[Ars] AMD confirms high-end Polaris GPU will be released in 2016

[Glo.]

https://forum.beyond3d.com/threads/nvidia-pascal-speculation-thread.55552/page-39#post-1895652 Good post about FinFET affect on GPU dies.

 

[maddie]

It's just a description in a shipping manifest, so it could very well be the typo that they article author suggests. Who knows, other than the random guy in the shipping department trying to fill the forms out and get these shipped out to India before he gets stuck in traffic on the way home and misses the first period of the Leaf's game.

Though I'm sure Maddie will have his ears perk up and suggest that it's three small dies on an interposer.

Someone called?

Nah, not the small ones, the P/N 102-C99398-00 one is the multi-die on interposer.

Seriously now, someone on semiaccurate claimed C7xxxx was original GCN, C8xxxx was the Fiji gen and C9xxxx is Polaris. With the demoed Polaris 10 card being fairly powerful, they would need something for someone just needing mainly video output and low 3D. Thus, reused 28nm lowest end and sub Polaris 10 performance. The prices are high though.

 

[JDG1980]

2GB and 4GB GDDR5 cards spotted.
http://videocardz.com/58308/amds-baffin-weston-and-banks-gpus-spotted-on-zauba

Back when "Arctic Islands" was still the codename for AMD's 2016 lineup, the rumors held that the chips would be (from smallest to largest) Baffin, Ellesmere, and Greenland. I suspect that the "Baffin" board in the above list is what is now being called Polaris 10.

Westin and Banks are new names, but I suspect there's a strong chance these are rebranded chips. Note the much lower prices assigned to the test samples ($150-$185) compared to the $705 and change they claim the Baffin prototype card is worth. We're probably looking at Cape Verde and Oland rebrands for the OEM market. These won't see much retail presence (nor should they).

 

[maddie]

Back when "Arctic Islands" was still the codename for AMD's 2016 lineup, the rumors held that the chips would be (from smallest to largest) Baffin, Ellesmere, and Greenland. I suspect that the "Baffin" board in the above list is what is now being called Polaris 10.

Westin and Banks are new names, but I suspect there's a strong chance these are rebranded chips. Note the much lower prices assigned to the test samples ($150-$185) compared to the $705 and change they claim the Baffin prototype card is worth. We're probably looking at Cape Verde and Oland rebrands for the OEM market. These won't see much retail presence (nor should they).

This sounds like a reasonable interpretation.

 

[jpiniero]

That does make sense, I guess that means that the cut Polaris 10 is coming first too.

Would people really be pissed if Polaris 10 is 380/380X performance @ $199/$229 @ 75W? I mean that is 190 / 2.5...

 

[PPB]

That does make sense, I guess that means that the cut Polaris 10 is coming first too.

Would people really be pissed if Polaris 10 is 380/380X performance @ $199/$229 @ 75W? I mean that is 190 / 2.5...

More like $140-150 @ 60w for 270x performance, $180 @ 75w for in between 380 to 380x performance.

 

[DownTheSky]

That and polaris 11 Fury performance.

 

[PPB]

AMD really needs this generation, and specially considering they are initially only making 2 dies for the consumer market, to make broader salvaging schemes. 7950 within 5-7% of 7970 perf at same clocks, 290 within 5% of 290x perf at same clocks, and the same with Fury to Fury X only hurts them badly as a lot of perf/$ conscious buyers would go for the harvested sku. And this is without even adding the shading unlock thing that happened with Cayman and with Fiji. A perfect example of good die harvesting segmentation is 970 and 980 (albeit the technical flaws that resulted in 970's gimped VRAM/bandwith/caches), a lot more people went 970 because of the incredible perf/$ at launch, but the 17-20% initial difference between them made a lot of people shell the extra bucks (and the price difference was indeed big) for the 980.

