What is Bitmain doing with the new 12nm chips they're buying in 2018?

[dlezama]

Article here: http://news.8btc.com/bitmain-urgent-order-100k-12nm-chips-from-tsmc-in-2018

TL;DR:

According to Digitimes, TSMC has received an urgent order from Bitmain for 100k 12nm HPC (high performance computing) chips at the beginning of 2018. 10k chips will be delivered every month in Q1 2018 and more in Q2.

So what are this chips going to become? If they are getting these chips in Q1, product sales will start soon. Or maybe they are producing next-gen hardware for them and we won't see the new miners for months?

[1714866038]

1714866038

[1714866038]

1714866038

[1714866038]

1714866038

[majlkcze]

Article here: http://news.8btc.com/bitmain-urgent-order-100k-12nm-chips-from-tsmc-in-2018

TL;DR:

According to Digitimes, TSMC has received an urgent order from Bitmain for 100k 12nm HPC (high performance computing) chips at the beginning of 2018. 10k chips will be delivered every month in Q1 2018 and more in Q2.

So what are this chips going to become? If they are getting these chips in Q1, product sales will start soon. Or maybe they are producing next-gen hardware for them and we won't see the new miners for months?

Interesting find, but I did not find any other source of this news. Link to "Digitimes" marked as a source is not working, direct search on DigiTimes shows nothing.
I'm not sure if this is reliable message to take care of.

It will be interesting to see more replies from members on this topic.

Thanks for the info anyway.

[philipma1957]

10k chips per month with 75% yield = 7500 chips/183 = 40 miners

So if they are putting them into an s-11 it is small batch

my guess is their sophon project  seems like   that gear  may be  10 chips or less  so 7500 good chips = 750 to maybe 1500 deep learning cards


https://sophon.ai/

[lrowland21093]

Interesting find, but I did not find any other source of this news. Link to "Digitimes" marked as a source is not working, direct search on DigiTimes shows nothing.

FWIW...
Here is the very brief release on the public Digitimes that did not mention the name of the client (Bitmain).
https://www.digitimes.com/news/a20180102VL201.html
There is a more detailed article on the paid side of Digitimes that mentioned Bitmain but I cannot link it since it is for paid user only.
Of course, there are no details on what Bitmain plans to do with the 12nm chips though.

[philipma1957]

the long link mentions bitmain  and it would be 10k chips a month for first 3 months  of year

and a total of 100k chip order.


which  if you are putting into a 180 chip s11  is not a lot of s11's

[majlkcze]

the long link mentions bitmain  and it would be 10k chips a month for first 3 months  of year

and a total of 100k chip order.


which  if you are putting into a 180 chip s11  is not a lot of s11's

Your counting is not correct IMHO,

100.000 wafers is not 100.000 chips... There are usually a lot of chips on one wafer, there is the same mistake on 8btc website who re-take this news...
Something big is happening there! It is a lot of chips.

Maybe it is 12nm-16nm for next batches of S9 (maybe S9+) because they hyped price and there is still no competing company, they have a monopol now. So why make new miner now?

This is why I think that:

"An enhanced version of TSMC's 16nm process was introduced in late 2016 called "12nm"
"In late 2016 TSMC announced a "12nm" process (e.g. 12FFC) which uses the same design rules as the 16nm node. The enhanced process is said to feature lower leakage better and cost characteristics and perhaps a better name (vs. "14nm"). 12nm is expected to enter mass production in late 2017."

[Source: https://en.wikichip.org/wiki/16_nm_lithography_process]

[the_electronrancher]

That's the question, is it 100K wafers or 100K chips?  100K chips is not even one lot of 25 wafers if it were bm1387-sized chips, so that's a debug run or metal split or something. 

100K wafers on the other hand is a stupidly large order, on the order of one billion chips in 4000 lots.  The upfront cost would be quite staggering, especially if it were expedited to TSMC's highest priority as suggested by the mention of urgency.

