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

[NTMBK]

http://arstechnica.com/gaming/2016/01/amd-confirms-high-end-polaris-gpu-in-development-for-2016/

AMD is working on two versions of its upcoming Polaris graphics architecture: Polaris 10 and Polaris 11. In an interview with VentureBeat, graphics chief Raja Koduri explained that one of those GPUs is aimed at thin-and-light laptops and entry-level desktops, while the the other is a larger, high-performance GPU designed to take back the premium graphics card market currently dominated by rival Nvidia.

 

[railven]

Sign me up! I'm in for a blood bath!

I want to take out the GTX 680 from the living room HTPC (man that thing is too loud), so a nice 1080p@60 competent card is worth it (though I just got a Steam Link and it works AWESOME, so I might not even need a PC in the living room anymore).

My 980 Ti is already disgusting to look at. Papa want's a new toy! (Actually want a new Rig, main parts are over a year old )

 

[maddie]

http://arstechnica.com/gaming/2016/01/amd-confirms-high-end-polaris-gpu-in-development-for-2016/

AMD is working on two versions of its upcoming Polaris graphics architecture: Polaris 10 and Polaris 11. In an interview with VentureBeat, graphics chief Raja Koduri explained that one of those GPUs is aimed at thin-and-light laptops and entry-level desktops, while the the other is a larger, high-performance GPU designed to take back the premium graphics card market currently dominated by rival Nvidia.

I have been getting a lot of flak from some for a speculative idea I have been pursuing.

Koduri is not some ignorant mid-level employee. He is a brilliant designer and head honcho at RTG.

OK folks, please explain how this will happen with two Polaris versions, if we know that one GPU is tiny [~110mm^2] and they want to take back the premium card market from Nvidia.

 

[Techhog]

I have been getting a lot of flak from some for a speculative idea I have been pursuing.

Koduri is not some ignorant mid-level employee. He is a brilliant designer and head honcho at RTG.

OK folks, please explain how this will happen with two Polaris versions, if we know that one GPU is tiny [~110mm^2] and they want to take back the premium card market from Nvidia.

You're wrong. This actually proves beyond doubt that there won't be a monolithic GPU. The smaller GPU uses GDDR5, so it's not viable for this idea you've been desperately pushing. Also, we've already known that there would be two GPUs, so this obviously won't change any minds. It's over, Maddie.

 

[railven]

I have been getting a lot of flak from some for a speculative idea I have been pursuing.

Koduri is not some ignorant mid-level employee. He is a brilliant designer and head honcho at RTG.

OK folks, please explain how this will happen with two Polaris versions, if we know that one GPU is tiny [~110mm^2] and they want to take back the premium card market from Nvidia.

I'd assume the larger one that has no disclosed information relative to size or performance?

And do they mean the performance from from the current Titan X/980 Ti kings or do they know future NV performance?

But I'll go with you.

They are going to fusion 400,000 tiny cores into one interposer and render a sun!

 

[Rvenger]

I have been getting a lot of flak from some for a speculative idea I have been pursuing.

Koduri is not some ignorant mid-level employee. He is a brilliant designer and head honcho at RTG.

OK folks, please explain how this will happen with two Polaris versions, if we know that one GPU is tiny [~110mm^2] and they want to take back the premium card market from Nvidia.

I could care less about markertshare etc etc. I just would like them to stay alive with a healthy marketshare to make cards in the foreseeable future. I am glad to see Polaris gaining some traction, this is welcomed news and can't wait until we see a dual GPU offering in this tech as well. (Not dual Fiji)

 

[railven]

I could care less about markertshare etc etc. I just would like them to stay alive with a healthy marketshare to make cards in the foreseeable future. I am glad to see Polaris gaining some traction, this is welcomed news and can't wait until we see a dual GPU offering in this tech as well. (Not dual Fiji)

Gaining traction on what?

If anything, to cite the long banned forum poster Scali - we're heading straight into Funny Season. Where the hype is going to become so unbearable the end results for every involved are going to be disappointing.

I mean, step into the CPU Subsection there is so much hope and expectations on the shoulders of Zen. Woof. Is AMD GPUs finally reaching that point?

This is the last great hurrah! If they fail at this, is it over? I swear if we get more of the "they are going to daisy link x-cores and beat NV" I might just leave until the cards are released haha.

The hype train crash and exploded once we saw Fury X results. And don't get me started on Hawaii and it's "with mantle will beat SLI 780's" nonsense.

 

[Arachnotronic]

Gaining traction on what?

If anything, to cite the long banned forum poster Scali - we're heading straight into Funny Season. Where the hype is going to become so unbearable the end results for every involved are going to be disappointing.

I mean, step into the CPU Subsection there is so much hope and expectations on the shoulders of Zen. Woof. Is AMD GPUs finally reaching that point?

This is the last great hurrah! If they fail at this, is it over? I swear if we get more of the "they are going to daisy link x-cores and beat NV" I might just leave until the cards are released haha.