 

[tential]

AMD really needs this generation, and specially considering they are initially only making 2 dies for the consumer market, to make broader salvaging schemes. 7950 within 5-7% of 7970 perf at same clocks, 290 within 5% of 290x perf at same clocks, and the same with Fury to Fury X only hurts them badly as a lot of perf/$ conscious buyers would go for the harvested sku. And this is without even adding the shading unlock thing that happened with Cayman and with Fiji. A perfect example of good die harvesting segmentation is 970 and 980 (albeit the technical flaws that resulted in 970's gimped VRAM/bandwith/caches), a lot more people went 970 because of the incredible perf/$ at launch, but the 17-20% initial difference between them made a lot of people shell the extra bucks (and the price difference was indeed big) for the 980.

Considering how well the HD7000 series held up and the R9 200 series held up, I'm not remotely worried about Polaris. To me, Fiji was simply a sideshow for Polaris. Nothing more than a "We can't release what we want to release, so here is just something in the mean time."

Even Fiji isn't bad except for Fury X. Nano at $450 now is a really good pick up, and so is Fury at $500 compared to the GTX 980.

The only impressive chip I've seen from Nvidia is the 980Ti.

The 970/980 is pretty meh considering they aren't even definitively faster than the older AMD chips they were supposed to be compete against. It's only the 980Ti that really was worth anything.

These are the first chips in a long time that are anything to be actually excited about.

Is anyone with a Kepler chip even happy about it? Seriously? I would NOT be considering how fast my 7950 felt.

 

[tviceman]

Even Fiji isn't bad except for Fury X. Nano at $450 now is a really good pick up, and so is Fury at $500 compared to the GTX 980.

Are these U.S.A. prices? You're spot on the Fury prices, but the cheapest Nano I'm seeing is $480. And are those prices really all that great? Looking at the latest TPU GPU review available, I'm seeing Fury and 980 tied at 1080, while the Fury is ahead by 10% at 1440p. When both cards are OC'd to their max stable, based on the scaling of past TPU reviews for each card, we're looking at 10% for Fury and 20% for open-air cooled 980's. Which means 980 wins at 1080p and both cards tie essentially 1440p. 980's are going for ~$480-500 with a free AAA game, so to me it looks like the 980 is a better value at 1080p and is tied at 1440p. As far as Nano's go, I think it's been shown that Nano's are ever so slightly slower than Fury at 1440p (3-5% slower), so take from that as you will.

The point is, people can hammer away at how bad of value the GTX 980 is in the face of GTX 980 TI, but Nano and Fury aren't bringing anything to the table of value unless the user goes CFX and 1440p / 4k. If you want to talk 4k on single card setups, then sure I totally concede Nano and Fury are better but we're talking near slide show performance in modern games.

The only impressive chip I've seen from Nvidia is the 980Ti.

The most impressive thing about 980 TI is it's price. I know that is what matters most to us users, but GM204 is IMO the most impressive chip put out. It's got the best perf/mm2 and perf/w and almost as much headroom as 980 TI. Unfortunately, Nvidia continues to price it too high in comparison to it's big brother.

Is anyone with a Kepler chip even happy about it? Seriously? I would NOT be considering how fast my 7950 felt.

Agreed. Kepler has gone down the tubes and it's sad.

 

[crisium]

AMD really needs this generation, and specially considering they are initially only making 2 dies for the consumer market, to make broader salvaging schemes. 7950 within 5-7% of 7970 perf at same clocks, 290 within 5% of 290x perf at same clocks, and the same with Fury to Fury X only hurts them badly as a lot of perf/$ conscious buyers would go for the harvested sku. And this is without even adding the shading unlock thing that happened with Cayman and with Fiji. A perfect example of good die harvesting segmentation is 970 and 980 (albeit the technical flaws that resulted in 970's gimped VRAM/bandwith/caches), a lot more people went 970 because of the incredible perf/$ at launch, but the 17-20% initial difference between them made a lot of people shell the extra bucks (and the price difference was indeed big) for the 980.

