Polaris foundry, the official word

[Arachnotronic]

Lots of discussion here on where Polaris is being built. Some think it will be GloFo exclusive, others think it will be GloFo/TSMC split.

I contacted AMD and was told that AMD is "engaging with multiple foundries" for Polaris and that the part they demo'd (the GTX 950 competitor) was built on GloFo 14nm.

 

[railven]

Source. Link.

Nah, but seriously, I have zero confidence in GloFlo. Until they actually deliver that uneasiness will not go away.

 

[Arachnotronic]

Source. Link.

Sadly, I can't link to my e-mail account

Nah, but seriously, I have zero confidence in GloFlo. Until they actually deliver that uneasiness will not go away.

Good thing they're engaging with multiple foundries

 

[Azix]

you can source your email. people do it all the time.

 

[Arachnotronic]

 

[Killrose]

You may want to protect your scource, his name is posted....

 

[Arachnotronic]

You may want to protect your scource, his name is posted....

Drew Praire is public relations @ AMD and his e-mail address is public. The information in the email is 100% meant for public consumption

 

[guskline]

Thank you Arachnotronic for the posting. Sounds like AMD is preparing to have different foundries do the Polaris10 and Polaris 11.

 

[Arachnotronic]

Thank you Arachnotronic for the posting. Sounds like AMD is preparing to have different foundries do the Polaris10 and Polaris 11.

That's what I'm thinking too. My bet is on 14nm GloFo for Polaris 10, and TSMC 16FF+ for Polaris 11.

 

[Azix]

I feel like they are going to mess up somewhere. There are a few products they can spread the manufacturing across though. shrink gcn 3 on tsmc for example and put that in the mid range with more VRAM. GCN 4 completely designed on a single process etc

 

[railven]

That's what I'm thinking too. My bet is on 14nm GloFo for Polaris 10, and TSMC 16FF+ for Polaris 11.

If TSMC can reach higher clocks in the current iteration of their node, it would make sense for them to bring out bigger Polaris.

If someone over at AMD is trying to retract the "And TSMC" line, I'd figure it was something that wasn't meant to be revealed yet.

 

[Arachnotronic]

I feel like they are going to mess up somewhere. There are a few products they can spread the manufacturing across though. shrink gcn 3 on tsmc for example and put that in the mid range with more VRAM. GCN 4 completely designed on a single process etc

Shrinking an architecture from 28nm to 16nm should be no easier than porting a 14nm architecture to 16nm...

 

[PPB]

Multiple foundries doesnt necesarily mean multiple nodes.

They may very well mean just Glofo and Samsung 14LPP. 2 foundries, 1 node.

 

[Arachnotronic]

Multiple foundries doesnt necesarily mean multiple nodes.

They may very well mean just Glofo and Samsung 14LPP. 2 foundries, 1 node.

Lol, okay, I'm sure that's what AMD meant.

 

[Techhog]

Lol, okay, I'm sure that's what AMD meant.

This is the most likely situation, actually. It's disappointing, but it's what you should realistically expect.

 

[Azix]

Shrinking an architecture from 28nm to 16nm should be no easier than porting a 14nm architecture to 16nm...

I was assuming that 28nm and 16nm being both tsmc would make that easier. rather than working between two manufacturers on the same architecture. Only going off random things heard/read in the past though. Maybe it's neither here nor there.

This is the most likely situation, actually. It's disappointing, but it's what you should realistically expect.

SAMSUNG was confirmed . but I dunno. Because he specified polaris and only mentioned glofo you could say for these GPUs its glofo and samsung. that makes most sense for a single or similar architectures. They would not need TSMC for polaris at that point

http://english.etnews.com/20151222200002

 

[PPB]

Lol, okay, I'm sure that's what AMD meant.

You can believe anything you want, I couldn't care less. Asking the PR department will only get you a PR response. Suddenly you started believing AMD PR after dismissing it so much in the CPU forum?

 

[Arachnotronic]

You can believe anything you want, I couldn't care less. Asking the PR department will only get you a PR response. Suddenly you started believing AMD PR after dismissing it so much in the CPU forum?

Combined with the claim from AnandTech that they will be using TSMC 16FF+, as well as the information I have seen on LinkedIn which shows GPU engineers working on 16FF+ dGPUs, I am very confident that we will see a TSMC 16FF+ dGPU.

 

[PPB]

Anandtech still thinks GCN 1.1+ can only do 8 compute queues in parallel Talk about quoting a tech site as the absolute truth.

Just google "confirmation bias", take a nap and think why you are so obsessed with Polaris being 2 different nodes.

 

[Arachnotronic]

Just google "confirmation bias", take a nap and think why you are so obsessed with Polaris being 2 different nodes.

Why don't you tell me? Be honest, I don't like it when people BS me or beat around the bush.

 

[ShintaiDK]

Its no surprise AMD only fabs the absolute amount of parts they need to at GloFo due to the WSA. Without the WSA I am sure every single AMD product from CPUs to GPUs would be made at TSMC. But that's what happens when you make a deal with the devil.

 

[PPB]

Why don't you tell me? Be honest, I don't like it when people BS me or beat around the bush.

Occam's razor. You can google that too.

Its just easier to make all your GPU dies in 1 node from a design standpoint. Whether that leaves you with a competitive stack against your competition is another matter entirely.

But the first gen 14/16nm A9 debacle (which was based totally on a case of statistical deviation) carved so deeply on some people's minds that are already making extrapolations on second gen 14/16nm performance and can't think of a big die being made in anything but 16nm ff+, even if they havent seen how a big die on 16nm ff+ performs.

But hey, if thinking that big die Polaris will be made on 16nm ff+ without any basis other than a ambiguous PR slide and PR response, whatever floats your boat is fine with me.

PD: What makes the poster above me think that AMD is filling the WSA quota after paying GLofo for not reaching the volume for many quarters in a row? That is even an argument in favour of going all in on 14nm LPP GF/Samsung.

 

[ThatBuzzkiller]

Multiple foundries doesnt necesarily mean multiple nodes.

They may very well mean just Glofo and Samsung 14LPP. 2 foundries, 1 node.

This guy gets it ...

OP still can't be certain whether or AMD will use another process node ...

 

[Stuka87]

Being that GloFlo uses the same 14nm tech as Samsung, I would surprised in AMD goes with TSMC 16nm.

 

[HurleyBird]

Being that GloFlo uses the same 14nm tech as Samsung, I would surprised in AMD goes with TSMC 16nm.

Prepare to be surprised. I mean, Anandtech claims they will, while the question the OP fielded specifically mentioned 16nm and the implied answer is yes. Most convincing is the fact that AMD is calling the process simply "FetFET." If they were only using 14nm, they'd call it 14nm. Especially given that Nvidia's entire stack is "only" on 16nm.

Source. Link.

Nah, but seriously, I have zero confidence in GloFlo. Until they actually deliver that uneasiness will not go away.

At this point I have very little confidence in GloFlo or TSMC. However, we know that the former is currently producing working 14nm GPUs.

I think I remember reading that both Polaris chips were on 14nm. If that's true, then given that AMD is also producing some chips on TSMC 16nm that haven't been shown off and Nvidia can't seem to get back anything demo-able from TSMC it appears GloFo is way ahead on FinFet GPUs.