Chinese semiconductor thread II

gotodistance

New Member
Registered Member
Why is Huawei's next chip called KIRIN 9100? It's the same 7nm chip, but why did they suddenly jump from 9010, 9020 to 9100?
 

gelgoog

Lieutenant General
Registered Member
The Kirin 9000S and 9010 are not the same chip. They use different Taishan large CPU cores and clockspeeds.
We still do not have actual data on the 9100 but the rumors say it might have a beefed up GPU.
I would not be surprised if they also designed their own small core instead of using one from ARM.
 

Xiongmao

Junior Member
Registered Member
It's well worth the temporary economic setback in exchange for semiconductor self-sufficiency.

Huawei has regained the #1 flagship crown in China, and AI NPUs don't require EUV or most advanced nodes, generative AI and LLMs have limited military implications. Training data quality and AI techniques matter as much if not more than raw processor power in the AI race.

So essentially, US has gifted China self-sufficiency in exchange for temporary short-term economic setback, and arguably negligible military AI advantage. US can only sustain it's leadership by being an innovative R&D leader, not stifling it's competitors with draconian sanctions. During peacetime too! What a gift that keeps on giving.
As I understand it, the US commerce secretary Raimondo had a lot to do with the chip sanctions, but that she is driven by a personal vendetta against China because her father lost his job when his employer relocated to China when she was a girl. It just makes it that bit sweeter when China emerges stronger, to know that her pettiness has caused great damage to her own country's competitiveness.
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
The Kirin 9000S and 9010 are not the same chip. They use different Taishan large CPU cores and clockspeeds.
We still do not have actual data on the 9100 but the rumors say it might have a beefed up GPU.
I would not be surprised if they also designed their own small core instead of using one from ARM.
lol, yep. The original comment sounds almost as ridiculous as saying QCOM SD 8 Gen 1, 2 and 3 are all the same chip.

Kirin 9100 is expected to use an improved process.
 

BlackWindMnt

Captain
Registered Member
The Kirin 9000S and 9010 are not the same chip. They use different Taishan large CPU cores and clockspeeds.
We still do not have actual data on the 9100 but the rumors say it might have a beefed up GPU.
I would not be surprised if they also designed their own small core instead of using one from ARM.
Talking about GPU is Moore Threads still in that market or did they completely pivoted to AI accelerators?
 

tphuang

Lieutenant General
Staff member
Super Moderator
VIP Professional
Registered Member
Please, Log in or Register to view URLs content!
5nm是定的,就是产量,希望能平稳

Please, Log in or Register to view URLs content!
某厂也在做内插帧技术,目前是在"5nm"新处理器上测试,主要应用于GPU高负载游戏,在渲染帧之间插预测帧并做缓冲补偿,可以削弱性能差异并降低功耗,玩游戏的感官流畅度更好,到时候看延迟会不会比通子低
Please, Log in or Register to view URLs content!
菊菊芯片在发布前一直没有真实且具体的参数信息,因为是国产工艺,更好做保密管理,同理……
bottom 2 are from the more trustworthy source. But obviously some people are calling the process used on 9100 a 5nm process, but it's not a real 5nm process like N5, that's why the quote around it.

The key here is a frame interpolation technology for high load game rendering. This tech allows games to be run smoother with lower latency and power consumption.

Remember that the key problem with 9000S and 9010 is the high end gaming performance. That's the main issue being addressed in 9100.
 

tokenanalyst

Brigadier
Registered Member

Embracing the AI era, Huatian Technology creates the "eSinC 2.5D packaging technology platform"​


For more than half a century, the semiconductor industry has been advancing by leaps and bounds in accordance with "Moore's Law", but as the advanced process technology of integrated circuits continues to shrink, it is getting closer and closer to the physical limit and facing technical bottlenecks; on the other hand, emerging industries such as artificial intelligence, 5G, and autonomous driving are booming, and the requirements for computing chip performance are gradually increasing, bringing multiple challenges.

Huatian Technology Creates eSinC 2.5D Platform

Public information shows that while expanding and improving the scale and level of its existing integrated circuit packaging business, Huatian Technology has vigorously developed advanced packaging technologies and products such as SiP, FC, TSV, Fan-Out, WLP, 2.5D, 3D, Chiplet, and FOPLP. In particular, for the 2.5D advanced packaging track, Huatian Technology is committed to building the eSinC (Embedded System in Chip) 2.5D packaging technology platform to meet the high-end packaging and testing needs of the artificial intelligence (AI) era.

Public information shows that Huatian Technology's eSinC 2.5D packaging technology platform includes three major 2.5D technology categories, namely, silicon interposer chiplet system SiCS (Silicon interposer Chiplet System), fan-out chiplet system FoCS (Fan out Chiplet System) and bridge chiplet system BiCS (Bridge interconnection Chiplet System):

1. SiCS is a 2.5D advanced packaging technology that uses silicon adapters to achieve multi-chip interconnection. This structure usually has high-density I/O interconnection and is suitable for the needs of high-performance computing and large-scale integrated circuits. The advantage of SiCS lies in its precise manufacturing process and superior electrical performance.

2. FoCS uses a rewiring layer (RDL) as an interposer to achieve interconnection between chips, mainly used to reduce costs and adapt to the connection requirements of different types of devices. It has greater design flexibility and can support more chip connections. The key features of FoCS technology include: using an RDL interposer, which is composed of polymers and copper wires and has relatively high mechanical flexibility.

3. BiCS key features include: using LSI chips to achieve high-density chip-to-chip interconnects. These chips can have a variety of connection architectures and can be reused for multiple products. The mold-based interposer has a wider RDL layer spacing and uses through-holes that penetrate the interposer to achieve low-loss and high-speed transmission of signals and power, and can integrate additional components.

Please, Log in or Register to view URLs content!
 
Top