[Erelas]

That's the question, is it 100K wafers or 100K chips?  100K chips is not even one lot of 25 wafers if it were bm1387-sized chips, so that's a debug run or metal split or something. 

100K wafers on the other hand is a stupidly large order, on the order of one billion chips in 4000 lots.  The upfront cost would be quite staggering, especially if it were expedited to TSMC's highest priority as suggested by the mention of urgency.

That's what I was trying to get my head around, the possibility that it is the actual number of wafers... I mean, SkyNet probably used less to destroy the world.

I wouldn't be surprised though if BitMain could actually handle that serious frontend cost, they've made quite a profit just in only taking BCH and running that racket.

[majlkcze]

That's the question, is it 100K wafers or 100K chips?  100K chips is not even one lot of 25 wafers if it were bm1387-sized chips, so that's a debug run or metal split or something. 

100K wafers on the other hand is a stupidly large order, on the order of one billion chips in 4000 lots.  The upfront cost would be quite staggering, especially if it were expedited to TSMC's highest priority as suggested by the mention of urgency.

That's what I was trying to get my head around, the possibility that it is the actual number of wafers... I mean, SkyNet probably used less to destroy the world.

I wouldn't be surprised though if BitMain could actually handle that serious frontend cost, they've made quite a profit just in only taking BCH and running that racket.

Yes, surprisingly they are releasing next batch of S9 today... I'm curious when they will come out with 12nm because if the news about new miner come out, nobody would buy S9 now...
or maybe they will keep them for their farms?

[philipma1957]

the long link mentions bitmain  and it would be 10k chips a month for first 3 months  of year

and a total of 100k chip order.


which  if you are putting into a 180 chip s11  is not a lot of s11's

Your counting is not correct IMHO,

100.000 wafers is not 100.000 chips... There are usually a lot of chips on one wafer, there is the same mistake on 8btc website who re-take this news...
Something big is happening there! It is a lot of chips.

Maybe it is 12nm-16nm for next batches of S9 (maybe S9+) because they hyped price and there is still no competing company, they have a monopol now. So why make new miner now?

This is why I think that:

"An enhanced version of TSMC's 16nm process was introduced in late 2016 called "12nm"
"In late 2016 TSMC announced a "12nm" process (e.g. 12FFC) which uses the same design rules as the 16nm node. The enhanced process is said to feature lower leakage better and cost characteristics and perhaps a better name (vs. "14nm"). 12nm is expected to enter mass production in late 2017."

[Source: https://en.wikichip.org/wiki/16_nm_lithography_process]

the article said chips not wafers

http://news.8btc.com/bitmain-urgent-order-100k-12nm-chips-from-tsmc-in-2018\\

my guess is  these are for testing purposes not large production.

[majlkcze]

the long link mentions bitmain  and it would be 10k chips a month for first 3 months  of year

and a total of 100k chip order.


which  if you are putting into a 180 chip s11  is not a lot of s11's

Your counting is not correct IMHO,

100.000 wafers is not 100.000 chips... There are usually a lot of chips on one wafer, there is the same mistake on 8btc website who re-take this news...
Something big is happening there! It is a lot of chips.

Maybe it is 12nm-16nm for next batches of S9 (maybe S9+) because they hyped price and there is still no competing company, they have a monopol now. So why make new miner now?

This is why I think that:

"An enhanced version of TSMC's 16nm process was introduced in late 2016 called "12nm"
"In late 2016 TSMC announced a "12nm" process (e.g. 12FFC) which uses the same design rules as the 16nm node. The enhanced process is said to feature lower leakage better and cost characteristics and perhaps a better name (vs. "14nm"). 12nm is expected to enter mass production in late 2017."