AMD's second tier card being within 5-7% at equal clocks was often great for consumers though. You're right from a business perspective it gives less incentive to buy the big boy though. And although harvesting more than you need to means you are selling a potentially more expensive card at a lower price, they need volume to get a hold of the market.

I think only going with 2 cards per die is a mistake. It means more rebadging of older feature cards, or large gaps in performance. While the 380X is there as a 960 4GB killer, what really could have been impressive was a further cut down Hawaii at 2304 Shaders, perhaps 48 ROPs, and 384-bit.

Interestingly AMD never plays around with ROPs, and rarely cuts bus width as well. GK104 had two different cards with 24 ROPs and 192-bits, for example. The most daring cut AMD did was the 7870XT, which further cut back Tahiti to 1536 shaders and from 384-bit to 256-bit. A hypothetical Tonga cut down further like Tahiti did could be a true GTX 950 killer, instead of the 370 and its GCN 1.0 limitations.

AMD needs new cards for their full range. I don't want to see Hawaii try to compete against Pascal at all. Even Tonga has to go when you consider how far the performance per watt is behind Maxwell. I can't imagine the hot and hungry AMD memes if 28nm still goes against 16nm Pascal. Everything above Bonaire performance needs to be killed, except the Nano.

 

[tential]

Are these U.S.A. prices? You're spot on the Fury prices, but the cheapest Nano I'm seeing is $480. And are those prices really all that great? Looking at the latest TPU GPU review available, I'm seeing Fury and 980 tied at 1080, while the Fury is ahead by 10% at 1440p. When both cards are OC'd to their max stable, based on the scaling of past TPU reviews for each card, we're looking at 10% for Fury and 20% for open-air cooled 980's. Which means 980 wins at 1080p and both cards tie essentially 1440p. 980's are going for ~$480-500 with a free AAA game, so to me it looks like the 980 is a better value at 1080p and is tied at 1440p. As far as Nano's go, I think it's been shown that Nano's are ever so slightly slower than Fury at 1440p (3-5% slower), so take from that as you will.

The point is, people can hammer away at how bad of value the GTX 980 is in the face of GTX 980 TI, but Nano and Fury aren't bringing anything to the table of value unless the user goes CFX and 1440p / 4k. If you want to talk 4k on single card setups, then sure I totally concede Nano and Fury are better but we're talking near slide show performance in modern games.




The most impressive thing about 980 TI is it's price. I know that is what matters most to us users, but GM204 is IMO the most impressive chip put out. It's got the best perf/mm2 and perf/w and almost as much headroom as 980 TI. Unfortunately, Nvidia continues to price it too high in comparison to it's big brother.



Agreed. Kepler has gone down the tubes and it's sad.

1080p is a resolution that does not exist if you're a Nano/980 purchaser in my mind. If you use it. Great. But even if you have a 1080p screen, 1440p VSR > 1080p native every day. No matter what. I never use 1080p resolution EVER on my 1080p screen.

Nano:
http://www.newegg.com/Product/Product.aspx?Item=N82E16814131679

There is no way I'd get a GTX 980 for 1440p over the Nano or Fury. And there is no way I'd game at 1080p with such a card. 1440p minimum.

I guess because GCN scales better with resolution, more people are thinking about using higher resolutions when they get R9 290/390/Fury/whatever. I just would never think of using a $500 card for 1080p....

And it's the latest games especially that I'm not impressed with the GTX 980.

 

[crisium]

1080p is a resolution that does not exist if you're a Nano/980 purchaser in my mind. If you use it. Great. But even if you have a 1080p screen, 1440p VSR > 1080p native every day. No matter what. I never use 1080p resolution EVER on my 1080p screen.

Nano:
http://www.newegg.com/Product/Product.aspx?Item=N82E16814131679

There is no way I'd get a GTX 980 for 1440p over the Nano or Fury. And there is no way I'd game at 1080p with such a card. 1440p minimum.

