Seeking Alpha


View as an RSS Feed
View TechResearch's Comments BY TICKER:
Latest  |  Highest rated
  • Do Leaked AMD Benchmarks Pose A Threat To Nvidia? [View article]
    This is my take on the rumours. I am quite sure that the chiphell benchmarks are of R9 380X. My view is that R9 380X is a 3072 sp GCN 2.0 chip with 4GB HBM and 512 GB/s. I am also thinking that it could be a salvage SKU from the flagship GCN 2.0 chip which is likely to be > 500sq mm die size. According to the charts R9 380X is 30% faster than R9 290X. The R9 390X is a 4096 sp GCN 2.0 chip with 4GB and 512 GB/s. Hynix is in production of 1 Gbps HBM with 128 GB/s bandwidth (4 Hi) already and the timing is perfect for a Q1 2015 launch. With 45% more stream processors than R9 290X (2816), GCN architectural enhancements (over and above the ones found in Tonga aka R9 285), state of the art high bandwidth memory (HBM) system I am betting that the R9 390X perf is > 50% faster (closer to 60%) than R9 290X .

    R9 290X was 35% faster than R9 280X with 37.5% more stream processors (sp) with much smaller architectural enhancements (GCN 1.1 vs GCN 1.0)

    Its not entirely unimaginable for AMD to improve perf/sp, perf/sq mm, perf/transistor and perf/watt with a significantly improved GCN 2.0 architecture. Add to it the massive bandwidth from HBM and we are likely to see a > 50% performance increase over R9 290X. I can see a 25% improvement over the Jack simple (3072 sp) with the fully enabled 4096 sp SKU. That puts perf in the range of 82 (very close to R9 290X CF in the chart) .

    As for manufacturing process I am betting AMD has gone with GF 28SHP. There are multiple reasons for that. GF 28SHP is a high performance process. The process is relatively mature with good yields. AMD is already manufacturing Kaveri, semi-custom game console chips and GPU products at GF 28SHP.

    "Joe Moore - Morgan Stanley
    Okay great and then secondly when you talked about the new GlobalFoundries agreement, you have talked about moving more GPUs and consoles to GlobalFoundries can you give a status update on what’s happening there?

    Devinder Kumar - SVP and CFO
    As of right now GlobalFoundries is shipping both the GPU products and semi-custom game console products. "

    Moreover AMD will not face any supply shortage should it deliver a GPU which beats the competition clearly. GF 28SHP is also significantly better than TSMC 28nm.

    "AMD claims a 19% reduction in core leakage/static current for Puma+ compared to Jaguar at 1.2V, and a 38% reduction for the GPU. The drop in leakage directly contributes to a substantially lower power profile for Beema and Mullins."

    Beema/Mullins are manufactured at GF 28SHP and have lesser power draw wrt Kabini/Temash which are manufactured at TSMC 28HP. Finally if you look at the AMD's 2.5D stacking R&D it goes all the way back to 2011 with GF, Amkor and Hynix as partners.
    Dec 2, 2014. 12:25 PM | 3 Likes Like |Link to Comment
  • AMD Finally Delivers Something Crucial [View article]
    A crucial development which has been revealed is that Carrizo-L based on Puma+ cores is built at GF 28SHP. So the previous plans revealed in May 2014 for AMD Puma+ based and ARM A57 based SoCs to be built at 20nm seems to have been scrapped.

    What is confusing is why did AMD announce 20nm SoCs in May 2014 if it did not have confidence that it could deliver on the roadmap ? Moreover since Apple is shipping A8 and A8X in high volume why did AMD skip 20nm. Is TSMC 20nm capacity a constraint or is AMD not happy with the state of the process and the cost/yield factors and the half node like power/perf gains ? Also what does this mean for ARM A57 SoC in 2015. Are they also made at GF 28SHP ?

    There is not much detail about both the Puma+ core and the GPU in Carrizo-L. With a single package and dual channel memory support AMD has the opportunity to go for a larger GPU of atleast 256 sp (2 compute units).

