the image of the cards as being hot and noisy.

2.)R9 285

Priced too highly and AMD stupidly sent out a few pre-overclocked models which had WORSE power consumption than the R9 280 models,making AMD look like they were going a step backwards with performance per watt when compared to Nvidia.

3.)R9 Fury X and Fury Nano

Unrealistic pricing for the performance and cooler QC problems.

Whatever the performance Polaris 10 and 11 have,they need to:
1.)Price them at an acceptable level for their performance
2.)Take into consideration what performance the competing cards BOTH old and new Nvidia will have
3.)Make sure the models sent for review have decent coolers which have passed QC
4.)Don't send bugged aftermarket models with poor power consumption for review
5.)Make sure they have some decent review drivers

6.)Make sure they get their sales pitch right, don't make claims that will come back to haunt them (overclockers dream).

Sadly NVidia made the above mistake with the 1080 running at 67c.

 

[TheRealDeal]

I think AMD should worry less about the name of the card,and try and get a perfect launch with realistic pricing. The last three have been subpar launches:

1.)R9 290/290X

Stock cooler was noisy and on lower fan speeds would throttle. Lack of third party cards for a month or two tarnished the image of the cards as being hot and noisy.

2.)R9 285

Priced too highly and AMD stupidly sent out a few pre-overclocked models which had WORSE power consumption than the R9 280 models,making AMD look like they were going a step backwards with performance per watt when compared to Nvidia.

3.)R9 Fury X and Fury Nano

Unrealistic pricing for the performance and cooler QC problems.

Whatever the performance Polaris 10 and 11 have,they need to:
1.)Price them at an acceptable level for their performance
2.)Take into consideration what performance the competing cards BOTH old and new Nvidia will have
3.)Make sure the models sent for review have decent coolers which have passed QC
4.)Don't send bugged aftermarket models with poor power consumption for review
5.)Make sure they have some decent review drivers

Hopefully the full crimson suite will be finished in time as another plus. A lot of people think Nvidia products are more polished due to Software. With the full Crimson suite ready along with some other goodies that Polaris has this could change perception a bit.

 

[CAT-THE-FIFTH]

6.)Make sure they get their sales pitch right, don't make claims that will come back to haunt them (overclockers dream).

Sadly NVidia made the above mistake with the 1080 running at 67c.

That too. RTG should look back at the HD4000 and HD5000 launches which pretty much were quite positive.

Hopefully the full crimson suite will be finished in time as another plus. A lot of people think Nvidia products are more polished due to Software. With the full Crimson suite ready along with some other goodies that Polaris has this could change perception a bit.

I suppose that too,although I tend not to look at the driver stuff unless there is a problem!!

 

[humbug]

1070 won't be under £300 anytime soon, the founders edition is likely to be £400 or so (straight USD to GBP conversion gives ~£375), AIB cards *may* dip closer to £310-320 or so, maybe.

I'm also kinda unsure why everyone keeps saying 980/390X performance, given the rumours so far seem to be a 2560 core for the full '480X' part, clocked around ~1.25-1.3GHz (if not higher based on some bits).

Compared to older cards the 390 (non-X) is 2560 cores, so take that as a base, add 25-30% clock speed, and 2 'generations' improvement in core design (GCN 1.1 to GCN '4'). The 390X is 5-10% faster it seems, Fury ~20% faster, Fury X ~25%, 980Ti ~30%, 1070 say 35-40%?

Even by die size if you take into account the disabled parts of the 1070's die (roughly 75% essentially) then it's 'effective' die size is similar to Polaris 10's rumoured 232mm^2.

Unless I'm missing something it'll either be comfortably ahead of the 390X (and hence into Fury/980ti/1070 territory) or AMD have managed to take a big step backwards this generation...

They know all this but ignore it, worse actually.

These people know very well the 2048 Shader parts are laptop parts and the prediction is full size parts are 2560 Shaders and yet they continue on as if its already a fact that P10 has a maximum 2048 Shaders and a maximum 390X performance.

Its the same people who hype up Pascal constantly. its like they are on some sort of roster taking turns downplaying Polaris and then up playing Pascal at regular turn intervals to make certain their message has no chance of getting lost on any reader.
You go back through this thread the same nonsense has been repeated at regular intervals over and over and over again.

Perhaps they are trying to steer people into pre-ordering £350+ 1070's

 

[Nutella33]

6.)Make sure they get their sales pitch right, don't make claims that will come back to haunt them (overclockers dream).

Sadly NVidia made the above mistake with the 1080 running at 67c.

Was going to say, No 1 in the list sounds awfully familiar

 

[GFX-Kid]

They know all this but ignore it, worse actually.

These people know very well the 2048 Shader parts are laptop parts and the prediction is full size parts are 2560 Shaders and yet they continue on as if its already a fact that P10 has a maximum 2048 Shaders and a maximum 390X performance.

Its the same people who hype up Pascal constantly. its like they are on some sort of roster taking turns downplaying Polaris and then up playing Pascal at regular turn intervals to make certain their message has no chance of getting lost on any reader.
You go back through this thread the same nonsense has been repeated at regular intervals over and over and over again.

