[Guru3d]Samsung to fab 14nm GPUs for AMD

[4K_shmoorK]

Not sure if this was posted in another thread or subsection yet. Interesting.

Not Global foundries, not TSMC .. but Samsung will be fabbing next years Arctic Islands family of GPUs. Since it's Samsung it'll be a 14nm FinFET fab, opposed to Nvidia who will be using 16nm for their Pascal architecture at TSMC.

Rumor is that not just the GPUs will be fabbed by Samsung, the APU and processors as well or at least part of the lineup.

According to industries on the 21st, Samsung Electronics’ System LSI Foundry Business Department will start mass-producing AMD’s new GPU ‘Greenland (development code name)’ along with Global Foundry (GF) starting from 2nd quarter of 2016. Greenland will be produced from Gen. 2 14-nano FinFET LPP (Low Power Plus) processing and its electricity efficiency per watt is 2 times higher compared to 28-nano GPU (code name: Fiji) that is currently being sold in markets. Proportion of production supplies is very fluid as AMD will regulate proportion between 2 businesses according to many conditions such as yield and others.
AMD had been asking TSMC from Taiwan to be in charge of GPU production but it decided to cut contract with TSMC with 28-nano being the last production after issues with yield and instable supplies had continued. Its alternatives were Samsung Electronics and GF. AMD, Samsung Electronics, and GF had finalized on such decision when Samsung Electronics and GF were signing on contracts on common license for 14-nano processing last year, Afterwards AMD decided to pass over 20-nano processing and go straight to 14-nano processing.


“Because Samsung Electronics and GF have same IP for 14-nano processing, chips that are designed by AMD will all be produced at both factories.” said a person who is familiar with this industry. “If products are produced from both factories, AMD won’t have to worry about a problem regarding lack of supplies.”
After starting production of Greenland, AMD will put out its new CPU ‘Zen’ right after. This product is also produced from 14-nano LPP processing and will be produced by Samsung Electronics and GF just like GPU.

Sales from toll manufacturing for AMD will be regularized starting from 3rd quarter of 2016. It is expected that considerable amount of sales will be made as AMD’s yearly sales in 2014 were $5.56 billion (about 6.4 trillion KRW).
It is heard that Samsung Electronics’ Foundry Business Department is very encouraged after it took over customers from TSMC, which is Samsung Electronics’ biggest competitor, one after the other. It is very meaningful that Samsung Electronics was able to expand its foundry products that were limited to mobile products to GPU and CPU for PCs. Foundry Business Department evaluated 14-nano FinFET processing as huge success. It brought back most of supplies of Apple’s application processor (AP) from TSMC and has taken charge of producing Qualcomm’s Snapdragon 820 AP. After succeeding in making surplus in 2nd quarter, System LSI Business Department increased amount of surplus in 3rd quarter. It is expected that its performance will increase even more in 4th quarter due to production of Qualcomm’s chips and others.
Director Shim Sang-pil of Austin Semiconductor Corporate (General Executive of System LSI Manufacturing Center) was awarded for proud Samsung Person for 2015 and was promoted to executive director 2 years early because he received credits for commercialization of 14-nano FinFET processing foundry. VP Michael Raiford, who is in charge of technologies, of Samsung Electronics Austin Semiconductor Corporate was also promoted to director because he received credits for mass-producing 14-nano System LSI products at the right time.

Samsung Electronics will speed up process of development of 10-nano processing that will be regularized in 2017. It will make an announcement of its thesis based on successful development of S-RAM that will be loaded into 10-Nano Logic at ISSCC Society that will be held in January of 2016 and do early promotion about it.


Regarding how Samsung Electronics was able to attract AMD as its customer, a person who is familiar with Samsung Electronics said that Samsung Electronics cannot talk about any details regarding its customers.

 

[railven]

This is good news. GloFlo was a suicide march.

 

[turtile]

This makes sense but I wonder how different 16nm and 14 will perform. If AMD went 16 for the GPU and 14 for the CPU, they'd still have extra work for the CPU on 16 for the APU. Making both the same process is a lot easier.

 

[Mondozei]

So if Greenland, the flagship GPU, will be on Samsung's 14 nm, will TSMC make the non-flagships?

I mean, since the story explictly mentions Greenland instead of the whole of the Artic Islands family, it does make you wonder why they only mentioned Greeland. I'd imagine it's quite costly, but I'd love for someone more knowledgable than me to jump in and enlighten us.