    I am more concerned about the 2016 APU products using Zen (x86-64) and K12 (ARMv8) architectures. If AMD cannot deliver next gen CPU core based FINFET products using either TSMC 16FF+ or Samsung 14 LPE process in 2016 then its a death knell as Intel will have second gen 14nm products (based on the new microarchitecture Skylake) and 10nm die shrink products out in 2017. This would put any hopes of AMD's recovery down and out for sure.
    Nov 21, 2014. 10:25 AM | 2 Likes Like |Link to Comment
  • Intel: More Good News About The Mobile Losses [View article]
    Somebody's going to run out of fuel, and TSMC appears the closest to empty.

    Thats what you are hoping or praying as a Intel pumper. But sorry the reality is different. TSMC has already grown 23% Y-o-Y for the first 10 months of 2014. My predictions are simple. ARMv8 takes marketshare from Intel in servers, notebooks and desktops. Microsoft will support their software stack on ARM. Foundries and ARM ecosystem benefits at Intel's expense. I expect this to play out over the next 5 - 6 years. :-)
    Nov 17, 2014. 10:20 PM | Likes Like |Link to Comment
  • Intel: More Good News About The Mobile Losses [View article]
    "With a built in $50 billion of business, Intel could afford to spend $12 billion a year on capex forever without pain. TSMC can do that for a while, but not much longer."

    What makes you claim that. You believe TSMC will run out of business when to the contrary they are growing much more rapidly than any other foundry or IDM. You also believe that only Intel will be able to keep up with leading edge process nodes R&D and capex. Boy are you hallucinating or praying or doing both. You obviously did not believe TSMC will get to 16FF+. Guess what you were wrong. You believe TSMC cannot keep up the pace. You are again woefully wrong.

    You did not even respond to the articles which state MS has a windows server version running on ARM. Its clear now that you don't want to see any competition to Intel and you don't want your beloved company to start losing marketshare in its most profitable business segment (servers). But like it or not it is going to happen. The ARM ecosystem will profit at the expense of Intel. You will see it from 2016 and the trend will pick pace after that.

    "Tech I checked your profile. Apparently you are an Intel hater and an AMD lover. I can't do anything about that. "

    I can't do anything about the fact that you are an Intel lover alongwith your buddy cincinnatus :-)
    Nov 17, 2014. 10:00 PM | Likes Like |Link to Comment
  • Intel's Server Business Makes The Stock A Strong Buy [View article]
    Its not about Intel coming up with an inferior product or having the lead in manufacturing technology. Until now ARM products have not been designed on leading edge nodes. 2016 will bring a contest between Intel 14nm and TSMC 16FF+. TSMC is closing the gap even further at 10nm. But the bigger point is Intel has almost the entire server market that its hard to see ARM vendors not carve a nice marketshare. like ta152h says when you are close to 100% market share you have everything to lose. For the rest like AMD, Applied Micro there is nothing to lose its all upside. Also given how IBM has closely aligned with Nvidia and Google to form OpenPower and given the Power 9 /Volta/Nvidia Nvlink technology with which they have secured the next 2 biggest supercomputer contracts from US DoE its clear to see that everyone wants an alternative to Intel. The industry will make sure that Intel is not allowed to dominate the server space.
    Nov 17, 2014. 09:51 PM | Likes Like |Link to Comment
  • Nvidia's Tough 2015: TSMC Delays 16 Nm Production, AMD Introduces New Boards, And Intel Tries To Gain Traction With IGpus [View article]

    "Many of the comments in the EE Times article tell the real story. The analyst quotes don't."

    oh is it. Maybe the commenters should try and write articles to refute the articles if they so strongly believe. But most of those commenters are trolls like you. :-)

    "Yield is a function of die size, so for an analyst to just say TSMC's 20nm is yielding 80% is telling you that analyst has no clue about the industry he's covering"

    Obviously the yields are for chips like Apple A8 which are in high volume production. TSMC got to 50% yields on A8 in Mar, 75% by Q3 and now its 80%. Anyway you are a person who wants to believe that TSMC is not doing well when the facts are the opposite. Thats what happens when you are a Intel pumper.
    Nov 17, 2014. 01:51 PM | Likes Like |Link to Comment
  • Nvidia's Tough 2015: TSMC Delays 16 Nm Production, AMD Introduces New Boards, And Intel Tries To Gain Traction With IGpus [View article]
    "I'll agree with you to the extent there are lies being told."

    everybody knows its you who is lying. so stop acting smart. Between a Intel pumper and troll like you and TSMC everyone knows whom to believe.

