Chinese semiconductor thread II

[gelgoog]

Will they used SMEE DUVi ? Cause ASML is not allowed to supply any new equipment to China.

It's just a chip packaging plant. They put the memory dies into plastic packages.

 

[measuredingabens]

Looks like the chip performance of the new Huawei P70s have been leaked.

Please, Log in or Register to view URLs content!

 

[AETHER]

Looks like the chip performance of the new Huawei P70s have been leaked.

Please, Log in or Register to view URLs content!

these are fake. model numbers dont match and the CPU info for MNA-AL00 is obviously an overclocked 8gen1

 

[FairAndUnbiased]

Don't know enough about the AI subject, but maybe this is a cheap Quantity to compete with quality situation?

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

Chinese researchers invented DLP architecture so this isn't a surprise.

Please, Log in or Register to view URLs content!

By 1988, Wei Zhang et al. had discussed fast optical implementations of convolutional neural networks for alphabet recognition.

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

In the 1990s, there were also attempts to create parallel high-throughput systems for workstations aimed at various applications, including neural network simulations.

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

This presentation covers a past attempt at neural net accelerators, notes the similarity to the modern SLI GPGPU processor setup, and argues that general purpose vector accelerators are the way forward (in relation to RISC-V hwacha project. Argues that NN's are just dense and sparse matrices, one of several recurring algorithms)

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

-based accelerators were also first explored in the 1990s for both inference and training.

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

In 2014, Chen et al. proposed DianNao (Chinese for "electric brain"),

Please, Log in or Register to view URLs content!

to accelerate deep neural networks especially. DianNao provides the 452 Gop/s peak performance (of key operations in deep neural networks) only in a small footprint of 3.02 mm2 and 485 mW. Later, the successors (DaDianNao,

Please, Log in or Register to view URLs content!

ShiDianNao,

Please, Log in or Register to view URLs content!

PuDianNao

Please, Log in or Register to view URLs content!

) are proposed by the same group, forming the DianNao Family

Please, Log in or Register to view URLs content!

 

[FairAndUnbiased]

It's just a chip packaging plant. They put the memory dies into plastic packages.

IDK there's advanced packaging of multiple dies into a single package, which does use lithographically defined features to connect large number of IO pins to each other.

 

[_killuminati_]

So their sanctions are not working, either China is servicing these tools themselves or the share of domestic made tools in higher end processes is higher that most D.C. pundits think and they are hoping/copeing that ASML is the only thin rope that keep China from making high end chips on their own.​

Maybe American companies are bypassing sanctions via backdoors to trade with China?

 

[Overbom]

Don't know enough about the AI subject, but maybe this is a cheap Quantity to compete with quality situation?

Please, Log in or Register to view URLs content!

Please, Log in or Register to view URLs content!

If this is really true then that's kinda of a big deal

Microsoft just some days ago said that for Copilot to run locally, PCs will require at least 40 TOPS of NPU performance. This processor is already 48 TOPS and is dirt cheap in comparison to other products.

At least for inference you dont need that overpriced NVIDIA expensive GPUs. Seems like a good opportunity for Chinese competition to overtake established Western players

 

[tokenanalyst]

Maybe American companies are bypassing sanctions via backdoors to trade with China?

US companies wont risk it doing directly, they could use third party companies for the services but even that is too much of a risk.

 

[sunnymaxi]

Guangyan Technology: An innovative pioneer in wafer-level test equipment ..​

Features

• Wafer-level CP test equipment : Our equipment supports high-precision uniformity testing, defect detection and inter-pixel sensing difference testing, providing a comprehensive solution for the quality control of sensors and optical components.
• Wafer-level quantum efficiency testing equipment : Through advanced quantum efficiency testing, photon transfer curve analysis and wafer/filter film defect testing technology, our equipment reveals every detail of photoelectric conversion efficiency and helps customers optimize product performance.
• Wafer-level oblique incident light measurement equipment : The equipment we provide includes unique oblique incident light measurement technology, supports Chief Ray Angle (CRA) analysis, phase focusing test (PDAF) and optical crosstalk testing, providing customers with comprehensive optical performance evaluation tool.
  

Guangyan Technology 's wafer-level test equipment not only enables customers to obtain accurate and reliable test data in the early stages of product development, but also provides the opportunity to re-wafer to further save R&D costs and accelerate product launch. We are committed to working with our customers to create a better future and promote the development of the semiconductor testing industry through continuous technological innovation and service optimization.

Please, Log in or Register to view URLs content!

 

[antonius123]

If this is really true then that's kinda of a big deal

Microsoft just some days ago said that for Copilot to run locally, PCs will require at least 40 TOPS of NPU performance. This processor is already 48 TOPS and is dirt cheap in comparison to other products.

At least for inference you dont need that overpriced NVIDIA expensive GPUs. Seems like a good opportunity for Chinese competition to overtake established Western players

But this chinese new AI GPU is made with 14nm while NVDIA is on lower nm such as 7nm or below? sound so amazing.