AMD 7nm in August 2018: What foundry?

[josep]

AMD Navi based Graphics cards Might Arrive in August 2018 and it will be 7nm.
AMD's next-gen Navi GPU launching in August 2018
If it is true, I'm wondering, who is fabbing them?

GF? I don't expect GF 7LP to be ready till late 2018 (zen2 is expected in early 2019)

TSMC? We know that AMD is using TSMC 12 nm for Graphics cards, but I don't think TSMC 7nm will be available for AMD so early. We know TSMC 7-nanometer process node has already had several tape-outs and expects to reach volume capacity in 2018, but we also know Apple (maybe Qualcomm too?) has higher priority.

Any additional info or rumors?

 

[raghu78]

AMD Navi based Graphics cards Might Arrive in August 2018 and it will be 7nm.
AMD's next-gen Navi GPU launching in August 2018
If it is true, I'm wondering, who is fabbing them?

GF? I don't expect GF 7LP to be ready till late 2018 (zen2 is expected in early 2019)

TSMC? We know that AMD is using TSMC 12 nm for Graphics cards, but I don't think TSMC 7nm will be available for AMD so early. We know TSMC 7-nanometer process node has already had several tape-outs and expects to reach volume capacity in 2018, but we also know Apple (maybe Qualcomm too?) has higher priority.

Any additional info or rumors?

Since these are just rumours there is no guarantee that AMD will ship 7nm Navi in 2018. AMD could ship 7nm Navi by late 2018 or early 2019 but only if they are manufactured at TSMC N7 as thats the node which will be going into volume production in Q2 2018. Apple would get almost all the TSMC 7nm wafers in 2018 but AMD could get a very small initial wafer allocation starting in Q3 or Q4 2018 with the real volume coming in 2019. I still do not think we will see a 7nm GPU before Q2 2019 at the earliest. imo AMD will use TSMC N7 for GPUs and GF 7LP for Zen 2 and Rome server CPUs. I do not think GF will be able to supply enough wafers for AMD to fab both CPU and GPU at GF 7LP. Moreover I do not expect a GF 7LP based AMD product to ship before mid-Q3 2019.

 

[astilo]

Moreover I do not expect a GF 7LP based AMD product to ship before mid-Q3 2019.

Can I ask you why and based on what? It is still October 2017 as we speak.

 

[raghu78]

Can I ask you why and based on what? It is still October 2017 as we speak.

Risk production for GF 7LP starts in Q2 2018. It usually takes 1 year to go from risk to high volume production (HVM) which puts HVM in Q2 2019 . Thus actual products fabbed at GF 7LP are unlikely to arrive before Q3 2019.

 

[Daniel Nenni]

AMD Navi based Graphics cards Might Arrive in August 2018 and it will be 7nm.
AMD's next-gen Navi GPU launching in August 2018
If it is true, I'm wondering, who is fabbing them?

GF? I don't expect GF 7LP to be ready till late 2018 (zen2 is expected in early 2019)

TSMC? We know that AMD is using TSMC 12 nm for Graphics cards, but I don't think TSMC 7nm will be available for AMD so early. We know TSMC 7-nanometer process node has already had several tape-outs and expects to reach volume capacity in 2018, but we also know Apple (maybe Qualcomm too?) has higher priority.

Any additional info or rumors?

Since AMD competes with Intel the ship dates are very different.You have to remember that a foundry going into production at a certain node does not mean a foundry customer will be in production anywhere near that time. Intel however ships chips so their node production date is much more definite. AMD also ships less mature silicon that Intel does so the dates are very misleading.

If AMD is in production with 7nm in 2018 it will most certainly be risk production and not HVM. At the GTC event last month I was told that AMD CPU and GPU products were being done at GF while AMD game console products were at TSMC 16nm and 12nm. No mention of TSMC 7nm was made. I can check again with engineering contacts inside AMD but I have no reason to doubt they will be at GF.

You should also know that the first version of TSMC 7nm is for SoCs (Apple and QCOM). The high performance version of 7nm that is being developed with NVIDIA will be out later. Since Nvidia is intimately involved with I highly doubt AMD has early access to it.

 

[ryeoh1969]

My take is tsmc will make CPU for AMD rather than GPU. This way it can appease one of their big customer Nvidia and at the same time take over shares of INTEL business

 

[raghu78]

My take is tsmc will make CPU for AMD rather than GPU. This way it can appease one of their big customer Nvidia and at the same time take over shares of INTEL business

AMD CPUs and APUs are likely to be manufactured at GF 7LP as AMD is their primary customer with wafer commitments to meet. Moreover GF 7LP process is designed for AMD's high performance CPU requirements. TSMC will probably manufacture AMD GPUs.

 

[Jozo035]

AMD CPUs and APUs are likely to be manufactured at GF 7LP as AMD is their primary customer with wafer commitments to meet. Moreover GF 7LP process is designed for AMD's high performance CPU requirements. TSMC will probably manufacture AMD GPUs.

What are these requirements? TSMC is manufacturing 5GHz SPARC M8. 4 GHz ARM was demonstrated using 7nm. So what more AMD needs?

 

[Portland]

If they have the mechanics that keep the machines going.

