digitimesAMD expected to unveil Polaris GPU in June

[maddie]

Yea i was going to say that the name is not Polaris but Vega and so far that is the only thing we know.

So if he was wrong or not is not established yet. Oh and Polaris could come with HBM1. Nobody said its GDDR 5 only.

Did'nt Koduri say that Polaris could use either GDDR5 or HBM? Do we know for sure that Polaris 10 is using GDDR5?

 

[maddie]

I don't think that's the case at all. Vega is rumored to be fabbed at TSMC as opposed to Polaris which is at GloFo. AMD's probably been focusing their efforts on getting Polaris out and Vega is just simply behind.

I have seen this posted quite a bit without the slightest bit of evidence. Do you have any source for this rumor?

 

[Silverforce11]

Did'nt Koduri say that Polaris could use either GDDR5 or HBM? Do we know for sure that Polaris 10 is using GDDR5?

Raja was asked this directly in the video from PCPER. Polaris is GDDR5, not HBM2 due to cost, this was quite clear.

HBM1 was a prototype, I doubt you will see it being used again.

 

[USER8000]

I sure hope Nvidia isn't late to release Pascal in May. I wouldn't want to have to explain to my customers why I'm delaying. Hopefully they release before AMD like you say they will. It would be a huge blunder for Nvidia to say they'll release before AMD, then release afterwards.

Edit: But yes, Nvidia is launching in May according to latest rumors, and I won't be surprised if Nvidia is first to market. This doesn't seem to be a complex design, it seems to be more of the same, with just a node shrink. I always expect NVidia to play it smart. I won't be surprised if Nvidia capitalizes on being first to market again, even if the product isn't as fast as the competition. People won't sell their GTX 1080 for Polaris if Polaris is faster. It'd have to be 25% faster for people to do that in droves, and I doubt that will happen. FIRST to market is the most important thing.

You can't sell what you don't have to sell. And gamers are impatient and don't like to wait. For how long have we got shoddy results from Preordering games? Yet we still do it, because, OMG I need to pledge my money to this game ASAP, or I won't get it (Not like you can't buy it digitally the day of... no, we need it PRELOADED at EXACTLY midnight to play it). So if anyone thinks the vast majority of gamers have the mental stamina to wait for both GPUs to be out to evaluate them you're kidding yourself.

Nvidia first to market, expect AMD to be in 10% marketshare just like Game_Dev says.
Polaris first to market? Expect AMD to do well.

The problem is Nvidia is using the same process as companies like Apple,so they will be competing with other companies.

AFAIK,both GlobalFoundries and Samsung have more excess capacity available,so I would be very surprised to see Nvidia release the chips first unless it is some kind of paper launch.

This is the issue - the company that can supply OEM demand the best will be the one who wins out in the end.

 

[jpiniero]

I have seen this posted quite a bit without the slightest bit of evidence. Do you have any source for this rumor?

I think part of it was when AMD started talking about the GPUs they mentioned FinFet without being specific about the node size. The implication being that they are fabbing products at both foundries. Since we can be pretty sure that Polaris is at GloFo, it's sort of a process of elimination that Vega would be at TSMC. Which makes sense, TSMC has been at 20P/16FF for 2+ years now and would be in better shape to do the bigger dies than GloFo which hasn't shipped anything in volume yet on Samsung's node.

 

[maddie]

Raja was asked this directly in the video from PCPER. Polaris is GDDR5, not HBM2 due to cost, this was quite clear.

HBM1 was a prototype, I doubt you will see it being used again.

And yet in that same PcPer video, Shrout seemed to believe this was implied.

Ryan Shrout:
http://www.pcper.com/news/Graphics-...past-CrossFire-smaller-GPU-dies-HBM2-and-more

Though he didn’t say it outright, it appears that Polaris will be using HBM1, leaving me to wonder about the memory capacity limitations inherent in that. Has AMD found a way to get past the 4GB barrier? We are trying to figure that out for sure.

By the way, there is no 4 GB barrier. Only 1 GB / HBM stack.

 

[Silverforce11]

I don't know how Ryan Shrout can reach that conclusion from what Raja said to him.

Polaris using HBM would also defeat it's target market segment on price.

 

[maddie]

I think part of it was when AMD started talking about the GPUs they mentioned FinFet without being specific about the node size. The implication being that they are fabbing products at both foundries. Since we can be pretty sure that Polaris is at GloFo, it's sort of a process of elimination that Vega would be at TSMC. Which makes sense, TSMC has been at 20P/16FF for 2+ years now and would be in better shape to do the bigger dies than GloFo which hasn't shipped anything in volume yet on Samsung's node.

At first some took the 2 foundries statement to mean TSMC and GloFlo, ignoring the benefits of having 2 foundries using the same process as hedges against production problems at and one.

The 2 foundries most likely means GloFlo and Samsung. It's a non-trivial task to design to two processes. A cash constrained company will not do both especially as the Samsung process appears to be denser and with possibly better electrical characteristics.

Also, don't forget that AMD is going to fab some very high core server CPUs on this very same process. Big dies.

 

[maddie]

I don't know how Ryan Shrout can reach that conclusion from what Raja said to him.

Polaris using HBM would also defeat it's target market segment on price.

Having watched the interview, I tend to agree with Shrout's interpretation.

Having said that, your reservation about HBM is also valid. ??????

 

[raghu78]

Wow, you got this way wrong. Would you like to make any more predictions?