    "By first-half 2016 TSMC *might* be ramping production on a process that's the same node that Intel had retail product launching in April 2012."

    What a joke. TSMC 16FF+ is a 15% reduction in die size wrt 20 SOC. TSMC 16FF+ will have a contacted gate length of 77 - 78nm, similar to Samsung's 78nm. Intel has a contacted gate length of 70nm. The M1 metal pitch for TSMC and Samsung is 64nm while for Intel is 52nm. But the key here is TSMC's choices have led to a solid 20nm ramp and the yield learning from 20SOC is helping TSMC with their 16FF+ SRAM ramp. Intel on the other hand is having trouble ramping 14nm in any meaningful sense of volume with only 4.5W tablet chips launched. Where the heck is the notebook and desktop chips which Intel used to launch in high volume as they did with ivybridge at 22nm ? So stop trolling.
    Nov 17, 2014. 01:42 PM | 1 Like Like |Link to Comment
  • Intel's Server Business Makes The Stock A Strong Buy [View article]
    I agree with what you have said. Intel has the challenge of maintaining their near complete dominance of servers against the ARM assault and an AMD which could deliver competitive x86-64 and ARMv8 architectures in 2016. But still the real war is just beginning and the results will be felt starting from 2016 and later when we see FINFET based server chips from AMD, Applied Micro and others. I also see a resurgence in POWER with IBM opening up the Power ISA to OpenPower foundation and strong support from Nvidia, Google, Tyan, Mellanox, Samsung and others.

    IBM and Nvidia have already won the deal for building the next supercomputers for US DoE

    Finally the tipping point will happen sometime in the next 2 years with Microsoft shipping Windows server for ARM.
    Nov 17, 2014. 01:31 PM | Likes Like |Link to Comment
  • Intel: More Good News About The Mobile Losses [View article]
    There are so many mistakes in your post. Firstly Apple is a fabless design company just as Qualcomm or Nvidia. The difference is Apple only designs for its own products and gets it manufactured by TSMC/Samsung while Qualcomm and Nvidia design and sell their foundry manufactured SOCs to phone and tablet manufacturers. I agree that Apple has a dominant share in the high end smartphone space and makes the majority of the smartphone industry profits but still Qualcomm is a bigger customer than Apple to the foundries as they sell across the price stack and dominate the Android high end smartphone space. The rapid growth in the smartphone market is in the developing countries such as China, India where price is a crucial factor. The majority of the volume growth is in low and mid range phones. I am in India where smartphones like Xaiomi Redmi 1S (Snapdragon 400) and Moto E (Snapdragon 200) are very popular and sell for USD 100. Thats the fastest growing segment. Apple does not have a phone for these rapidly growing segments. So all said and done the threat to Qualcomm is not Apple. I would say its Mediatek and the other Chinese/Taiwanese fabless design companies. Qualcomm will lose a bit of high end market share as Apple iPhone 6 Plus takes market share from flagship Android phablets.

    "Once we make that step, the question turns into one of funding the semiconductor technology. Again, Intel is alone with that much maligned PC and server chip business generating $24 billion each year in operating profits to spend on the battle with TSMC."

    You seem to be under an illusion that TSMC is not profitable and cannot invest in future nodes. Contrary to it TSMC is accelerating 10nm FINFET development.

    Intel's operating profits for 2013 were USD 9.6 billion (page 27) on revenues of USD 52.7 billion

    TSMC reported profits of USD 6.16 billion from revenues of USD 20 billion . page F-5

    TSMC has already reported more than USD 20 billion sales with a 23.5% Y-o-Y growth for first 10 months of 2014. Oct 2014 posted 55.9% growth compared to Oct 2013. let that number sink in a bit.

    TSMC invested USD 9.7 billion in capex for 2013 and a similar amount for 2014. Intel is slightly higher at USD 10.6 and 11 billion for the same years. So contrary to what you think TSMC is doing very well. oh btw TSMC will go into 16FF/16FF+ production by mid - 2015 with high volume shipments of 16FF/16FF+ in Q4 2015 (10% of wafer revenues). Also where the heck is Cherrytrail when Apple is ramping the living daylights out of A8 (iPhone 6) and Snapdragon is starting to ramp S810 this quarter for device availability in late Q1 2015.

