Other sources wrong, Pascal not taped out in June, potentially still in testing

[Notional]

2 minutes ago, Sakkura said:

Right. If AMD launch in their vague "mid-2016" release window, and Nvidia is just a month or two behind with their mainstream GPUs, then Big Pascal slipping to 2017 seems very reasonable. It took 11 months from the GTX 680 to the GTX Titan.

Polaris is confirmed for August release.

[Sakkura]

3 minutes ago, Notional said:

Polaris is confirmed for August release.

I believe it's "confirmed" WCCF-style. I don't think AMD has said anything other than mid-2016.

[-BirdiE-]

Lovely. "Other speculation is wrong because <insert speculation here>".

Le sigh.

[Humbug]

28 minutes ago, Sakkura said:

Maybe the flagship will be roughly Tahiti-sized. On paper, the planar density should about quadruple, so that size still gives you a heck of a lot of transistors to work with.

yes but simply beating the 980ti is not the goal. That's a forgone conclusion... AMD knows it has to be fast enough to compete with Pascal flagship which is on 16nm finfet. So I don't think they will hold back.. Will be an arms race as we have seen before.

[ThreePinkApples]

32 minutes ago, Sakkura said:

I believe it's "confirmed" WCCF-style. I don't think AMD has said anything other than mid-2016.

As far as I understood they've said "Summer 2016", so June-August is the best ETA we have

[DocSwag]

45 minutes ago, Sakkura said:

Maybe the flagship will be roughly Tahiti-sized. On paper, the planar density should about quadruple, so that size still gives you a heck of a lot of transistors to work with.

I remember hearing around 16 bill, twice the fury X, but I could be wrong

[Notional]

1 hour ago, Sakkura said:

I believe it's "confirmed" WCCF-style. I don't think AMD has said anything other than mid-2016.

Back to school is the target, according to Lisa Su, for both Desktop and Laptop graphics solutions.

Quote

Angestrebt wird Mitte 2016, insbesondere zum in den Vereinigten Staaten wichtigen Back-to-School-Zeitraum sollen nicht nur Desktop-Lösungen, sondern allen voran auch Notebook-Grafikchips zur Verfügung stehen.

http://www.computerbase.de/2016-01/amd-ceo-lisa-su-interview/

4 hours ago, Tomsen said:

And who is saying AMD scrapped TSMC? I'm expecting AMD to validate their GPU design on TSMC too for APU/SoC and perhaps some dGPU.

Had to revisit your comment. I've done a little digging, and apparently Polaris 10 and 11 are the old Ellesmere and Baffin, which are 14nm, mid and low end. The koth card Greenland, is apparently known as Vega 10 and could be 16nm FF+ from TSMC, if Glofo's 14nm LPP cannot handle the performance. We will see.

[samcool55]

Well i called pascal release in june BS before, seems like i'm not the only one.

And if you think about it, if the release is in june before AMD then Nvidia should tell the world about it.

Want a reason? Investors. Every company wants to make money, the more the better. And you need money to make money.

Why do you think AMD and RTG are showing the world what their future is? Mostly to get the attention from investors and hope they invest in AMD.
The hype train is more of a bonus and not really their main focus.

Every company can greatly benefit from investors, and the best way to get their attention is showing the world you are the best in the market you are focused on.

So if Nvidia can release before AMD, we would have known. And the fact Nvidia hasn't said anything can mean 2 things.

1. they know what they have is sh*t and gets destroyed by polaris.

Or 2. It's far from ready and it will take a while before they will launch pascal.

Those are probably the only 2 options why pascal isn't announced yet. 

[2FA]

54 minutes ago, DocSwag said:

I remember hearing around 16 bill, twice the fury X, but I could be wrong

That would make sense, they managed to fit roughly 8.9 billion transistors on the 28nm node and with 14/16nm they fit roughly twice as many on the same sized die.

[Sakkura]

54 minutes ago, DocSwag said:

I remember hearing around 16 bill, twice the fury X, but I could be wrong

If you quadruple the transistor count of Tahiti, you end up in the 16 billion ballpark. It's all very plausible in theory.

22 minutes ago, Notional said:

Back to school is the target, according to Lisa Su, for both Desktop and Laptop graphics solutions.

http://www.computerbase.de/2016-01/amd-ceo-lisa-su-interview/

Had to revisit your comment. I've done a little digging, and apparently Polaris 10 and 11 are the old Ellesmere and Baffin, which are 14nm, mid and low end. The koth card Greenland, is apparently known as Vega 10 and could be 16nm FF+ from TSMC, if Glofo's 14nm LPP cannot handle the performance. We will see.

She's saying mid-2016 in that article. Not August. Just no later than September. Mid-2016 would be roughly May through August.

