AMD Piledriver rumours ... and expert conjecture

[Reynod]

We have had several requests for a sticky on AMD's yet to be released Piledriver architecture ... so here it is.

I want to make a few things clear though.

Post a question relevant to the topic, or information about the topic, or it will be deleted.

Post any negative personal comments about another user ... and they will be deleted.

Post flame baiting comments about the blue, red and green team and they will be deleted.

Enjoy ...

 

[Cazalan]

will definately shut up the intel fanbase saying all other fabs are 3+ years behind if they complete on schedule.

We can only hope they do catch up.

The announcement was focused on extreme mobile which might just be smaller cell phone chips. We'll have to see if they can make 2+ billion transistor chips at 14nm.

 

[mayankleoboy1]

And the chips are not complete 14nm. They are a mix of 14 and 20nm. Sort of like 14nm transistors on a substrate of 20nm transsitors.
Only the FinFET transistors will be made in 14nm node. The rest are in 20nm.

 

[fazers_on_stun]

Considering how long it takes Intel to get a new node good enough for production, I doubt GF will get 20nm on time, never mind 14nm.

I'd agree with that.

But the biggest announcement came at the end of the week. In a presentation to media, GlobalFoundries announced the future of manufacturing for the company. The 20nm process, which is scheduled to debut in 2013 will only last for a few months as the bleeding edge process, to be improved with the 14nm process coming in 2014. As you can see, this comes only a year after 20nm and on par with Intel's own 14nm process and Broadwell processors. Note that Intel will not have a low power process before 2015.

So if GF only spends a few months on 20nm, they won't come anywhere near a decent ROI and be deeper in the red (according to the article) than they already are. This seems to be the same idea that Samsung announced earlier - practically skipping a generation. Dunno for sure but I've never heard of a fab or foundry successfully accomplishing that, since each node is a feedback-type learning process for the next one.

 

[fazers_on_stun]

Read in the financial news last week that AMD has yet another sizeable payment they have to cough up to GloFlo, I think this quarter or maybe next. I just wonder if AMD actually got anything out of this "sale", or if because Hector Ruiz engineered it (and then promptly left AMD to become CEO of GF - rather suspicious, no??), AMD got the short end of the stick and actually had a net payment to GF to take the fabs...

IIRC Hector got some hefty golden parachute payment when he left AMD..

 

[cgner]

What still puzzles me is how they successfully compete on GPU market while only promising on CPU front :/

 

[Ags1]

How come AMD CPUs are at 32nm but their GPUs are at 28nm?

 

[noob2222]

I'd agree with that.

But the biggest announcement came at the end of the week. In a presentation to media, GlobalFoundries announced the future of manufacturing for the company. The 20nm process, which is scheduled to debut in 2013 will only last for a few months as the bleeding edge process, to be improved with the 14nm process coming in 2014. As you can see, this comes only a year after 20nm and on par with Intel's own 14nm process and Broadwell processors. .

So if GF only spends a few months on 20nm, they won't come anywhere near a decent ROI and be deeper in the red (according to the article) than they already are. This seems to be the same idea that Samsung announced earlier - practically skipping a generation. Dunno for sure but I've never heard of a fab or foundry successfully accomplishing that, since each node is a feedback-type learning process for the next one.

I haven't seen any news of samsung skipping any process nodes.

Austin texas is being retrofitted and expanded to 28nm and beyond (4B this year just after spending 3.5B on 32nm expansion), thier memory line 16 is on 20nm, they are "buildng a new fab somewhere" (located in Hwaseong, South Korea to be precise) for 20nm and 14nm SoC (1.9B) and building a 14nm meory fab in china(7B). Two of its memory lines are being upgraded to 20nm and 14nm System LSI for SoC production (didn't disclose amount on that)

they plan on releasing exynos 5 (arm 15) next year at 20nm, making qualcomm chips at 28nm, are making apples currently at 32nm. what are they skipping?

Aside from that Samsung is throwing some awefully large numbers out there, they have the means to back it up.

