Chinese semiconductor industry

siegecrossbow said:

This is because China does not want decoupling.

Click to expand...

something is better than nothing.

China always play long term game. PRC wants to keep relations with US high tech firms as long as it benefit local industries in technology 'know how'..

its like dual strategy. continue to work with foreign firms and simultaneously develop own local industry too.

The days of unique components and low level coding to fringe API is dying or dead. Modern consoles and handhelds now are mostly using off the shelf or tweaked cpu/gpu/memory/ssd these days. Even the games are moving to online stores. Basically just tweaked "smartphones" and PCs.

sunnymaxi said:

https://twitter.com/i/web/status/1566132077954625538

Click to expand...

Just as an fyi, this happened over a year ago. The tweet here is a little misleading. A little more its deployment.

Kunlun, which specializes in artificial intelligence chips, has recently started mass producing microchips that pack more computational power by using 7-nanometer technology. The firm has deployed its AI chips to power Baidu’s core searching services

Click to expand...

I don't think Baidu has publicized its performance for fear of getting greater sanctions. They just use arbitrary measures like 2 to 3 times faster than kunlun 1. After the Hisilicon sanctions, I think all the Chinese tech firms got more secretive with chips that they only intend to use internally.

On Kunlun 1
They produced over 20000 of it by the time Kunlun 2 got announced in Aug 2021. It's comparable in performance to V100. So if you look at this, Baidu/Alibaba/Huawei have all already been producing AI chips at V100 level. It seems to me that by the time A100 supply is cut off, there is no reason to procure anymore Nvidia chips. And note that Kunlun 2 chips work with Domestic OS.

I mean the figures listed here for R200 AI accelerator (you can also find this on Baidu cloud website) really isn't that impressive. I mean its 32 FP performance is pretty good and 16 FP performance is okay, but 8 FP performance isn't any better than Baidu 1. It claimed to be able to do 1.7 times the processing of market leading 150 w CPU

The one good thing about kunlun 2 is that it is definitely already in mass production and in use. Here is an article of it winning award with Shandong. Lol, this part saying it already has over 100 million orders. That can't be real. I think maybe that's 100 million RMB in orders since 2021. Used in GPU, AI framework and OS

昆仑芯2代产品已与多款通用处理器、操作系统、AI框架等完成端到端的适配,2021年已启动互联网及各行业客户交付,目前在手订单过亿,商业化进展顺利。

Click to expand...

looks like they are developing kunlun-3, kunlun-4 and self driving kunlun chips. Again even back in March, it claims to have well over 10 million orders. I'm not sure how this can be real.

tphuang said:

Please, Log in or Register to view URLs content!

The one good thing about kunlun 2 is that it is definitely already in mass production and in use. Here is an article of it winning award with Shandong. Lol, this part saying it already has over 100 million orders. That can't be real. I think maybe that's 100 million RMB in orders since 2021. Used in GPU, AI framework and OS

Please, Log in or Register to view URLs content!

looks like they are developing kunlun-3, kunlun-4 and self driving kunlun chips. Again even back in March, it claims to have well over 10 million orders. I'm not sure how this can be real.

Click to expand...

Could it mean the amount of chips rather than the number of orders? Because 10 and 100 million RMB is about 1.4 and 14 million USD. Something like that seems way to low amount of chips that is supposed to be used in various industries.

The exhibited products have been mass-produced and applied, and have been deployed on a large scale in the Internet, smart computing centers, smart cities, biological computing and other fields, helping the government and enterprises to accelerate the intelligent layout of the industry.

Click to expand...

Interesting comments here by Americans - on the Nvidia gpu ban - that are devoid of the fanboyism you see on Reddit.

Topazchen said:

Interesting comments here by Americans - on the Nvidia gpu ban - that are devoid of the fanboyism you see on Reddit.

Please, Log in or Register to view URLs content!

Click to expand...

What are you talking about? Hackernews has tons of fanboyism.

Appix said:

Washington has told Nvidia that the new curbs are designed to prevent advanced AI gear being used for or diverted to military purposes by China or Russia. In June of this year, Washington, DC-based think tank The Center for Security and Emerging Technology

Please, Log in or Register to view URLs content!

almost all of the 97 AI chips in public Chinese military purchase records between April and November 2020 were designed by US firms Nvidia,

Please, Log in or Register to view URLs content!

,

Please, Log in or Register to view URLs content!

or Xilinx, which is now part of AMD.

Click to expand...

I'm not particularly worried about China's ability to design GPUs compared to the west, I'm more worried about China's ability to design CPUs and FPGAs. The research document linked

FYI Intel, Microsemi and Xilinx/AMD are the largest FPGA manufacturers in the world.

Does someone have more in-depth knowledge about the development of localized FPGA situation in China? How did the AMD merger with Xilinx pass China's regulations?

horse said:

This stacking of current nodes, will be very interesting when those chips inside the final products come onto the scene.

If the specs and how it will work in the final product are accurate, then this could change the IC game a lot.

The single chip from an advanced node is still a more elegant solution. But that is just another way of saying that the chip stacking will make the advanced node seemed over-engineered.

Click to expand...

olalavn said:

next year, SMIC will use new method, and Polysiliicon to manufacture chip reduce chip temperature... i hope it will reduce temperature when stacking chip

Click to expand...

Absent China's access to EUV, I think the most logical path for them is to use advanced packaging of 2.5D/3D at 14nm and try to engineer IC logic with dozens of layers NAND style. I imagine Huawei would have to pursue an unconventional cooling solution like embedding liquid coolant as part of the packaging or use thermoelectric semiconductors with some kind of Peltier cooling. Is this what you are saying?

tphuang said:

Huawei should use the 14 nm and then the 7 nm fabs to first produce the server chips, the AI chips and the smartphone CPUs. Until they can get enough fabs into production, they need to focus on ramping production for the chips they really need. At a very minimum, they should get Kirin/Kunpeng chips back into production.

And once they can first restore their previous chip production that got cut off by TSMC sanction, then they need to get into the computer/laptop CPU market.

Click to expand...

Another strategy China could pursue is increasing their world cloud market share and total world cloud computing capacity. This would capitalize on 5G infrastructure to gradually move computer programs and mobile applications to the global cloud in client-server mode. This would offload computing to China's supercomputer servers. This would also circumvent performance disadvantages for all non-graphics intensive applications as well as facilitating the 4th Industrial Revolution to China's favor. I wouldn't be surprised if this is what the Americans are planning for themselves and they want to circumvent China doing the same by grabbing world cloud market share in this domain before China does it.

US govt banned selling PlayStation 2 to China in 1999 ... so banning A100/H100 shouldn't be a surprise to anybody
you can see how ridiculous some (many) US politicians are