[Source: https://en.wikichip.org/wiki/16_nm_lithography_process]

the article said chips not wafers

http://news.8btc.com/bitmain-urgent-order-100k-12nm-chips-from-tsmc-in-2018

Digitimes is original source:
https://www.digitimes.com/news/a20180102VL201.html

"The short lead-time orders total about 100,000 wafer starts, and will be built using a 12nm process at TSMC"

There is a mistake in re-taken article on 8btc 

So this is a hell lot of chips... Maybe BM is trying to overtake some cryptocurrency or make private hard-fork? I'm just speculating...

[philipma1957]

the long link mentions bitmain  and it would be 10k chips a month for first 3 months  of year

and a total of 100k chip order.


which  if you are putting into a 180 chip s11  is not a lot of s11's

Your counting is not correct IMHO,

100.000 wafers is not 100.000 chips... There are usually a lot of chips on one wafer, there is the same mistake on 8btc website who re-take this news...
Something big is happening there! It is a lot of chips.

Maybe it is 12nm-16nm for next batches of S9 (maybe S9+) because they hyped price and there is still no competing company, they have a monopol now. So why make new miner now?

This is why I think that:

"An enhanced version of TSMC's 16nm process was introduced in late 2016 called "12nm"
"In late 2016 TSMC announced a "12nm" process (e.g. 12FFC) which uses the same design rules as the 16nm node. The enhanced process is said to feature lower leakage better and cost characteristics and perhaps a better name (vs. "14nm"). 12nm is expected to enter mass production in late 2017."

[Source: https://en.wikichip.org/wiki/16_nm_lithography_process]

the article said chips not wafers

http://news.8btc.com/bitmain-urgent-order-100k-12nm-chips-from-tsmc-in-2018

Digitimes is original source:
https://www.digitimes.com/news/a20180102VL201.html

"The short lead-time orders total about 100,000 wafer starts, and will be built using a 12nm process at TSMC"

There is a mistake in re-taken article on 8btc  

So this is a hell lot of chips... Maybe BM is trying to overtake some cryptocurrency or make private hard-fork? I'm just speculating...

that article appears to refer to the entire 12nm production of the foundry with more then 1 company in on that amount


well if you think every chip from 100,000 wafers  is for Bitmain  damn  that would be a lot of gear.

Havw ro wait and see.

1 wafer = $3500  I am  not sure that bitmain order 350 million  dollars in chips

3500 x 100000 =  350,000,000  I think that is all wafers for all companies  but  if it is just for bitmain  that is very impressive

maybe you are correct.

that would be 100,000,000 chips at $3.50 each   which could build

72% x 100 mill =  72 mill/180 chips a s-11 =

 400,000 s-11 or s9+  number wise  at say 18th  each

4 = 72th
40 = 720th
400 = 7.2ph
4000 = 72ph
40000 = 720ph
400000 = 7200ph or 7.2eh

actually  your numbers seem correct  since the network is at 13.2eh  adding 7.2 to go to 20.4eh makes sense

[majlkcze]

the long link mentions bitmain  and it would be 10k chips a month for first 3 months  of year

and a total of 100k chip order.


which  if you are putting into a 180 chip s11  is not a lot of s11's

Your counting is not correct IMHO,

100.000 wafers is not 100.000 chips... There are usually a lot of chips on one wafer, there is the same mistake on 8btc website who re-take this news...
Something big is happening there! It is a lot of chips.

Maybe it is 12nm-16nm for next batches of S9 (maybe S9+) because they hyped price and there is still no competing company, they have a monopol now. So why make new miner now?

This is why I think that:

"An enhanced version of TSMC's 16nm process was introduced in late 2016 called "12nm"
"In late 2016 TSMC announced a "12nm" process (e.g. 12FFC) which uses the same design rules as the 16nm node. The enhanced process is said to feature lower leakage better and cost characteristics and perhaps a better name (vs. "14nm"). 12nm is expected to enter mass production in late 2017."