    The way I see TSMC is more profitable (profit as a % of revenue) than Intel and is growing at a much faster rate and is spending capex at similar levels. Intel has done nothing to prove they can take market share from Qualcomm, Mediatek and the others. FYI Intel has a 0.2% market share in smartphones.

    Intel picked up a good amount of market share in tablet chips due to their contra revenue scheme but that has only worsened their losses. Moreover retaining that market share position is not going to be easy.

    "The non-Apple tablet AP market leadership position continues to change hands and during Q2 2014 it was Intel’s turn. Strategy Analytics notes that previously six companies held the non-iPad tablet AP leadership position, which continues to be a challenging one to sustain"

    "Yup, this is a high stakes game, worthy of a bet. My bet is on Intel. "

    I am going to wait and see how this plays out over the next 5 years. But I am willing to bet that Intel will not dominate the entire semiconductor space and there will be a very healthy market share for foundries and fabless design companies like Apple, Qulacomm, AMD, Nvidia. As Intel will gain some market share in phones I also believe that Intel's server market share will fall from what it is now today to a much lower share, though still it will be No.1 . Eventually I see ARMv8 based chips competing with x86 across the entire spectrum of devices - phones, tablets, notebooks, desktops and servers. The tipping point will be Microsoft support for ARMv8 . It will happen in time. Did anybody think that Microsoft will bring Office for iPad. But it did happen, right.
    Nov 17, 2014. 12:38 PM | 1 Like Like |Link to Comment
  • Nvidia's Tough 2015: TSMC Delays 16 Nm Production, AMD Introduces New Boards, And Intel Tries To Gain Traction With IGpus [View article]
    "However you completely miss that TSMC's 16nm is really 20nm FinFET."

    stop lying. TSMC 16FF+ brings die size reduction wrt 16FF/20SOC. ARM customers are taping out 16FF+ designs now for production in mid-2015 and launch in late 2015 / early 2016. TSMC expects the vast majority of its customers to go straight to 16FF+ skipping 16FF. ARM talks of 2.5 - 2.6 Ghz A57 clocks at 16FF+ with a power consumption of just 750mw.

    "16FF+ can provide an additional 15% die size reduction compared to 20SoC."

    Hou said that TSMC was able to improve power, performance, and area in this "second generation" FinFET technology for four reasons:

    Learning from 20SoC production has allowed for better process control, and as a result, signoff corners have been tightened so as to reduce the need for over-design
    Device enhancement
    Middle end of line (MEOL) improvements
    Back end of line (BEOL) improvements

    Combine all these factors, Hou said, and a 16FF+ ring oscillator simulation will show a 20% to 23% speed improvement compared to 16FF. More specifically, standard cells show a 16% to 18% speed improvement, memory shows a 17%-19% speed improvement, eFUSE shows a 13% speed improvement, and I/O devices provide a 3% speed improvement. However, the 16FF+ technology significantly reduces I/O device leakage. "

    ""TSMC and ARM® today announced the results from a key FinFET silicon validation of the ARM big.LITTLE™ implementation, using ARM Cortex®-A57 and Cortex-A53 processors on TSMC’s advanced 16nm FinFET (16FF) process technology.

    Silicon results on 16FF show the “big” Cortex-A57 processor achieving 2.3GHz for sustained mobile peak performance, as well as the “LITTLE” Cortex-A53 processor consuming only 75mW for most common workloads. The performance improvements are a result of the collaboration between ARM and TSMC to jointly optimize the 64-bit ARMv8-A processor series on FinFET process technologies and build on the successful tape-out of the Cortex-A57 processor on TSMC’s 16FF process last year.

    Ongoing collaborative efforts are focused on TSMC’s 16FF+ process technology which will deliver an additional 11% gain in performance for the Cortex-A57 at the same power as the 16FF process, along with a further 35% power reduction for the Cortex-A53 when running low-intensity applications. This further increases the dynamic performance range and power savings for big.LITTLE platforms. 16FF+ is scheduled to be delivered by Q4 2014. Early big.LITTLE implementations of Cortex-A57 and Cortex-A53 processors on 16FF+ are supported by ARM POP™ IP technology."