TSMC spends about 6B a year

 

[fazers_on_stun]

I haven't seen any news of samsung skipping any process nodes.

Austin texas is being retrofitted and expanded to 28nm and beyond (4B this year just after spending 3.5B on 32nm expansion), thier memory line 16 is on 20nm, they are "buildng a new fab somewhere" (located in Hwaseong, South Korea to be precise) for 20nm and 14nm SoC (1.9B) and building a 14nm meory fab in china(7B). Two of its memory lines are being upgraded to 20nm and 14nm System LSI for SoC production (didn't disclose amount on that)

they plan on releasing exynos 5 (arm 15) next year at 20nm, making qualcomm chips at 28nm, are making apples currently at 32nm. what are they skipping?

Aside from that Samsung is throwing some awefully large numbers out there, they have the means to back it up.

TSMC spends about 6B a year

http://hexus.net/business/news/general-business/40681-samsung-ready-play-catch-up-funding-14nm-fabrication/

For those who have been keeping up-to-date with the firm, you'll know that Samsung has recently moved from 45nm to 32nm production for its ARM-based SoCs; with the Exynos 4 Quad, featured in Samsung's GALAXY S III smartphone, being one of the first chips to benefit from the decrease in size.

When we take a look at the grander picture, however, Samsung is a little behind its competitors, with chip-maker, TSMC, already churning out 28nm parts for competing firms like Qualcomm and NVIDIA. Intel is even further ahead with its Ivy Bridge processors being produced at 22nm and utilising 3D transistor technology.

It makes sense then, that the firm doesn't wish to fall behind and yesterday the Samsung announced that it would be investing £1.24 billion in a new production line that will have nodes for both 20nm and 14nm production on 300mm wafers, bypassing both the 28nm and 22nm way-points.

With a completion target set for the close of 2013, this investment should see Samsung placed on-par with Intel, who has also confirmed its intention to move to 14nm next year. We wonder if TSMC intends to match Samsung's announcement, as there will no doubt be many eager customers who don't wish to fall behind.

 

[$hawn]

How come AMD CPUs are at 32nm but their GPUs are at 28nm?

CPU's are generally manufactured on the main full node...90nm->65->45->32->22 etc
GPU's on the other hand, are made on the half node, 80nm->56->40->28 etc.

These are not rules written in stone, but that's just the trend that has been followed over the years.
AMD might break this chain for the first time, by using 28nm to make its future SR CPU's, because it probably won't have access to a fully working, sufficiently high yield 22nm process anytime soon, neither from GF, nor from TSMC. It doesn't want to suffer like it had to with GF's 32 process initially.

Also, if u observe carefully, there's always almost √2, ie, 1.414, of shrinking between newer and older technologies. This means that for a fixed die design, the area will be reduced by a factor of 2, (√2 *√2), every time a shrink happens.

 

[esrever]

TSMC also does the GPUs and glofo does the CPUs. Its just how it lines up. Intels the one pushing moore's law and the rest of the industry is following.

 

[sarinaide]

I would rather they perfect a process rather than jumping to a new process which if it is GloFoundries then I have my reservations about. The simple fact that Bulldozer was cut down on from initial spec releases is enough to suggest GF was not ready for 32nm let alone 20 and 14nm.

 

[noob2222]

http://hexus.net/business/news/general-business/40681-samsung-ready-play-catch-up-funding-14nm-fabrication/

For those who have been keeping up-to-date with the firm, you'll know that Samsung has recently moved from 45nm to 32nm production for its ARM-based SoCs; with the Exynos 4 Quad, featured in Samsung's GALAXY S III smartphone, being one of the first chips to benefit from the decrease in size.

When we take a look at the grander picture, however, Samsung is a little behind its competitors, with chip-maker, TSMC, already churning out 28nm parts for competing firms like Qualcomm and NVIDIA. Intel is even further ahead with its Ivy Bridge processors being produced at 22nm and utilising 3D transistor technology.