I guess because GCN scales better with resolution, more people are thinking about using higher resolutions when they get R9 290/390/Fury/whatever. I just would never think of using a $500 card for 1080p....

I used a 1080p 60Hz screen for nearly the first 4 months of my new build, and you are quite right that I used 1440P VSR in every game.

PC Part Picker has

390X: $378.98
980: $463.97
Nano: $464.98
Fury: $499.99

But...

390: $274.99
980 Ti: $599.99

...is the real kicker. It's hard to justify any of these middle cards over the radically cheaper 390 and significantly faster 980 Ti.

 

[tential]

I used a 1080p 60Hz screen for nearly the first 4 months of my new build, and you are quite right that I used 1440P VSR in every game.

PC Part Picker has

390X: $378.98
980: $463.97
Nano: $464.98
Fury: $499.99

But...

390: $274.99
980 Ti: $599.99

...is the real kicker. It's hard to justify any of these middle cards over the radically cheaper 390 and significantly faster 980 Ti.

Just at every price bracket I listed.

Yes, the 2 cards you listed are the ONLY ones that matter. If you aren't getting R9 390 or a GTX 980Ti, you're pretty crazy.

 

[tviceman]

1080p is a resolution that does not exist if you're a Nano/980 purchaser in my mind. If you use it. Great. But even if you have a 1080p screen, 1440p VSR > 1080p native every day. No matter what. I never use 1080p resolution EVER on my 1080p screen.

Nano:
http://www.newegg.com/Product/Product.aspx?Item=N82E16814131679

There is no way I'd get a GTX 980 for 1440p over the Nano or Fury. And there is no way I'd game at 1080p with such a card. 1440p minimum.

I guess because GCN scales better with resolution, more people are thinking about using higher resolutions when they get R9 290/390/Fury/whatever. I just would never think of using a $500 card for 1080p....

And it's the latest games especially that I'm not impressed with the GTX 980.

I'm not going to argue your mindset (it's a good one), I was just arguing the numbers. It's of my opinion that Fury, Nano, or GTX 980 are currently barely competent 1440p. I don't think Fury or Nano is capable at 4k even though they're faster than the 980, which leaves 1440p and 1080p comparisons. And since 4k is out of the question IMO, so is 4k DSR/VSR and 1440p VSR, as I pointed out already, results in essentially ties across the board between the 3 cards.

I say this as a GTX 980 (with OC) owner on a 1440p monitor.

Just at every price bracket I listed.

Yes, the 2 cards you listed are the ONLY ones that matter. If you aren't getting R9 390 or a GTX 980Ti, you're pretty crazy.

You are exactly right and this really is what the current selection boils down to. Talking about the 980, Fury, and Nano as a potential buy right now is absurd.

 

[AtenRa]

There are people that keep their Graphics Cards for more than a year you know. They keep them for 3-4 years easily, those users thinking is to buy the fastest they can at the time of purchase in order to have enough performance for the next games down the road 3-4 years later. Those users will buy the Fury/Nano and GTX980 for 1080p monitors easily.
There are also those with 120/144Hz 1080p monitors, those will also buy Fury/Nano and GTX 980 easily.

 

[maddie]

AMD really needs this generation, and specially considering they are initially only making 2 dies for the consumer market, to make broader salvaging schemes. 7950 within 5-7% of 7970 perf at same clocks, 290 within 5% of 290x perf at same clocks, and the same with Fury to Fury X only hurts them badly as a lot of perf/$ conscious buyers would go for the harvested sku. And this is without even adding the shading unlock thing that happened with Cayman and with Fiji. A perfect example of good die harvesting segmentation is 970 and 980 (albeit the technical flaws that resulted in 970's gimped VRAM/bandwith/caches), a lot more people went 970 because of the incredible perf/$ at launch, but the 17-20% initial difference between them made a lot of people shell the extra bucks (and the price difference was indeed big) for the 980.