I mean, dual-sourcing is when you have two different companies manufacturing the same product. In the case of Apple, it is the A9.
As I understood it, dual-sourcing is quite expensive.

But what if you have different companies doing different products in the same GPU family? Is that still considered "dual-sourcing"? Or is it sufficiently different as to not be more expensive?

To my mind, it would seem AMD would either go all in on TSMC or all in on Samsung, and not divide the division of labor. It may be possible that AMD did decide to go all in on Samsung, but I'm just curious how the potential cost structure would be affected if they decided to do non-flagship GPUs on TSMC.

 

[Mondozei]

BTW, the latest stories point to Samsung starting mass production in Q2 2016. Q2 extends to the end of June, and you have to add a month or two for there to be adequate supply and give time to transport sufficient GPUs across the world in the tankers.

So it does look like the back to school season is the time when we'll see the hard, actually available to consumers, launch of the next-gen GPUs. So around september or thereabouts. More or less 9 months from now. And that assumes that AMD actually has learned their lesson from the past and have stocked up enough GPUs.

If they haven't, most buyers can count on an extra month or two before supply becomes normal. Meaning we're talking almost a full year before you'd be able to get a new GPU at normal prices(MSRP, that is) without hitting F5 on Newegg sixty times a day to see if there's new supply.

If we see the cut-down versions from both NV/AMD next year, it might just be better to wait until Spring of 2017 when you'd see the next full-fat flagship launch and by then I'm guessing supply will be much better and the 14/16 nm maturity and yields will be better, too. At least if you have an upper-mid range GPU today.

 

[Magee_MC]

So if Greenland, the flagship GPU, will be on Samsung's 14 nm, will TSMC make the non-flagships?

I mean, since the story explictly mentions Greenland instead of the whole of the Artic Islands family, it does make you wonder why they only mentioned Greeland. I'd imagine it's quite costly, but I'd love for someone more knowledgable than me to jump in and enlighten us.

I mean, dual-sourcing is when you have two different companies manufacturing the same product. In the case of Apple, it is the A9.
As I understood it, dual-sourcing is quite expensive.

But what if you have different companies doing different products in the same GPU family? Is that still considered "dual-sourcing"? Or is it sufficiently different as to not be more expensive?

To my mind, it would seem AMD would either go all in on TSMC or all in on Samsung, and not divide the division of labor. It may be possible that AMD did decide to go all in on Samsung, but I'm just curious how the potential cost structure would be affected if they decided to do non-flagship GPUs on TSMC.

As I understand it the Samsung and GF 14nm process is the same, so that might mitigate some of the usual cost differences in dual-sourcing from two different companies with two different processes.

 

[kawi6rr]

Good news I guess next year will be when I start making some upgrades.

 

[RussianSensation]

This makes sense but I wonder how different 16nm and 14 will perform.

Impossible to say at this point since Arctic Island is a new/heavily revised GCN architecture. That means perf/watt improvements will be intertwined with any advantages 14nm may have over 16nm.

Not necessarily directly comparable go high-end GPUs, but A9 showed a significant difference between Samsung's 14nm and TSMC's 16nm nodes.

Samsung's A9 chip was 3.5-10.8% more efficient, and TSMC's A9 chip was 8.85% larger.

http://www.tomshardware.com/news/iphone-6s-a9-samsung-vs-tsmc,30306.html

That means under SoC applications, Samsung's 14nm is better than TSMC's 16nm.

BTW, the latest stories point to Samsung starting mass production in Q2 2016. Q2 extends to the end of June, and you have to add a month or two for there to be adequate supply and give time to transport sufficient GPUs across the world in the tankers.

So it does look like the back to school season is the time when we'll see the hard, actually available to consumers, launch of the next-gen GPUs. So around september or thereabouts. More or less 9 months from now. And that assumes that AMD actually has learned their lesson from the past and have stocked up enough GPUs.

If so, AMD is playing a dangerous game, because it would mean outside of Fiji they will be competing with 2012-2013 GPU designs for 3/4 of 2016 and still have no viable mobile dGPU strategy. With the competitor aiming for 2X the perf/watt over Maxwell, if NV pounces early Q2/3 2016, it won't take a lot of effort to wipe out the entire AMD product stack with 16nm Pascal chips. Unlike NV owners who would wait 6-9 months for NV cards, AMD cannot afford to be MIA for most of 2016 while relying on Tonga, Hawaii and Fiji.