The names are different but their product stack now has 2 Polaris GPUs based on GDDR5 and 2 Vega GPUs based on HBM2. Polaris 10 and 11 are GDDR5 based and meant to hit mainstream price points. I am guessing that means USD 149 - USD 349. Vega 10 and 11 are expected to be HBM2 based and complete the entire product stack. anandtech believes that Vega GPUs will be complementary to Polaris and will address the higher price segments. I find that very plausible.

http://anandtech.com/show/10145/amd-unveils-gpu-architecture-roadmap-after-polaris-comes-vega

"Vega is currently scheduled to come relatively quickly after Polaris. Depending on how literal you interpret this chart, the far left edge of the Vega box does fall into 2016, though obviously AMD intends to leave themselves some wiggle room here and not tie themselves down to specific dates. The fact that Vega comes this soon after Polaris is interesting; it seems hard to believe that it’s a direct successor to Polaris – I can’t see AMD replacing Polaris parts in less than a year – so this points to Vega being more of a cousin, and is where AMD’s naming system isn’t especially helpful in deciphering anything further.

With Polaris confirmed to use GDDR5, Vega is notable for being the first AMD architecture to use HBM2, and the first parts in general to use HBM tech since Fiji. I’m presuming these are higher-end GPUs to complement the Polaris GPUs (the smaller of which we know to be a low-power laptop design), which is where HBM would be more cost-effective, at least at current prices.

Meanwhile AMD has also confirmed the number of GPUs in the Vega stack and their names. We’ll be seeing a Vega 10 and a Vega 11. This follows Polaris GPU naming – which has finally been confirmed – with Polaris 10 and Polaris 11. I have also been told that Polaris 11 is the smaller of the Polaris GPUs, so at this point it’s reasonable to assume the same for Vega."

Anyway this is how I see the chips fall

GP107 vs Polaris 11
GP106 vs Polaris 10
GP104 vs Vega 11
GP100 vs Vega 10

 

[Game_dev]

NVIDIA will probably have a refresh ready by the time Vegan gets here

 

[crisium]

They'll probably have Volta by the time Vegeta gets here.

 

[swilli89]

They'll probably have Volta by the time Vegeta gets here.

Shit it will probably be here before Goku and Piccolo even finish off Raditz /nerd

 

[exar333]

They'll probably have Volta by the time Vegeta gets here.

 

[Mahigan]

Shit it will probably be here before Goku and Piccolo even finish off Raditz /nerd

I still think that when Potter and Narnia hit, things will all of a sudden get magical

 

[itsmydamnation]

They'll probably have Volta by the time Vegeta gets here.

over 9000 GFLOPS confirmed!!!!

 

[AtenRa]

I dont expect Vega for 2016 release.

 

[Mahigan]

over 9000 GFLOPS confirmed!!!!

Over 9000? What Warp Factor is that?

 

[jpiniero]

At first some took the 2 foundries statement to mean TSMC and GloFlo, ignoring the benefits of having 2 foundries using the same process as hedges against production problems at and one.

You are right that it would make more sense to do Samsung. TSMC looks like it has better yield so that might be the reason.

Also, don't forget that AMD is going to fab some very high core server CPUs on this very same process. Big dies.

Yeah but not in the timeframe that Vega is going to be released.

 

[USER8000]

TSMC looks like it has better yield so that might be the reason.

Thats why AMD has shown of both Polaris 10 and 11 already! :sneaky:

We can't say really anything about yields apart from the fact that even though Apple wanted to go with TSMC they needed to dual source with Samsung too.

 

[Silverforce11]

Thats why AMD has shown of both Polaris 10 and 11 already! :sneaky:

We can't say really anything about yields apart from the fact that even though Apple wanted to go with TSMC they needed to dual source with Samsung too.

That's actually the loudest, because actions > words. Apple is not in the business of enabling Samsung unless they absolutely have to.

 

[ShintaiDK]

We can't say really anything about yields apart from the fact that even though Apple wanted to go with TSMC they needed to dual source with Samsung too.

You got that part wrong. Apple wanted A9 to be Glofo/Samsung. Ended up being Samsung/TSMC and A9X is TSMC exclusive. A10 is TSMC exclusive.

 

[Erenhardt]

Apple wanted A9 to be Glofo/Samsung.

Hard to believe anyone would want to tie his business with Glofo.

 

[arandomguy]

Hard to believe anyone would want to tie his business with Glofo.

http://www.forbes.com/sites/timwors...ke-apples-processors-by-samsung/#32df57535dc1

Sources close to the situation said the deal that appears to be taking shape looks more like this: Samsung will use GlobalFoundries for what is known as “flex capacity.” This is a long-standing industry practice under which a chip manufacturer pays to occasionally use another company’s factories when demand on their own factory is running higher than they would like, and they need a little help.
This would be a good time to point out that Apple is not Samsung’s only foundry customer. The Samsung fab in Austin, Texas, also turns out chips for Samsung. Occasionally there will be times when Samsung has to balance the demand on that fab in order to meet both the needs of its primary foundry customer — Apple — as well as its own internal needs for smartphone and tablet chips.
​

That’s where GlobalFoundries will come in, picking up the additional work on an as-needed basis. Samsung would basically hire GlobalFoundries as a subcontractor, and continue to manage the relationship with Apple. This is a very different business relationship than, say, if Apple were to tap GlobalFoundries as a “second source” for chips. Apple would of course have to give its blessing to the arrangement.
​

 

[exar333]

Hard to believe anyone would want to tie his business with Glofo.

Would definitely want to add Samsung to that as well. They just lost a big case regarding a key engineer who defected TSMC and gave up many trade secrets to Samsung. Apple also had issues with Samsung which caused them to sever a lot of manufacturing ties.

Sorry, a bit off-topic...