I have come to the belief that AMD designs for the harvested die and sells the full one as sort of "icing on the cake" with regards to revenue.

 

[Glo.]

I know its hype train but...

AMD lately likes to play with public through Twitter, right?

https://twitter.com/GChip/status/700420476070301697

8.6 TFLOPs of Compute power?
4096 GCN cores clocked at 1050 MHz?
Well, so far everything lines up...
232 mm2, 125W TDP, HBM2 4096 GCN cores.

Price, performance? If next gen SIMD blocks reflect Maxwell Performance even by 0.9 ratio, we could expect 20% faster GPU than... Titan X. IF all of this is correct.

 

[Dice144]

I sure hope so. My 290x is struggling with my 4k screen. Almost got a fury but didn't think it was a big enough jump to justify the cost. Some games I just play at 1080 for now

 

[MrTeal]

I know its hype train but...

AMD lately likes to play with public through Twitter, right?

https://twitter.com/GChip/status/700420476070301697

8.6 TFLOPs of Compute power?
4096 GCN cores clocked at 1050 MHz?
Well, so far everything lines up...
232 mm2, 125W TDP, HBM2 4096 GCN cores.

Price, performance? If next gen SIMD blocks reflect Maxwell Performance even by 0.9 ratio, we could expect 20% faster GPU than... Titan X. IF all of this is correct.

How do you get 8.6TFLOPS, 4096 shaders HBM2 and 232mm² from those tweets?

 

[tential]

How do you get 8.6TFLOPS, 4096 shaders HBM2 and 232mm² from those tweets?

Same question I looked and found nothing...

 

[Glo.]

Not from these tweets. It is apparent that it is game of words there in that tweet. 8 and 6 are key numbers here.

232 mm2 is from previous rumors about AMD 14 nm FinFET GPU.
4096 GCN cores is the value of core count which gives 8.6 TFLOPs of compute power at 1050 MHz.
And HBM 2 from Sisoft table of contents, about GPU tested in december with that core count, and 2048 Bit memory bus and 3 GB of VRAM.

 

[littleg]

16 million too, what could that be about?

 

[MrTeal]

Not from these tweets. It is apparent that it is game of words there in that tweet. 8 and 6 are key numbers here.

1.232 mm2 is from previous rumors about AMD 14 nm FinFET GPU.
2.4096 GCN cores is the value of core count which gives 8.6 TFLOPs of compute power at 1050 MHz.
3.And HBM 2 from Sisoft table of contents, about GPU tested in december with that core count, and 2048 Bit memory bus and 3 GB of VRAM.

16 million too, what could that be about?

That's all a hell of a stretch. 8-methyl-N-vanillyl-6-nonenamide is capsaicin, the active ingredient in hot sauce. Pure cap is 16M scoville units, which is the defacto "Heat Rating" for peppers. Raja and Chris Hook have been tweeting back and forth about hot sauce for over a month.
https://twitter.com/GChip/status/684603003307859968

About the other points
1. All the 232mm² came from a linked in profile, which while not nothing is a pretty flimsy evidence, especially considering it's not linked to any GPU in particular. It's possible they will launch a GPU at around that size though.
2. See above. 8.6 doesn't mean anything. They might launch a 4096 core GPU (though @ 232mm² would be really, really good scaling), but it has nothing to do with hot sauce.
3. SiSoft is just always wrong, in its memory configurations, as discussed before. Fury's regularly show up with that same memory config.

http://ranker.sisoftware.net/top_ru...c7facbed85b888aed6ebdafc99fcc1f1d7a499a9&l=en
Reading anything into them is silly. Besides, there is absolutely zero chance of AMD shipping a >Fiji class GPU with 3GB of VRAM with some goofy 2GB+1GB asymmetric topology.

 

[Magee_MC]

16 million too, what could that be about?

16 million Scoville units is equal to pure capsaicin. My guess is that they're hinting at the name of the next card.