Soldato
- Joined
- 2 Jan 2012
- Posts
- 12,453
- Location
- UK.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD To Tape Out 14nm and 20nm Process Chips In Next 2 Quarters
- Thread starter Boomstick777
- Start date
Permabanned
- Joined
- 9 Aug 2009
- Posts
- 12,234
- Location
- UK
in b4 delays
I thought AMD already had 20nm CPUs (or atleast APUs) in the works with GF?.
As GF is over a year late on delivering 20nm I can't see 14nm appearing from them that soon.
"Both the 20nm and 14nm chips will enter production in 2014. 20nm production with TSMC is expected to start in February 2014 while 14nm production with Global Foundries will also take place before the first half of 2014."
From most other sources Apple has the first round of 20nm production from TSMC and nVidia and Qualcomm share allocation on the next round (approx. - production starting March to June 2014) so doesn't fit with any thing else I've seen.
As GF is over a year late on delivering 20nm I can't see 14nm appearing from them that soon.
"Both the 20nm and 14nm chips will enter production in 2014. 20nm production with TSMC is expected to start in February 2014 while 14nm production with Global Foundries will also take place before the first half of 2014."
From most other sources Apple has the first round of 20nm production from TSMC and nVidia and Qualcomm share allocation on the next round (approx. - production starting March to June 2014) so doesn't fit with any thing else I've seen.
Last edited:
That's why I'm still on Sandybridge...I can't justify the cost for the +10% (and then lose on overclocking!)
This. Come on FX 8 core steamroller!!!
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
14nm at GloFo and I think it's 16nm at TSMC should come exceptionally quickly after 20nm at both companies.
While most places spend billions on R&D, then billions on equipment to get to a new process, 20/16 or 20/14nm were developed together. They will use almost entirely the same equipment, there is no refitting labs entirely or spending years and billions on new lithography methods. The 16/14nm from the two companies are keeping the same metal gate pitch which is basically 20nm(and afaik it will be noticeably smaller than Intel's 22nm gate pitch... which is much more important than most other things) with smaller 16/14nm features on other layers. So it will improve costs, yields, power usage and likely frequencies but with very little reworking of the fab to get going. Both companies have said for quite a long time that their 14/16nm variants are due within the year after then 20nm and there isn't much reason to think it isn't true at this stage.
Also I've heard that GloFo is already producing 20nm A57's for an unnamed customer, risk production is going on with low yields, there is usually a customer or two doing so. They'll have fab equipment going, not make certain of yields, often offer per good chip pricing(but still very expensive this early) and then they don't just pour money down the drain as they tweak the process but use a customer to do it. Apple couldn't rely on the output of risk production so would be fairly useless to them, but a smaller customer who would take anything they can get and release new products, particularly something using tiny tiny chips would be useful. If Apple can't count on X millions of chips per month then they can't launch a product, most companies don't have such high requirements.
A huge number of customers are skipping 20nm entirely because 14/16nm is where ultimately the process was designed to go with 20nm not being brilliant(nor bad) and because due to the extremely short time period between "processes" the cost of taping out at 20nm for most customers then almost immediately working on 16/14nm tape outs would be cost prohibitive. The few parts this won't be true for are for very high margin sales, so Apple selling their own devices that outmatch what anyone else can make, and gpu's which are so die size limited any drop instantly means improvement.... there are few to no other customers who required 350mm2+ chips at foundries outside gpus. Also makes gpu makers unsuitable for risk production as Nvidia well know. They tried to make runs of 480gtx's with risk production, but that is why it turned out they got extremely lucky on the good die deal which then TSMC refused to do again for them
TSMC/GloFo would do risk production for tiny chips as you'll still get enough off a wafer, when its a 500mm2 part and you're talking 2-3 good dies(if lucky) then it's just not worth it at all.
In terms of AMD taping out products, the link doesn't actually have a quote saying anything other than AMD working on 20/14nm products.
I would expect they would have started work on 20nm gpu's and 20nm apu's and 20nm cpu's quite some time ago, but they will have more products to do as well. I would think ps4/x1 shrinks will be far quicker than previous consoles(particularly with rumours of Sony thinking of a MUCH shorter generation for consoles this year, they want the PS4 cheaper asap, sales growing at lower price and then seemingly a new upgraded box in probably at max half the time the PS3 was around).
Pretty much the second 20nm parts are being shipped to AMD they'll be working on taping out 14/16nm products.
Last edited:
From what I hear it isn't going to happen on AM3+ (maybe I'm just starting to side with the naysayers?). I mean, at one point AMD did pretty much promise it but it seems they're moving away from just a CPU now.
I have a feeling Steamroller may show up in the guise of a really low clocked 8 core APU using the next architecture they have lined up, and, will mimic the consoles so well that it will be all you need for your gaming.
I've been doing loads of reading up on AMD's HSA (Heterogeneous system architecture) and from the sound of it the next APUs will support it fully..
Very briefly - HSA allows the hardware to hand shake at a hardware level meaning ultra low latency and massively improved speed.
All very exciting, but not so good for AM3+
- Joined
- 2 Jan 2012
- Posts
- 12,453
- Location
- UK.
