Chinese semiconductor industry

Status
Not open for further replies.
Blitzo

Blitzo

Lieutenant General
Staff member
Super Moderator
Registered Member
Orthan said:
Perhabs @WTAN could add more information about this?
Click to expand...

Even taking the article itself at face value, there actually isn't too much to say that is inconsistent with what we know.

Arguably the most important paragraph is this one:
"由于《瓦森纳协定》的限制，上海微电子装备很难从国外进口用于生产高端光刻机的部件在02专项光刻机项目二期中，设定的时间为：2021年验收193nmArF浸没式DUV光刻机，对标产品为ASML现阶段最强DUV光刻机：TWINSCAN NXT:2000i。各子系统拆分如下：上海微电子装备负责光刻机设计和总体集成，北京科益虹源提供光源系统，北京国望光学提供物镜系统，国科精密提供曝光光学系统，华卓精科提供双工作台，启尔机电提供浸没系统。"
 
Fedupwithlies

Fedupwithlies

Junior Member
Registered Member
And now for something completely different:

A little while ago I talked about the Supercomputing Conference that was going to happen Nov. 14-18. (post here: Chinese semiconductor industry)

I said "There's not a lot of immediate "newsworthy" stuff that comes out of these"

I was... pretty goddamn wrong.

Remember that NextPlatform article about China reaching 2 exa-scale supercomputers? (here it is:
Please, Log in or Register to view URLs content!
)

Well part of their reporting was from "outstanding authority" but at least this part, "This simulation... highlights extensive use of mixed-precision math, including 16-bit floating point performance of a reported 4.4 exaflops."

Is from this paper that was presented to the conference (paper here:
Please, Log in or Register to view URLs content!
)

This paper btw won the Gordon Bell Prize. Just thought people would like to know.

As a further follow up, however, this article from TomsHardware :

(
Please, Log in or Register to view URLs content!
)

Well the title itself gives it away. The author believes that the Chinese supercomputers did not reach exascale according to the TOP500 standards because the Conference Paper says "with a sustained performance of 1.2 Eflops (single-precision)"

He believes that the single-precision means it was 32-bit LINPACK operation and not the 64-bit LINPAK operations that TOP500 requires.

He also said they cheated, with the implication that they defrauded the Gordon Bell prize.

Just a couple of things: First, some systems run 64-bit LINPACK as single-precision. (I'll admit, I've never seen one, its like something out of the 90s. But I'm not a supercomputing expert)

Second, Gordon Bell prize isn't awarded based on benchmarking, its based on what you do with it. And, from the award announcement itself, "The project far outpaced state-of-the-art approaches to simulating an RQC. "

This is just another example of the western attitude of "if a Chinese person beats us, they must've cheated!"

Normally I like reading Anton Shilov's articles because he reports happenings in the computing world that I'm totally out of touch with. However, this article reads like so much sour grapes.

Edit: Also why are we acting like the TOP500 standards are the only standards? Remember, these computers were never submitted to TOP500. Its the western media that's claiming them to be exascale by TOP500, not the makers themselves (other than some anonymous source by NextPlatform).

ALSO also, nice to see China have more supercomputers in general. Its not just depth (how fast the top computers are) but breadth (how accessible the middle of the pack supercomputers are) that can drive research.
 
Last edited:
FairAndUnbiased

FairAndUnbiased

Brigadier
Registered Member
Fedupwithlies said:
And now for something completely different:

A little while ago I talked about the Supercomputing Conference that was going to happen Nov. 14-18. (post here: Chinese semiconductor industry)

I said "There's not a lot of immediate "newsworthy" stuff that comes out of these"

I was... pretty goddamn wrong.

Remember that NextPlatform article about China reaching 2 exa-scale supercomputers? (here it is:
Please, Log in or Register to view URLs content!
)

Well part of their reporting was from "outstanding authority" but at least this part, "This simulation... highlights extensive use of mixed-precision math, including 16-bit floating point performance of a reported 4.4 exaflops."