If we see the cut-down versions from both NV/AMD next year, it might just be better to wait until Spring of 2017 when you'd see the next full-fat flagship launch and by then I'm guessing supply will be much better and the 14/16 nm maturity and yields will be better, too. At least if you have an upper-mid range GPU today.

If you shift the focus on the software driving GPU upgrades, then it makes GPU upgrades for the sake of upgrading more pointless. If you have a 980 OC and it maxes out all your games, no point in upgrading unless you want to. :thumbsup: Then you could easily wait to Sprint 2017, etc.

Good news I guess next year will be when I start making some upgrades.

The market thought so. AMD stock is up almost 10% with almost 25M shares exchanged.

^"citing unnamed sources, as per reported by South Korea’s Electronic Times reported."

 

[Arachnotronic]

That means under SoC applications, Samsung's 14nm is better than TSMC's 16nm.

Um, no. Tom's Hardware seems to be the only site that showed Samsung 14nm A9 more efficient than TSMC A9. Also note that Apple chose TSMC 16nm exclusively for the higher performance A9X (larger die, higher clock speeds) and has chosen TSMC 16nm exclusively for the A10.

The Samsung chip is a little denser than the TSMC one, but word in the industry is that the TSMC 16nm FF+ version of the A9 was basically a last-minute port, so it is relatively un-optimized compared to the 14nm version.

 

[RussianSensation]

Um, no. Tom's Hardware seems to be the only site that showed Samsung 14nm A9 more efficient than TSMC A9. Also note that Apple chose TSMC 16nm exclusively for the higher performance A9X (larger die, higher clock speeds) and has chosen TSMC 16nm exclusively for the A10.

There is probably a lot more to it than just technical characteristics but also cost and yields. Snapdragon 820 is also built on Samsung's 14nm FinFET. This suggests TSMC's and Sammy's processes have their own advantages/disadvantages. How that will translate to GPUs is too hard to predict. Since Samsung is trying to attract many customers from TSMC, one of the most effective ways to do so is with lower price/volume priority. If AMD can be more effective in price/performance due to lower manufacturing costs and have sufficient supply, it may make sense to switch out of TSMC to at least try out another fab. This is still just a rumour from "unnamed sources."

 

[jpiniero]

Considering the WSA, doing everything on Samsung's process makes a ton of sense rather than doing GPUs at TSMC and CPUs at GloFo.

 

[boozzer]

I don't know much about this. Q: is this good news or just normal news?

 

[Tuna-Fish]

A: It's hard to tell, because there is not enough public information.

 

[CakeMonster]

Isn't "LPP" something that would have had the whole crowd here very suspicious 1-2 years ago? Not a process typically expected for high performance GPUs.

 

[RussianSensation]

Isn't "LPP" something that would have had the whole crowd here very suspicious 1-2 years ago? Not a process typically expected for high performance GPUs.

No, because for 14nm node, the old HPP term is not relevant in this context. For, 14nm GloFo/Samsung changed the terminology to mean this:

14nm FinFET Technology
14LPE – Early time-to-market version with area and power benefits for mobility applications
14LPP – Enhanced version with higher performance and lower power; a full platform offering with MPW, IP enablement and wide application coverage

This video briefly explains how the distinction of 28nm HPP vs. SLP is no longer relevant for 14nm. 14nm high performance is now called LPP. This 14nm LPP node is positioned for usage in complex large die CPU, GPU designs.

 

[CakeMonster]

Yeah, I kind of figured. I seem to remember the talk being about moving beyond 28nm since 2012 was about the lack of "high power" processes.. That there were none suitable for large GPUs in the pipeline (which I guess has turned out to be true).

 

[RussianSensation]

Yeah, I kind of figured. I seem to remember the talk being about moving beyond 28nm since 2012 was about the lack of "high power" processes.. That there were none suitable for large GPUs in the pipeline (which I guess has turned out to be true).

Ya, the marketing is confusing. I agree with you.

 

[mysticjbyrd]

AMD Greenland GPU Coming Next Summer On 14nm – To Enter Mass Production In Q2 2016

Read more: http://wccftech.com/amd-greenland-14nm-production-q2-2016/#ixzz3v6PoubGy

 

[railven]

I just got my Christmas Bonus. I'm ready!!!

 

[crisium]

August-September a good guess for availability with that schedule?