[Source: https://en.wikichip.org/wiki/16_nm_lithography_process]

the article said chips not wafers

http://news.8btc.com/bitmain-urgent-order-100k-12nm-chips-from-tsmc-in-2018

Digitimes is original source:
https://www.digitimes.com/news/a20180102VL201.html

"The short lead-time orders total about 100,000 wafer starts, and will be built using a 12nm process at TSMC"

There is a mistake in re-taken article on 8btc  

So this is a hell lot of chips... Maybe BM is trying to overtake some cryptocurrency or make private hard-fork? I'm just speculating...

that article appears to refer to the entire 12nm production of the foundry with more then 1 company in on that amount


well if you think every chip from 100,000 wafers  is for Bitmain  damn  that would be a lot of gear.

Havw ro wait and see.

1 wafer = $3500  I am  not sure that bitmain order 350 million  dollars in chips

3500 x 100000 =  350,000,000  I think that is all wafers for all companies  but  if it is just for bitmain  that is very impressive

Is there somebody who knows what is BM1387 chip size? And what is TSMC 12nm wafer size?   maybe some deeper search on the internet can help to figure estimated chips amount  
The main question is: is it only for BM?
In case of $350m invest, I'm not skeptical, because they made a lot of money via BCH hype.

Some fast digging for 16nm:

If you know the die-size of the chip, you can figure out exactly how many chips you can print per wafer. The Bitmain BM1387 chip is somewhere below 20 mm² in die size, and though it isn’t published, looks to be 3mm x 4mm by measuring the inside.

We can plug in these numbers to a die-per-wafer calculator and get 5158 chips per wafer (we’re using the 300mm/12" wafers as is industry standard). Each S9 miner uses 189 such chips, so each wafer can make enough chips for a little over 27 S9's. Each wafer costs around $8000, so the chips for each S9 costs roughly $300.

source: https://medium.com/@jimmysong/just-how-profitable-is-bitmain-a9df82c761a

So if 12nm would be the same size (probably should be even smaller?), this is for 27*100.000 = 2.700.000 units
If it stays on same hashrate (14TH/s) they have: 2.700.000 * 14 = 37.800.000 TH/s = 37.800 PH/s
This would be >100% attack  

BTW. its not bad to overtake $16.279.200.000 total network cap with $350.000.000 investment
In other side, if I would overtake whole BTC network, I would not get this message leaked

[philipma1957]

 or it is simply 100k worth of test chips and a meh.

Besides  I have fully explained that bitmain can hold all out btc hostage by fee manipulation.  Fact not fiction

https://bitcointalk.org/index.php?topic=2634505.msg27049906#msg27049906


read        the post above   the fees here for a 1335 byte transaction that creates 30 to 50      227 byte sized transaction
most of those 1335 byte transaction were white listed at 0.0001 btc

yet a normal 1000 byte transaction cost 0.004 to 0.009 if you

Study these charts below

[fanatic26]

This lines up with the rumors ive heard that Bitmains next miner will be 12nm. They are supposed to announce it this month.

[majlkcze]

This lines up with the rumors ive heard that Bitmains next miner will be 12nm. They are supposed to announce it this month.

Hopefully, better to announce new miner than overtake network 

[BenRickert]

I think we all know that Bitmain will not take a new and enhanced technology directly to the public. There will be an initial "testing" , mining phase in their own facility. To launch directly to outside markets would be an insane risk on the QA/QC front, not to mention they would benefit from the better efficiency within their own mines initially. Like they have always done. BTW, from 16nm to 12 nm is not what I would call a quantum leap. The 14ths S9 will remain extremely relevant if that be the case. Lets face it. Moores law is hitting a wall. Further advancements will be harder and harder to come by with less and less increase in efficiency.

[QuintLeo]

10000 chips a month is a drop in the bucket - that would be ballpark 500 miners a month.

 I'd bet on these chips being intended for use in Bitmain's "AI Deep Learning" type new project, NOT in a cryptocoin miner.