    "Manufacturing at TSMC 16nm FinFET+ provides the highest performance SoCs for premium mobile or enterprise applications. Even while some next generation mobile designs require reduced voltage domains down to 0.7V, you can see that the Cortex-A57 can hit over 2.1GHz (@TT / 85C) at those reduced voltage levels. This high frequency, combined with the Cortex-A57’s capacity to deliver more performance per MHz results in top of the line performance while staying well under the 750mW budget. "

    "Our customers have planned multiple tape-outs this year in 16FF+, and more big.LITTLE based designs to drive next-generation mobile applications! "
    Nov 17, 2014. 05:43 AM | 1 Like Like |Link to Comment
  • Intel's Server Business Makes The Stock A Strong Buy [View article]
    "Could you please point out to me who is suggesting that. I can't find it."

    Andreas must have been hallucinating that this is an article about AMD. So he jumped in and started trolling. he needs professional help to get over his obsession/hatred for AMD.
    Nov 15, 2014. 11:51 PM | 2 Likes Like |Link to Comment
  • Nvidia's Tough 2015: TSMC Delays 16 Nm Production, AMD Introduces New Boards, And Intel Tries To Gain Traction With IGpus [View article]

    "There's no adoption of TSMC's 20nm beyond the Apple A8 in 2014"

    you are flat out lying. QCOM and Xilinx have been shipping 20nm products from Q2 2014.

    "There's no 20nm volume, and there's not the GPUs on 20nm that you claimed would prove a volume ramp."

    You did not reply when i pointed out that Oct 2014 revenue was up 55.9% wrt Oct 2014. The 20nm ramp is the primary reason according to TSMC. So I know you are trolling. As for GPU these are high performance devices and are power constrained. Since 20nm provides half node like power efficiency gains Nvidia and AMD are looking to go straight to 16/14nm FINFET to get the traditional full node power efficiency gains and density gains.

    "Tape-outs that happen in late 2015 mean you won't see validated, finished products until late 2016 or early 2017. It will take 9-12 months from a tape-out to see a product enter production."

    Products are taping out now and in Q1 2015. TSMC has more than 60 customer designs lined up. btw TSMC now expects Q4 2015 revenue from FINFET to be close to 10%. Wafers will go into production in late Q2 and early Q3 2015 with a 1.5 quarter lead time (FINFET processing has higher lead times) and revenue will be realised in Q4 2015. so stop trolling. pathetic Intel pumper. btw I have not invested in stocks. so stop talking crap.

    "We believe 20nm yields are approaching 80%, and 16 nanometer FF+ SRAM yields exceed 90%, improving confidence that FinFET will reach 1% of TSMC's third-quarter 2015 sales, 10% of fourth-quarter 2015 sales, and approaching 20% by the first quarter of 2016," Credit Suisse analyst Randy Abrams said in a Nov. 12 report following the TSMC announcement. TSMC "also may secure Apple for the next iPad in the fourth quarter of 2015 and iPhone 7 orders in 2016."

    "We see signs of key clients re-engaging with TSMC." The company "is also pulling forward FinFET capacity builds." The pact between Samsung and GlobalFoundries "offers little assurance of smooth production" next year. "TSMC's prior view of it having a smaller initial [FinFET] share is no longer valid; its focus on tech production excellence works again."
    Nov 15, 2014. 11:41 PM | 1 Like Like |Link to Comment
  • Intel: More Good News About The Mobile Losses [View article]
    Here is an article on FINFET progress. Apple and Qualcomm are likely to split their order between Samsung and TSMC.

    "We believe 20nm yields are approaching 80%, and 16 nanometer FF+ SRAM yields exceed 90%, improving confidence that FinFET will reach 1% of TSMC's third-quarter 2015 sales, 10% of fourth-quarter 2015 sales, and approaching 20% by the first quarter of 2016," Credit Suisse analyst Randy Abrams said in a Nov. 12 report following the TSMC announcement. TSMC "also may secure Apple for the next iPad in the fourth quarter of 2015 and iPhone 7 orders in 2016.