[Sakkura]

1 hour ago, Humbug said:

yes but simply beating the 980ti is not the goal. That's a forgone conclusion... AMD knows it has to be fast enough to compete with Pascal flagship which is on 16nm finfet. So I don't think they will hold back.. Will be an arms race as we have seen before.

If we go back of the envelope with a Tahiti-sized GPU and quadrupled planar density, then you're looking at 16-17 billion transistors - twice as much as the GTX 980 Ti and Fury X pack. Quadrupled density may be a little on the optimistic side, but it's still going to be a BIG jump.

1 hour ago, aerandir92 said:

As far as I understood they've said "Summer 2016", so June-August is the best ETA we have

Mid-2016. But pretty close to the same thing.

[Notional]

4 minutes ago, Sakkura said:

She's saying mid-2016 in that article. Not August. Just no later than September. Mid-2016 would be roughly May through August.

Back to school is in August. Either way, Polaris will be on the shelves before September, and that is nice.

[Sakkura]

2 minutes ago, Notional said:

Back to school is in August. Either way, Polaris will be on the shelves before September, and that is nice.

Back to school is typically in early September.

[Khvarrioiren]

I found nothing in the article that supports the claims in the title. Clickbait so stronk.

[dalekphalm]

3 hours ago, Notional said:

Back to school is in August. Either way, Polaris will be on the shelves before September, and that is nice.

 

3 hours ago, Sakkura said:

Back to school is typically in early September.

I think you two are disagreeing on semantics on this School, in most places (North America in particular), starts first week of September. However, "Back to School" season, at least here in Canada, is considered to start in August - particularly the last two weeks of August, since that's when all the big retailers do their huge back to school shopping/sales push.

So to say they're aiming for Back To School, could very well mean by August, since they'll want to have it on the shelf, ready to purchase, before the end of August.

[Metro]

Some of you are saying AMD has a working high end sample of their GPU, however this could be the R9 490x, it might not be the successor for the Fury X

[Tomsen]

19 hours ago, Notional said:

GPU's are much simpler than CPU's in the architecture. That much is true, but afaik they also tend to be bigger. It might cause tape out to shelf to decrease from the normal 1 year, but half it? I seriously doubt that. That would put polaris on the shelves in a month or two, since they already have fully functional cards in demo.

We don't really know what TSMC is capable of atm, as only NVidia are getting big chips made. As mentioned before, phone chips worked fine on 20nm as well. The fact they work on low power, small size, low performance says little of the potential of 16nm ff+ on big high performance chips like dGPU's.

AMD has not officially stated they are using 16nm/TSMC on anything. On the contrary, they have officially stated that Polaris 10 and 11 (low and high end) are 14nm FF, and that they will use Samsung foundries on top of Global Foundries, meaning their projected production exceeds that of Global Foundries capacity.

There is 0 change of any AMD apu being TSMC manufactured. Not gonna happen. Maybe a SOC, but I honestly doubt it. Both ZEN and Polaris seems to be optimised for the 14nm FF node by Samsung/GloFo. It would make little sense to change that. The only scenario I can see, is if AMD exceeds production capabilities of both GloFo and Samsung. I seriously doubt that could happen.

They are indeed quiter simpler, but that extra validation for CPUs can take some time, especially on enterprise components.
They do tend to get bigger, because it is a big parallel powerhouse, and thats why it smart to start out with a smaller design, which is what we are seeing from AMD.
6 months is if you are experiencing no issues on either end, it might not be a big launch with multiple SKUs and whatnot. Of course there is more to it, like have developed stable drivers, getting OEM deals landed, get the chips to the warehouses and so on. Well, AMD was already showing off demos in december, from what I'm hearing..

 

Going from that perspective, we also dont know what Samsung/GloFo is capable of. Now, I dont expect that only phone chips are been manufactured at 16/14nm FinFet. There are a few fabless designhouses, I would expect them to be on the highest performance node. Don't know the die sizes of those, but I can try looking them up sometimes.

 

http://www.anandtech.com/show/9886/amd-reveals-polaris-gpu-architecture/3
"The group [ radeon technologies group ] has confirmed that they will be utilizing both traditional partner TSMC’s 16nm process and AMD fab spin-off (and Samsung licensee) GlobalFoundries’ 14nm process, making this the first time that AMD’s graphics group has used more than a single fab"
Now perhaps it might be because of volume, but I also think that been able to offer semicustom deals on more nodes is playing a huge part. If for example your customer has some IP not validated on GloFo/samsung.

 

Zero chance? Many things can happen, you know!
I'm most certainly going to expect a SoC. Also I expect some semi-custom deals been fabbed at TSMC/Samsung.
Where did you get that its been optimized specifically for 14nm? All I get is that they are optimizing for FinFet.
Again, it does not necessarily have to be volume issues for having a ride range of manufacturing capabilities.