Soldato
- Joined
- 21 Jul 2005
- Posts
- 21,123
- Location
- Officially least sunny location -Ronskistats
Well with a bit of logical work it's not hard to figure out. SR apparently gives the same boost over PD that PD got over BD. 15% or so.
So just add 15% on top of your 8320 benchmarks and you should have a rough guide
IE it'll sit just below the 4770k, yet cost around £80 less.
From what I hear it isn't going to happen on AM3+ (maybe I'm just starting to side with the naysayers?). I mean, at one point AMD did pretty much promise it but it seems they're moving away from just a CPU now.
I have a feeling Steamroller may show up in the guise of a really low clocked 8 core APU using the next architecture they have lined up, and, will mimic the consoles so well that it will be all you need for your gaming.
I've been doing loads of reading up on AMD's HSA (Heterogeneous system architecture) and from the sound of it the next APUs will support it fully..
Very briefly - HSA allows the hardware to hand shake at a hardware level meaning ultra low latency and massively improved speed.
All very exciting, but not so good for AM3+
8 Core APU + HSA + iGPU fitted in an ITX = Definite purchase!
Soldato
- Joined
- 21 Jul 2005
- Posts
- 21,123
- Location
- Officially least sunny location -Ronskistats
I think it's definitely the way forward but probably under 2 years from being there.
They should unify the sockets and stick to 1 for consumers. Kill of AM3+ and expand on FM2 or FM2+ or whatever they call it and on that socket alone offer Quad core APUs SixCore APUs and pure 8 core CPUs. Would really love to see what a 14nm or even a 20nm CPU from AMD would be capable off.
14nm at GloFo and I think it's 16nm at TSMC should come exceptionally quickly after 20nm at both companies.
While most places spend billions on R&D, then billions on equipment to get to a new process, 20/16 or 20/14nm were developed together. They will use almost entirely the same equipment, there is no refitting labs entirely or spending years and billions on new lithography methods. The 16/14nm from the two companies are keeping the same metal gate pitch which is basically 20nm(and afaik it will be noticeably smaller than Intel's 22nm gate pitch... which is much more important than most other things) with smaller 16/14nm features on other layers. So it will improve costs, yields, power usage and likely frequencies but with very little reworking of the fab to get going. Both companies have said for quite a long time that their 14/16nm variants are due within the year after then 20nm and there isn't much reason to think it isn't true at this stage.
Also I've heard that GloFo is already producing 20nm A57's for an unnamed customer, risk production is going on with low yields, there is usually a customer or two doing so. They'll have fab equipment going, not make certain of yields, often offer per good chip pricing(but still very expensive this early) and then they don't just pour money down the drain as they tweak the process but use a customer to do it. Apple couldn't rely on the output of risk production so would be fairly useless to them, but a smaller customer who would take anything they can get and release new products, particularly something using tiny tiny chips would be useful. If Apple can't count on X millions of chips per month then they can't launch a product, most companies don't have such high requirements.
A huge number of customers are skipping 20nm entirely because 14/16nm is where ultimately the process was designed to go with 20nm not being brilliant(nor bad) and because due to the extremely short time period between "processes" the cost of taping out at 20nm for most customers then almost immediately working on 16/14nm tape outs would be cost prohibitive. The few parts this won't be true for are for very high margin sales, so Apple selling their own devices that outmatch what anyone else can make, and gpu's which are so die size limited any drop instantly means improvement.... there are few to no other customers who required 350mm2+ chips at foundries outside gpus. Also makes gpu makers unsuitable for risk production as Nvidia well know. They tried to make runs of 480gtx's with risk production, but that is why it turned out they got extremely lucky on the good die deal which then TSMC refused to do again for them
TSMC/GloFo would do risk production for tiny chips as you'll still get enough off a wafer, when its a 500mm2 part and you're talking 2-3 good dies(if lucky) then it's just not worth it at all.
In terms of AMD taping out products, the link doesn't actually have a quote saying anything other than AMD working on 20/14nm products.
I would expect they would have started work on 20nm gpu's and 20nm apu's and 20nm cpu's quite some time ago, but they will have more products to do as well. I would think ps4/x1 shrinks will be far quicker than previous consoles(particularly with rumours of Sony thinking of a MUCH shorter generation for consoles this year, they want the PS4 cheaper asap, sales growing at lower price and then seemingly a new upgraded box in probably at max half the time the PS3 was around).
Pretty much the second 20nm parts are being shipped to AMD they'll be working on taping out 14/16nm products.
Yeah but all other sources of information points to 16/14nm production being towards the end of that year later - not just 1 quarter after 20nm production i.e. nVidia seems to have pushed back the full fat Maxwell to late 2014 so as to get it onto 16nm (probably) or 14nm.
Lol. Epic fail.
How mature. Though it's hard to take anyone serious with numbers in their name.
Associate
- Joined
- 31 Oct 2012
- Posts
- 2,243
- Location
- Edinburgh
To be fair in a thread opening with a quote about Global Foundries to think GF meant GeForce when that would make the sentence you picked on totally illogical is a bit of an oops moment. Yep, we all do them, but saying 'Epic fail' at it isn't the worst
Be interesting to see when these die shrinks actually appear on desktop parts
Be interesting to see when these die shrinks actually appear on desktop parts