 

[astilo]

What are these requirements? TSMC is manufacturing 5GHz SPARC M8. 4 GHz ARM was demonstrated using 7nm. So what more AMD needs?

Ryzen on 14LPP was a great success (they have 16 cores/32 threads running above 4GHz, very cost effective, good density and high yield). I see no reason to change foundry for them.
I do not like monopoly.

 

[astilo]

Risk production for GF 7LP starts in Q2 2018. It usually takes 1 year to go from risk to high volume production (HVM) which puts HVM in Q2 2019 . Thus actual products fabbed at GF 7LP are unlikely to arrive before Q3 2019.

In the meanwhile TSMC should start risk production end of 2017 and have volume production in the first half of 2018 (6-9 months). GF should be just 2 quarters late, since they start risk production in the first half of 2018. Applying the same logic, we should expect volume production by end of 2018, but let's say Q1 2019. That's still 6 months earlier than what you are guessing.
At the end, I just wanted to know if you had some first hand details. But I guess not, it looks like it was just your feeling.

 

[ryeoh1969]

Ryzen on 14LPP was a great success (they have 16 cores/32 threads running above 4GHz, very cost effective, good density and high yield). I see no reason to change foundry for them.
I do not like monopoly.

For AMD to gain market shares in CPU and GPU, the main consideration is the performance of the chips are up to par with competition and cost competitive. The key factor here is yield, tsmc may charge more but if their yield is significantly better then the decision is a no brainer.

Its been out in the market for more than a year now, tsmc 7nm yield and performance is out of the chart. Its very unlike their situation with 10nm. Furthermore, if from EYPC innovation to leading Intel in zen2 aka "Rome" they need a complete revamp of micro architecture. Compared with tsmc lead in yield and process maturity in FE plus ground braking tech in 3DIC, my guess is simple -->
AMD CPU design with new architecture +tsmc 7nm yield maturity +new advance packaging knowhow will win the next round of datacenter refresh.

 

[astilo]

Its been out in the market for more than a year now, tsmc 7nm yield and performance is out of the chart.

Well, considering TSMC 7nm risk production is planned for end of the year (2017), I do not know where did you get that.

 

[ryeoh1969]

Well, considering TSMC 7nm risk production is planned for end of the year (2017), I do not know where did you get that.

Of course from the rumor mills.....and the likes of Qualcomm switching camp and their 7nm taping out like mad

 

[Lodix]

Well, considering TSMC 7nm risk production is planned for end of the year (2017), I do not know where did you get that.

TSMC started risk production of 7nm in May this year.

But I dont think AMD will use TSMC 7nm in 2018 because the fabs will be full of smaller mobile chips that have higher volumen and yields.

 

[Jozo035]

Smaller or mobile? Choose one but don't use both.

Mobile SOCs are normally over 120 mm2 (2016 Iphone was 125 mm2 IIRC). It is similar size as Intel desktop 4-core processors (such as 7700K). So I don't think that there is any significant difference given by their size.

 

[Lodix]

No, they are normally under 100mm^2.

Anyway other designers will take all the initial volume in 2018. They are using GF 12nm for their 2017 products.

 

[astilo]

TSMC started risk production of 7nm in May this year.

But I dont think AMD will use TSMC 7nm in 2018 because the fabs will be full of smaller mobile chips that have higher volumen and yields.

Are you sure? Maybe initial tape-outs in May, risk production seems a bit premature to me, since I would assume that the very first 7nm product will be as usual the September 2018 iPhone SoC.

 

[Jozo035]

From their q1 earnings:

Now let's move to N7 and N7 Plus. TSMC N7 will enter risk production in second quarter this year. So far we have more than 30 customers actively engaged in N7 and we expect about 15 tape-outs in this year with volume production in 2018.

From q2, answer regarding volume ramp-up.

Yes, we estimate it's still about a year, not really the technology on our side. It's really on their design, product qualification and system verification get to the market. It typically takes that long. Although for the mobile, it's about one year. Some of the high-performance computing or PC application will be shorter. So -- but that 7 and 7+, I don't see much difference in the lead time up to the volume.

September Iphone release is not usually first product on new process. This year it was Ipad in Q2 and even earlier some Android devices at the end of Q1.

September Iphone release is driven by Apple marketing (update cycle). It has no relation to process availability.

Plus you probably did not realized, that respin in fab takes several weeks (maybe even months here). Plus They need time to integrate and test it into PCB. So any mobile phone release is not such a credible sign of process maturity.

For 7nm there will be Xilinx probably as early adopter. They are not driven by marketing and they relay mostly on pure silicon and packaging availability, so you can take a look at their announcements.

 

[Daniel Nenni]

Are you sure? Maybe initial tape-outs in May, risk production seems a bit premature to me, since I would assume that the very first 7nm product will be as usual the September 2018 iPhone SoC.

Remember, TSMC 7nm is using the 10nm fabs that are already in mass production so I do not expect delays due to yield. Many 7nm tape-outs including QCOM and Apple have already been completed so I expect to see TSMC 7nm chips in the first half of 2018. GF is about 6 months behind so the second half of 2018 is possible. But remember GF did not do 10nm so they will be starting with a greenfab.