Perhaps they are trying to steer people into pre-ordering £350+ 1070's

Yep, I have noticed this also.

 

[Seanspeed]

These people know very well the 2048 Shader parts are laptop parts and the prediction is full size parts are 2560 Shaders

There is no 'the prediction', as if there's only one grand predictor that we all bow down to.

And no, the people saying this are not ALL the same people hyping up Pascal. Some of us just aren't too eager to get hyped over predictions that may not turn out true at all. It's called being cautious.

And if you want to go down that route of what people are saying what, there's some consistencies I could also point out here...

 

[D.P.]

AMD claim a mainstream part not a performance part.
The leaked benchmarks shows performance at the expect 390X level and those leaks were very accurate for the 1080.
There have been a load of other leaks and rumours that are consistent wit that performance level.

Do one now anything for sure but I would go hyping up some kind of 1080 killer given the current rumors and press releases, it will only lead to disappointing and more accusations flown at AMD. Polaris just needs to be the right price for the performance and have decent efficiency to be a winner.


\

 

[SKILL]

Define mainstream vs performance and the 1070/1080 is also the 'mid-range' part, although the performance is very good it's not the successor to the 980ti, it's the successor to the 970/980.

I don't think anyone really expects it to beat the 1080, I'm pretty sure nvidia have the very top end all wrapped up until the 'big die' parts come out end of the year/early 2017.

Just that given the rumours of the 2560 cores/1.25-1.3GHz part, and the rumoured performance of the 1070, they should be within 5-10% of each other or AMD have massively dropped the ball, ignoring price for the moment just in terms of technology and performance it would be a big step back from what they had years ago.

EDIT: Of course if yields are so bad that we only get a 2316/2048 core as the best Polaris then that would make more sense (in terms of the 'no faster than 390X' bit)

 

[Overlag]

and the 1070/1080 is also the 'mid-range' part

I stopped reading there. £650 for a mid range part? LOL.

 

[Mauller]

AMD claim a mainstream part not a performance part.
The leaked benchmarks shows performance at the expect 390X level and those leaks were very accurate for the 1080.
There have been a load of other leaks and rumours that are consistent wit that performance level.

Do one now anything for sure but I would go hyping up some kind of 1080 killer given the current rumors and press releases, it will only lead to disappointing and more accusations flown at AMD. Polaris just needs to be the right price for the performance and have decent efficiency to be a winner.


\

The latest leaks were showing one at fury (nonx) performance at 1.2ghz for the 2316 shader part. As long as the clocks scale well then this part will reach furyx - 980ti performance let alone a full 2560 sp part if there is one in the end.

 

[Sargatanas2511]

I stopped reading there. £650 for a mid range part? LOL.

It technically is their high mid-range card. the 1080ti and Titan will be based on a different chip. Price doesn't suddenly make it high end.

 

[SKILL]

I stopped reading there. £650 for a mid range part? LOL.

It's the direct successor to the 970/980, which most would class as a 'mid-range' part, just because it's priced highly doesn't change that. I'm not saying the price is wrong/bad, but it's priced at that level due to both the good performance and lack of competition, not because it's a high-end/big die. We already know that there will be a 1080ti/Titan variant which will be the 'big Pascal' chip, that's the high end, not the 1080.

So yeah, LOL

 

[D.P.]

The latest leaks were showing one at fury (nonx) performance at 1.2ghz for the 2316 shader part. As long as the clocks scale well then this part will reach furyx - 980ti performance let alone a full 2560 sp part if there is one in the end.

That is some pretty big caveats there.

 

[humbug]

AMD claim a mainstream part not a performance part.
The leaked benchmarks shows performance at the expect 390X level and those leaks were very accurate for the 1080.
There have been a load of other leaks and rumours that are consistent wit that performance level.

Do one now anything for sure but I would go hyping up some kind of 1080 killer given the current rumors and press releases, it will only lead to disappointing and more accusations flown at AMD. Polaris just needs to be the right price for the performance and have decent efficiency to be a winner.


\

In this thread some pages back those same people who predicted the 1080 also predicted the 2048 Shader part with performance around a 390/X, leaked benchmarks put it there, they predicted a clock speed of 1375Mhz and made it clear these are discrete Laptop parts.

From the size of the chip its predicted the full fat desktop chips would have 2560 Shaders (25% more) and be clocked at around 1600Mhz (+15%)
Thats a total of 40% more GPU, that would put them right in the 980TI level, at least.

 

[D.P.]

Things never scale that linearly though.

We will see when they are reviewed, I just think it makes sense to be realistic, especially given what AMD themselves have mentioned.

 

[Phixsator]

Things never scale that linearly though.

We will see when they are reviewed, I just think it makes sense to be realistic, especially given what AMD themselves have mentioned.

What makes me wonder about those 480x leaks is the crossfire scaling in 3dmark. The lowest result i could find gave 60% scaling for the fury(picked a card close to the polaris card) but the CF setup in that slide as a good deal less than that which makes me believe either drivers are it holding back, slides are fake, some sort of throttling issue in CF due to heat or perhaps something else i cant think of right now. With at least 60% scaling it should have come out at the very top just edging out the 1080..