The hype train crash and exploded once we saw Fury X results. And don't get me started on Hawaii and it's "with mantle will beat SLI 780's" nonsense.

:thumbsup:

People are acting like NV has nothing in the pipeline while AMD has all of the goods.

Anyway, like I said in an earlier post, whoever is first out of the gate with a high end large GPU on 14/16nm will be in a good position/get my money. A lot of people tired of 28nm GPUs will jump at the chance to grab some 14/16nm goodness and whoever can deliver first will enjoy a nice boost in the marketplace, IMO.

 

[AnandThenMan]

:thumbsup:

People are acting like NV has nothing in the pipeline while AMD has all of the goods.

Who has said this? All I've seen is it looks like AMD will be first on 14/16nm going by the fact that AMD has shown working silicon Nvidia has shown non-working mockups. But I'd like to see those posts saying Nvidia has nothing in the 16nm pipeline provide a few links. :thumbsup:

 

[maddie]

You're wrong. This actually proves beyond doubt that there won't be a monolithic GPU. The smaller GPU uses GDDR5, so it's not viable for this idea you've been desperately pushing. Also, we've already known that there would be two GPUs, so this obviously won't change any minds. It's over, Maddie.

Am I in the twilight zone?

I asked a question and I'm wrong.

This actually proves there won't be a monolithic GPU

I think that is what I was proposing for the multi-die option.

The smaller GPU uses GDDR5, so it's not viable for this idea you've been desperately pushing

Never claimed the small die GDDR5 version was to be designed for multi-die. Where did you ever get that idea?

 

[maddie]

I could care less about markertshare etc etc. I just would like them to stay alive with a healthy marketshare to make cards in the foreseeable future. I am glad to see Polaris gaining some traction, this is welcomed news and can't wait until we see a dual GPU offering in this tech as well. (Not dual Fiji)

This is exactly how I feel. The option to buy competing reasonably priced graphic cards is valuable to me.

 

[LTC8K6]

Glad I didn't buy a Fury card...

 

[maddie]

Glad I didn't buy a Fury card...

Unless they have money to burn, I think any top end buy will be in the same boat.

 

[guskline]

I'm anxious to see the highest end Polaris. It should have HBM2, at least 8Gb ram and a new Polaris core.

My twin R9 290s bench slightly faster than my single GTX980TI but produce a lot of heat (watercooling helps)

I wonder how much faster a single Highest end Polaris will be?

 

[GodisanAtheist]

The chip Kourdri is talking about is almost certainly a 250mm-350mm chip which will pack in moderately more performance than the 980ti/Fury X (something like 20-30% faster).

It will be to current high end cards what the 7970 was to the 6970/580.

 

[Techhog]

Am I in the twilight zone?

I asked a question and I'm wrong.

I think that is what I was proposing for the multi-die option.

Never claimed the small die GDDR5 version was to be designed for multi-die. Where did you ever get that idea?

There would be no point in using a big die for the multi-die solution. If there are only two dies, one can't be used for this solution, and the other makes no sense to use, the only possible option is that this solution does not exist. It's that simple. You seem to be attempting to imply that there will be an HBM version of the small die, but that's also not going to happen due to limited yields and cost. Even if the two combine into a mid-range die, this idea would not be viable given the state of HBM in 2016. On top of this, different memory controller = different die, and AMD clearly said that there would be two only new 14nm dies this year. Creating a hybrid HBM/GDDR5 die isn't possible either. You can continue to deny it, but at the end of the day your "speculation" simply isn't a possibility.

 

[Stormflux]

OK folks, please explain how this will happen with two Polaris versions, if we know that one GPU is tiny [~110mm^2] and they want to take back the premium card market from Nvidia.

Small die Salvaged GDDR5 = Low End - 380 Successor
Small die Full/Binned GDDR5 = Mid-Tier - 390 Successor

Large die Salvaged HBM2 = High-Tier - Fury Successor
Large Die Full/Binned HBM2 = Premium-Tier - FuryX successor

/shrug Just guessing. Looking forward to Q3.

 

[Techhog]

The chip Kourdri is talking about is almost certainly a 250mm-350mm chip which will pack in moderately more performance than the 980ti/Fury X (something like 20-30% faster).

It will be to current high end cards what the 7970 was to the 6970/580.

You must be expecting this to be an extremely inefficient node, then. The difference between this node and 28nm should be bigger than the difference between 40nm and 28nm, so 30-50% is what should be expected assuming a TDP of ~160-200W.

 

[maddie]

There would be no point in using a big die for the multi-die solution. If there are only two dies, one can't be used for this solution, and the other makes no sense to use, the only possible option is that this solution does not exist. It's that simple. You seem to be attempting to imply that there will be an HBM version of the small die, but that's also not going to happen due to limited yields and cost. Even if the two combine into a mid-range die, this idea would not bde viable given the state of HBM in 2016. You can continue to deny it, but at the end of the day your "speculation" is going to be exposed for the baseless pipedream it is.

My question was very specific. Answer it.