Is from this paper that was presented to the conference (paper here:
Please, Log in or Register to view URLs content!
)

This paper btw won the Gordon Bell Prize. Just thought people would like to know.

As a further follow up, however, this article from TomsHardware :

(
Please, Log in or Register to view URLs content!
)

Well the title itself gives it away. The author believes that the Chinese supercomputers did not reach exascale according to the TOP500 standards because the Conference Paper says "with a sustained performance of 1.2 Eflops (single-precision)"

He believes that the single-precision means it was 32-bit LINPACK operation and not the 64-bit LINPAK operations that TOP500 requires.

He also said they cheated, with the implication that they defrauded the Gordon Bell prize.

Just a couple of things: First, some systems run 64-bit LINPACK as single-precision. (I'll admit, I've never seen one, its like something out of the 90s. But I'm not a supercomputing expert)

Second, Gordon Bell prize isn't awarded based on benchmarking, its based on what you do with it. And, from the award announcement itself, "The project far outpaced state-of-the-art approaches to simulating an RQC. "

This is just another example of the western attitude of "if a Chinese person beats us, they must've cheated!"

Normally I like reading Anton Shilov's articles because he reports happenings in the computing world that I'm totally out of touch with. However, this article reads like so much sour grapes.

Edit: Also why are we acting like the TOP500 standards are the only standards? Remember, these computers were never submitted to TOP500. Its the western media that's claiming them to be exascale by TOP500, not the makers themselves (other than some anonymous source by NextPlatform).

ALSO also, nice to see China have more supercomputers in general. Its not just depth (how fast the top computers are) but breadth (how accessible the middle of the pack supercomputers are) that can drive research.
Click to expand...
these are some jealous haters. that's K, they need the constant copium supply more than we do.
 
A

ansy1968

Brigadier
Registered Member
tokenanalyst said:
View attachment 79187

If I were the NIKON CEO i would be begging on my knees in front of the Japanese government building to let me sell lithography machines in China. Nikon is dying slowly and painfully.
Click to expand...
@tokenanalyst bro your chart is from 2020, my diagnosis, the patient is in his/her last straw. The window of opportunity had definitely close. Nikon latest DUVL that is capable of producing 5nm is scheduled for introduction in 2023, by that time an equivalent SMEE DUVL machine the SSA900 22nm DUVL maybe in large scale production, I'm speculating but 2 years is enough time to improve on SSA800 DUVL especially with end user feedback from SMIC, YMTC, HUA HONG and Huawei. Those partnership will help SMEE greatly just like what ASML and TSMC had done to become number 1.
 
T

tokenanalyst

Brigadier
Registered Member
Conflicting goals:

Nikon:
1637370296919.png
China:
1637370376840.png

United States:

1637370509352.png
Is up to the Japanese government if they want to continue to be Lapdogs of the U.S. to the point of letting their poster child company die in 5-7 years once SMEE immersion lithography machines become dominant on China. At the end of the day Japan will lose their ability to make advance litho machines. Letting just two countries with that ability.
 
A

ansy1968

Brigadier
Registered Member
tokenanalyst said:
Conflicting goals:

Nikon:
View attachment 79190
China:
View attachment 79191

United States:

View attachment 79192
Is up to the Japanese government if they want to continue to be Lapdogs of the U.S. to the point of letting their poster child company die in 5-7 years once SMEE immersion lithography machines become dominant on China. At the end of the day Japan will lose their ability to make advance litho machines. Letting just two countries with that ability.
Click to expand...
@tokenanalyst bro the US stated goal is to destroy the Japanese dominance during the 80's. The sign is there with an EUVL consortium being form without them. The same thing that is happening now with the Chinese, but the difference, China is playing catch up, fully motivated and in survival mode. I think Japan as a whole had accept its fate and give up, an unfortunate event cause as Asian we should celebrate her success.
 
Status
Not open for further replies.
Top