It makes sense then, that the firm doesn't wish to fall behind and yesterday the Samsung announced that it would be investing £1.24 billion in a new production line that will have nodes for both 20nm and 14nm production on 300mm wafers, bypassing both the 28nm and 22nm way-points.

With a completion target set for the close of 2013, this investment should see Samsung placed on-par with Intel, who has also confirmed its intention to move to 14nm next year. We wonder if TSMC intends to match Samsung's announcement, as there will no doubt be many eager customers who don't wish to fall behind.

http://news.softpedia.com/news/Samsung-Ramps-Up-28nm-Apple-A6-SoC-Production-Says-Report-228220.shtml
http://www.engadget.com/2012/07/04/qualcomm-samsung-28nm-snapdragon-s4/
http://www.samsung.com/global/business/semiconductor/news-events/press-releases/detail?newsId=11861

28nm is alive and well at samsung.

Hexus article was retarded, took apart what samsung said and posted 1/4 of it. Here is the samsung PR that hexus butchered. http://www.samsung.com/us/aboutsamsung/news/newsIrRead.do?news_ctgry=irnewsrelease&news_seq=20185

So, who would you believe, hexus or samsung when it comes to what samsung is doing?

as for 22nm seems now that only Intel is going there.

tsmc isn't (they usually skip all full nodes), gf isn't, samsung isn't, UMC from what I can find may or may not have a 22nm (talk of it in 2008, but recent articles point to 20nm finfet). Looks like 22nm is alone in the world and will only exist in one place.

 

[mayankleoboy1]

Looks like 22nm is alone in the world and will only exist in one place.

It is also the first actual mass-produced product on 22nm and FinFET.
For all we know, PR from GLOFO is vaporware. Or they are criminally overestimating their capability to get some shares.

Intel has had 22nm for about an year now.
Even TSMC cant compare to Intel in this regard. They have had trouble with 28nm.

Even if Intels implementation of 22nm FinFET would have been piss-poor , it is miles ahead of a vaporware tech. And they already are working on 14nm.

 

[fazers_on_stun]

http://news.softpedia.com/news/Samsung-Ramps-Up-28nm-Apple-A6-SoC-Production-Says-Report-228220.shtml
http://www.engadget.com/2012/07/04/qualcomm-samsung-28nm-snapdragon-s4/
http://www.samsung.com/global/business/semiconductor/news-events/press-releases/detail?newsId=11861

28nm is alive and well at samsung.

Hexus article was retarded, took apart what samsung said and posted 1/4 of it. Here is the samsung PR that hexus butchered. http://www.samsung.com/us/aboutsamsung/news/newsIrRead.do?news_ctgry=irnewsrelease&news_seq=20185

So, who would you believe, hexus or samsung when it comes to what samsung is doing?

as for 22nm seems now that only Intel is going there.

tsmc isn't (they usually skip all full nodes), gf isn't, samsung isn't, UMC from what I can find may or may not have a 22nm (talk of it in 2008, but recent articles point to 20nm finfet). Looks like 22nm is alone in the world and will only exist in one place.

Guess it would be believable if and when Samsung does it, but I have my doubts. Interesting how both GF and Samsung are doing a lot of arm-flapping "Lookit me!! LOOKIT ME!" recently with marketing ads for 14nm, whereas Intel has the only working CPU already taped out at that node.. Maybe it was that announcement at IDF that triggered their nervousness and got the marketing dept fired up 😛.

Plus I suspect that 1.24BN pounds (maybe $1.5BN dollars) is way too little for both 20nm and 14nm, when Intel is spending what - $7BN?

 

[fazers_on_stun]

Since there were some posts a few pages back about AMD stock, thought people might get a chuckle out of Seeking Alpha, which has back-to-back articles with one stating "stay away from AMD" http://seekingalpha.com/article/882261-amid-cfo-resignation-and-intel-s-guidance-we-d-stay-away-from-amd?source=msn and the other stating "Buy AMD" http://seekingalpha.com/article/882841-cfo-resignation-and-intel-warning-are-irrelevant-buy-amd?source=msn .