"OK folks, please explain how this will happen with two Polaris versions, if we know that one GPU is tiny [~110mm^2] and they want to take back the premium card market from Nvidia."

It appears that Nvidia is prepping a full line of products. From GP100 to GP 107. All I want to know is how will AMD expect to do this below and still compete in the middle range seeing that they claimed that the demoed Polaris is intended to compete in the GP106 range. Except for the lowest market segment, will they only produce one big Die and and then harvest all the way down the price range using one design? Remember where the big earners are sold. [GTX970]

while the the other is a larger, high-performance GPU designed to take back the premium graphics card market currently dominated by rival Nvidiahttp://arstechnica.co.uk/gadgets/20...-review-all-the-power-of-the-titan-x-for-650/

http://arstechnica.co.uk/gadgets/20...-review-all-the-power-of-the-titan-x-for-650/

 

[.vodka]

The chip Kourdri is talking about is almost certainly a 250mm-350mm chip which will pack in moderately more performance than the 980ti/Fury X (something like 20-30% faster).

It will be to current high end cards what the 7970 was to the 6970/580.

I'd consider first gen Polaris cards more like late 7970GHz with better drivers vs 6970/580.

Although this time it'll probably be different on launch day. They said Polaris includes a new, much improved revision of GCN. They're still calling it GCN, so the architecture at a high level is probably similar to what we have today, not a total paradigm shift as VLIW5/4 -> GCN 1.0 was.

That implies drivers this time will probably be much more suited to take advantage of the hardware, a completely different situation as we'd seen with the 7970 right out of the gate which took a lot of driver revisions to get its performance up to what it could actually do.... Just a guess, all of this.



Well, every day that passes is one less day to leave the 28nm generations behind once and for all. Bring it on!

 

[Headfoot]

My guess is it will be a delayed, multi-family approach on 14nm. First gen, 100-150mm2 die + 250-325mm2 big die. Next gen, add another bigger 400-425mm2 big die. Maybe finish it off with the reticle limit 600mm2 die in the spirit of gm200 and fiji. It worked on 28nm, and they are predicting 14nm to have a similarly long life. Only makes sense to pursue the same strategy

 

[MrTeal]

You must be expecting this to be an extremely inefficient node, then. The difference between this node and 28nm should be bigger than the difference between 40nm and 28nm, so 30-50% is what should be expected assuming a TDP of ~160-200W.

Should be, but likely isn't at least when it comes to density. Pretty much everything released has pointed to 16/14nm FF being about twice as dense as 28nm planar. Cayman was 6.8 million transistors per mm² while Tahiti was 12.3, so 1.8 times as dense. 14 nm isn't getting anywhere near the theoretical two node jump you'd expect just looking at 28->14.

 

[MrTeal]

My question was very specific. Answer it.



It appears that Nvidia is prepping a full line of products. From GP100 to GP 107. All I want to know is how will AMD expect to do this below and still compete in the middle range seeing that they claimed that the demoed Polaris is intended to compete in the GP106 range. Except for the lowest market segment, will they only produce one big Die and and then harvest all the way down the price range using one design? Remember where the big earners are sold. [GTX970]

[/URL]

They have two designs, Polaris 10 and Polaris 11. As you say, Polaris 10 will be the competitor to GP106. Likely Polaris 11 will be the competitor to GP104.

What leads you to believe nVidia will be releasing GP100 into the consumer market this year? GK100 wasn't released as a consumer GPU until a year after GK104, and even then it cost $1000.

 

[tential]

Unless they have money to burn, I think any top end buy will be in the same boat.

The 980ti is definitely worth it.... Fury x sucked but the 980ti was definitely worth it. Just like the 290 and even 390 is worth it, and worth it to go cf on both in my opinion due to how well cf scales.
The 970/sli is more debatable as we can see instances where it's hindered by its 3.5 in sli. In single card it's fine though, although in this price bracket I do think you should keep sli/cf open since it is great price to performance usually.

 

[Techhog]

My question was very specific. Answer it.



It appears that Nvidia is prepping a full line of products. From GP100 to GP 107. All I want to know is how will AMD expect to do this below and still compete in the middle range seeing that they claimed that the demoed Polaris is intended to compete in the GP106 range. Except for the lowest market segment, will they only produce one big Die and and then harvest all the way down the price range using one design? Remember where the big earners are sold. [GTX970]

[/URL]

The answer is simple: They won't. They don't have the money for a full new line-up, so we're going to see more rebrands. The multi-die talk is pointless because it's not currently feasible even if we accept the slim chance that it's already possible. You're just trying to be optimistic, hoping that AMD hs some kind of secret sauce to make them more competitive. Trust me, i wish that what you were saying were possible so AMD would stand a chance. Sadly, there are plenty of facts against it and none for it. The logic that you're using behind that question also could have applied to the 200 and 300 series, which is why we saw this exact same speculation both times. Nothing has changed. If you want to make tis arguement, find some actual, specific evidence. Your leading question is only proof that you lack actual evidence.