[azurezeed]

We'll see, but this makes sense, no nvidia rumors or proof of a working Pascal is really concerning

[Briggsy]

On 2/2/2016 at 9:41 AM, Humbug said:

yes but simply beating the 980ti is not the goal. That's a forgone conclusion... AMD knows it has to be fast enough to compete with Pascal flagship which is on 16nm finfet. So I don't think they will hold back.. Will be an arms race as we have seen before.

 

I really hope so. 

 

while 28nm was milked by both companies, where flagship dies started off small and more transistors were added each year to increase performance, 14/16nm may never see massive die sizes like 28nm did due to the amount of power required to get electrons to every transistor - I read a paper a while back talking about how the power requirement increases exponentially beyond 100mm^2 with first gen finfet, and while I'm sure the latest finfet may have doubled the maximum die size before power draw goes crazy, I sincerely doubt we will see 600mm^2 gpu dies anytime soon at 14nm. I hope we see a crazy jump in performance, but I'm still skeptical about them not milking this process node even worse.

[Humbug]

50 minutes ago, Briggsy said:

 

I really hope so. 

 

while 28nm was milked by both companies, where flagship dies started off small and more transistors were added each year to increase performance, 14/16nm may never see massive die sizes like 28nm did due to the amount of power required to get electrons to every transistor - I read a paper a while back talking about how the power requirement increases exponentially beyond 100mm^2 with first gen finfet, and while I'm sure the latest finfet may have doubled the maximum die size before power draw goes crazy, I sincerely doubt we will see 600mm^2 gpu dies anytime soon at 14nm. I hope we see a crazy jump in performance, but I'm still skeptical about them not milking this process node even worse.

hmm it's not AMD/Nvidia who are to blame though regarding the milking of 28nm. They are not happy about the situation. They were let down by the fabs who struggled with the transition.

[dalekphalm]

3 minutes ago, Humbug said:

hmm it's not AMD/Nvidia who are to blame though regarding the milking of 28nm. They are not happy about the situation. They were let down by the fabs who struggled with the transition.

One of the obvious problems with a Fabless chip designer - they're literally at the mercy of the fabs. When the main fabs dropped the ball with 20nm, it screwed over both NVIDIA and AMD - costing them millions of dollars each, no doubt.

[DocSwag]

I feel like AMD is ahead of Nvidia by a couple of months. AMD has already shown working prototypes of Polaris, while Nvidia has done nothing of the sort. This article serves to show it is possible Pascal silicon still doesn't exist, and on top of the scandal with the Drive PX 2 actually not being Pascal GPUs but gtx 980ms, Pascal really might not exist. The last time Nvidia faked GPUs was with Fermi, when Fermi ended up getting delayed. If Polaris hits the market a couple of months before Pascal, AMD could hold a huge advantage.

[dalekphalm]

2 hours ago, DocSwag said:

I feel like AMD is ahead of Nvidia by a couple of months. AMD has already shown working prototypes of Polaris, while Nvidia has done nothing of the sort. This article serves to show it is possible Pascal silicon still doesn't exist, and on top of the scandal with the Drive PX 2 actually not being Pascal GPUs but gtx 980ms, Pascal really might not exist. The last time Nvidia faked GPUs was with Fermi, when Fermi ended up getting delayed. If Polaris hits the market a couple of months before Pascal, AMD could hold a huge advantage.

Indeed, in the past, NVIDIA has loved dropping new advanced products early, and rubbing it in AMD's face (Eg: 980 Ti and Titan X).

 

The fact that we haven't seen any demos at all, tells me that they're pretty far behind. Unless AMD drops the ball and runs into huge issues fine tuning their chips, they're going to have a huge advantage this year.

 

I've bought both AMD and NVIDIA GPU's, but I'm not a fan of NVIDIA's business practices. I don't want NVIDIA to go away or die or get crushed - but rather, I want AMD to succeed so it pressures NVIDIA to adopt more consumer friendly policies.

 

I also want to see AMD's marketing really take a turn for the positive, because despite having incredibly competitive products in the 300 series vs NVIDIA's 900 series, they sell a tiny fraction of the amount of cards. That makes no sense from a purely price/performance perspective. Many people won't ever even consider an AMD GPU because "they suck" or have "bad drivers" - despite neither being true (Both AMD and NVIDIA have occasional driver problems). It's baffling, but NVIDIA has really marketed themselves so much better.

[CTR640]

nVidia=nLazydia, sitting on asses, picking noses to see their precious Pascal testing.

[corsairian]

On 2/2/2016 at 4:17 PM, AresKrieger said:

Reads "semiaccurate.com", can't take this seriously anymore because too funny 

Lol.