    TSMC's technology position with its strong 20nm yield ramp and development on 16nm FF+ has resulted in an early and steeper ramp of 16nm FF+, Abrams said."

    "We see signs of key clients re-engaging with TSMC, given its 16nm FinFET (FF) yield ramp-up; TSMC is also pulling forward FF capacity builds. The SEC-GF pact offers little assurance of smooth production in 2015. TSMC’s prior view of it having a smaller initial FF share is no longer valid; its focus on tech production excellence works again."

    In hindsight TSMC's plan to do the transition to 16 FINFET in 2 stages has proven brilliant - First solve the double patterning immersion litho challenge at 20nm. Then solve the FINFET transistor device challenge at 16 FINFET using the 20nm BEOL and changes to MEOL and FEOL. 16FF+ is now expected to ramp in July 2015 and the bulk of TSMC customers are going straight to 16FF+ .
    Nov 15, 2014. 04:27 AM | 2 Likes Like |Link to Comment
  • Intel: More Good News About The Mobile Losses [View article]
    "The question that analysts need to ask on Nov 20, and that Intel needs to answer is, exactly where are you spending nearly $4 billion a year on mobile and communications. "

    Good question. But I don't think you will get a honest answer.But there are a few reasons.
    1. Intel's contra revenue is a large portion of that USD 1 billion loss. Ashraf Eassa concluded that Intel's contra revenue per unit of Baytrail platform is USD 20. So 40 million tablets x 20 = USD 800 million

    2. The other major reason is mobile application processors with integrated baseband have eaten into stand alone mobile baseband sales. Intel had a substantial stand alone baseband business in 2012 which is now close to zero. In fact your article states that Q4 2012 revenue was USD 1 billion while Q3 2014 was USD 1 million. The biggest beneficiary of this trend has been Qualcomm which has a highly integrated AP stack with integrated baseband from top to bottom.

    "Strategy Analytics estimates that stand-alone AP unit share dropped to 21 percent in Q2 2014 compared to 28 percent in Q2 2013 and this can be attributed to increased shipments of low-to-mid range smartphones, which often feature baseband-integrated APs"

    3. Intel has close to 0% market share in smartphone application processors. But continues to invest in integrated apps processor R&D.

    "Intel captured a minor 0.2 percent revenue share in the smartphone applications processor market in 2013.'

    Starting in Q1 2015 Intel can show some revenue with Sophia. The problem is it is built at TSMC 28 HPM and will compete against competing products from Qualcomm and Mediatek. Qualcomm is quite strong with their overall mobile AP stack and OEM relationships. Mediatek is an aggressive competitor and Intel is not going to have it easy. Apple is growing and continues to be out of Intel's reach.

    The foundries (TSMC/Samsung/GF) are gearing up for FINFET production in 2015. The Apple A9 is rumoured to have gone to Samsung 14 FINFET. But TSMC seems to be in a much better situation than Samsung with FINFET yields. Samsung denies yield problems in its latest Q3 2014 earnings call and says its on track to start production by year end with 30% capacity at 14 FINFET by 2015 end, though I am skeptical of Samsung's claims.

    "Nicolas Gaudois - UBS - Analyst
    Yes. Thank you. On the LSI side, could you comment a little bit more in details, please, on 14 - nanometer FinFET preproduction yields and your confidence in having three or more large customers by H2 2015, including Samsung LSI itself. And do you think this is typically the scale we view as necessary to actually improve profitability?

    Unidentified Company Representative
    (Interpreted). As I mentioned during my presentation, the products are already in full function sampling with a customer that's already been secured. And in terms of yield and other plans, we have no problems and everything is on track. We are leveraging the experience and expertise and technology that we have acquired over a long period of time working with 3D elements. And we are able to adopt the new technology FinFET an d also scale down the nodes. And our progress in the 14 - nano space is moving smoothly, as originally scheduled. In terms of the ramp - up, already the wafers for mass production are fed in for the target date of going mass production in -- or at the end of this year. And we believe that the ramp - up will be going on as scheduled in 2015. Our plan is to ramp up very quickly during 2015 so that by the end of 2015 this would account for 30% of our 12 - inch capacity"

    In the low end Samsung and GF are looking at 28 FD-SOI production for cost sensitive devices. 28-FDSOI bring 20nm perf and efficiency at 28nm wafer costs and yields.