Talk about schizoid viewpoints 😛..

 

[davemaster84]

Well we're so damn close to October and still there are no real benchmarks or a price list, they should let us know already 😛

 

[guskline]

I would rather they perfect a process rather than jumping to a new process which if it is GloFoundries then I have my reservations about. The simple fact that Bulldozer was cut down on from initial spec releases is enough to suggest GF was not ready for 32nm let alone 20 and 14nm.

sarinaide: Is you 8150 Bulldozer overclocked?

 

[sarinaide]

sarinaide: Is you 8150 Bulldozer overclocked?

If I bench then yes but mostly run at stock.

 

[mayankleoboy1]

BD launch confirmed to October2 :
http://semiaccurate.com/2012/09/24/amd-announces-launch-date-of-2nd-gen-apus-2/


also the birthday of this man :

 

[gamerk316]

^^ Why? That didn't work so well last time...

Again, I'm expecting about an 8% nominal performance increase over BD. I'm sure some here will cherry-pick a single bench that shows a 20% increase, but I expect a <10% nominal performance gain.

 

[de5_Roy]

may be.... instead of having low yields early on (credit(!) to glofo) in a july launch, amd may have stockpiled enough apus, and then there is windows 8...
i am interested to see how amd's version of xmp turns out. apus might benefit from them more than the cpus.

 

[sarinaide]

I have officially seen a genuine FX pre release engineering sample, but have gotten to test AMD Radeon RAM with BEMP and it is a little exciting.

 

[cgner]

^^ Why? That didn't work so well last time...

Again, I'm expecting about an 8% nominal performance increase over BD. I'm sure some here will cherry-pick a single bench that shows a 20% increase, but I expect a <10% nominal performance gain.

All I expect is higher clocks and MAYBE 3-5% per-clock performance. They just tighten up existing archeh

 

[noob2222]

Guess it would be believable if and when Samsung does it, but I have my doubts. Interesting how both GF and Samsung are doing a lot of arm-flapping "Lookit me!! LOOKIT ME!" recently with marketing ads for 14nm, whereas Intel has the only working CPU already taped out at that node.. Maybe it was that announcement at IDF that triggered their nervousness and got the marketing dept fired up 😛.

Plus I suspect that 1.24BN pounds (maybe $1.5BN dollars) is way too little for both 20nm and 14nm, when Intel is spending what - $7BN?

read the samsung pr, not the stupid hexus crap.

Its 1.9B for a new fab LINE (not an entire plant) being built on samsungs already owned $33B + property in south korea. If you already own the property, pretty much all your doing is buying the equipment. most likely its a small line for 20 and 14nm early production runs, obviously the test phase is already been done.

funny you say samsung is arm flapping, they did one PR saying they were going to spend 1.9B on a 20 & 14nm production line, and all the other tech websites parroted it, half of them not even reading what was actually said (1.5B instead of 1.9B) and adding in their own "half-baked theory" of samsung skipping 28nm and 22nm because it "only" mentioned 20 & 14.

I don't see samsung's marketing team making pretty little slides about their roadmap, no, they are just doing it. In fact they didn't even show their 14nm wafer until after IBM showed theirs.

The only arm flapping is from GF, they are the ones losing business from AMD. GF is the only one that needs to arm flap, in order to get ARM vedors back.

Samsung doens't need any "marketing stunts", they are getting all the business they want from companies that TSMC & GF can't keep up with (Apple, and soon Qualcomm going to Samsung for 28nm chips) And I don't see TSMC being able to keep up with all of Apple's needs, especially considering Qualcomm is leaving TSMC, some of it sure, but not 100% without alienating the rest of tsmc's customers.

http://thedroidguy.com/2012/08/apple-and-qualcomm-cannot-take-exclusive-chip-manufacturing-from-tsmc/

 

[cgner]

pwnin troll yo