    So Intel's products are going to face tough competition. If Intel does not show good progress in the next 2 years then its quite difficult to keep thinking that Intel can reverse these losses.
    Nov 15, 2014. 02:48 AM | 4 Likes Like |Link to Comment
  • Nvidia's Tough 2015: TSMC Delays 16 Nm Production, AMD Introduces New Boards, And Intel Tries To Gain Traction With IGpus [View article]
    you are just a clueless troll. QCOM is ramping S810 production at TSMC in Q4 2014 for availability in devices in H1 2015. As for R9 390X yeah I can say that I was wrong about it being built on a 20nm process.

    "Unlike the rest of the 64-bit Snapdragon family, the 808 and 810 likely won't show up in devices until the first half of 2015 (410 devices will arrive in Q3 2014, while 610/615 will hit in Q4). The 810 will come first (and show up roughly two quarters after the Snapdragon 805, which will show up two quarters after the recently released 801). The 808 will follow shortly thereafter."

    Snapdragon 805 started appearing in devices in July 2014. You can expect S810 in devices in late Q1 or early Q2 2015.

    Samsung Galaxy S6 is expected to be one of the first Android flagship phones with Snapdragon 810. So if you don't know anything just shut up.

    btw did you see TSMC 20nm ramp contributing to the amazing revenue growth in Oct 2014. 55.9% higher revenue wrt Oct 2013 and and 7.9% wrt Sep 2014.

    So next time don't talk trash that TSMC cannot ramp 20nm in high volume and don't speculate that TSMC cannot do 16 FF and 16FF+ production.

    "TSMC (TWSE: 2330, NYSE: TSM) today announced its 16-nanometer FinFET Plus (16FF+) process is now in risk production. This enhanced version of TSMC’s 16FF process operates 40% faster than the company’s planar 20-nanometer system-on-chip (20SoC) process, or consumes 50% less power at the same speed. It offers customers a new level of performance and power optimization targeted at the next generation of high-end mobile, computing, networking, and consumer applications.

    TSMC’s 16nm process offers an extended scaling of advanced SoC designs and is verified to reach speeds of 2.3GHz with ARM’s “big” Cortex®-A57 in high-speed applications while consuming as little as 75mW with the “LITTLE” Cortex-A53 in low-power applications. It is making excellent progress in yield learning, and has achieved the best technology maturity at the same corresponding stage as compared to all TSMC's previous nodes.

    Our successful ramp-up in 20SoC has blazed a trail for 16FF and 16FF+, allowing us to rapidly offer a highly competitive technology to achieve maximum value for customers’ products,” said TSMC President and Co-CEO, Dr. Mark Liu. “We believe this new process can provide our customers the right balance between performance and cost so they can best meet their design requirements and time-to-market goals.”

    The 16FF+ process is on track to pass full reliability qualification later in November, and nearly 60 customer designs are currently scheduled to tape out by the end of 2015. Due to rapid progress in yield and performance, TSMC anticipates 16FF+ volume ramp will begin around July in 2015."

    btw TSMC in their Q3 2014 earnings call confirmed that they now expect high single digit % of revenue at 16 FINFET in Q4 2015 (just as 20nm hit 10% wafer revenue in Q3 2014). And we are talking about 16FF+ products ramping. The majority of customers have confirmed they are going direct to 16FF+. Its going to be fun watching Russ Fischer's and your claims that TSMC cannot ramp 16 FINFET products in 2015 being torn apart by TSMC's execution. We can expect the next gen QCOM FINFET Snapdragons to launch in early Q1 2016 and drive the Samsung Galaxy S7.

    Interestingly TSMC now has better progress at FINFET yields than Samsung and is in the clear lead. Thats understandable since 16FF / 16FF+ benefit from 20nm ramp and yield learning. They share the same BEOL (Back end of line) tools with modifications to MEOL (Middle EOL) and FEOL (the actual FINFET transistor device).
    Nov 15, 2014. 01:57 AM | 2 Likes Like |Link to Comment