antiterror13
Brigadier
Well, AI technology is developing fast unlike x86 ... so 200% perf increase is impressive but not phenomenal
I think it will be banned soon ....
Chinese semiconductor industry
- Thread starter Hendrik_2000
- Start date
- Status
Well, AI technology is developing fast unlike x86 ... so 200% perf increase is impressive but not phenomenal
I think it will be banned soon ....
"...Chinese semiconductor makers have produced 203.6 billion chips so far in 2021, a 47.3% increase year-over-year."
This means doubling production in two years. Amazing...
A real pity... Huawei is being slaughtered for the greater good of China. If it had not taken such a beating, China would not have waked up. China owes a lot to Huawei.
I agree, but Huawei is also guilty of doing the same, they should have invested a long time ago to improve China's semiconductor manufacturing capabilities and not leave everything to the government, they should have taken positions in different companies in the Chinese semiconductor industry and force that their machinery, software and chemicals became more competitive, they should have worked with the different Chinese institutes to develop technology such as EUV lithography. However Huawei preferred to work with untrustworthy western or westernized companies such as TSMC and Qualcomm. The good thing is that they are working fast to remedy that.
To be fair, most Chinese companies and people had a high regard for the US back then despite the US and the rest of the Western world not reciprocating such respect. Trump's trade war and the events that occurred in 2020 really shook the Chinese to their core.
You are misreading this… this is user level equipment, modems. This is equipment that Huawei already knows is dead. These are things like portable hotspots or USB modems that absolutely require 7nm chips due to size and power. If you don’t buy the Qualcomm modems, then you render the entire 5G network pointless which is even worse for Huawei.
US can try to retard the development in technology, but in race, you don’t win by slowing the opponent, you win by being faster. Huawei is not letting up in R&D.
- "In the early days of exascale computing and AI, these customer-configurable integrated circuits played a key role. Organizations could program and reprogram FPGAs onsite to handle a range of changing demands. As time went on, however, their performance and market growth got outpaced by faster GPUs and specialized ASICs."
- "Now, innovations like high-speed AI tensor logic blocks, configurable embedded SRAM, and lightning-fast transceivers and interconnects are putting this early leader back in the race."
- “FPGAs offer hardware customization with integrated AI and can be programmed to deliver performance similar to a GPU or an ASIC,” explains Kuppuswamy. “The reprogrammable, reconfigurable nature of an FPGA lends itself well to a rapidly evolving AI landscape, allowing designers to test algorithms quickly and get to market fast and scale quickly”.
- "FPGAs are often used where data must traverse many different networks at low latency. They’re incredibly useful at eliminating memory buffering and overcoming I/O bottlenecks — one of the most limiting factors in AI system performance. By accelerating data ingestion, FPGAs can speed the entire AI workflow."
- "The company also announced a partnership with Intel on a series of AI projects, including FPGA-backed workload acceleration, a deep learning framework based on Xeon scalable processors. Intel did not identify which FPGA series Baidu would use, but the chip maker recently announced the integration of its Arria family with its mainstream Xeon server chip."
- "This portion of the deal is a cloud-based partnership, as Baidu said it was looking to develop a “heterogeneous” cloud computing platform based on Intel FPGAs."
- "Ouyang Jian also showed the killer feature of the Kunlun chip and its compatibility with the China-made processor Feiteng via video."
- "Baidu AI chip accumulation was due to its FPGA to do the accumulation of AI acceleration, but also thanks to its software-defined accelerator and XPU architecture years of accumulation."
- "Ouyang Jian said, "Compared to GPUs, the Kunlun chip has done a good job of being versatile and programmable, and we're still working on making the programmability better."
- "After the release of Kunlun, news about it was released one after another. Architecture-wise, Kunlun has 2 computing units, 512GB/S of memory bandwidth and 16MB SRAM/unit."
- "According to Ouyang Jian, 16MB SRAM is good for AI inference, XPU-SDNN on XPU architecture is designed for Tensor and so on, and XPU-Cluster can meet the needs of general processing."
Baidu's AI chip initiative is part of a larger heterogeneous cloud computing platform. Meaning, they make use of more than 1 chip for their cloud computing services. The Kunlun chip is the main chip in this architecture but they also use the Feiteng CPU, and Intel FPGAs in conjunction for FPGA acceleration. The Kunlun chip itself is a sort of XPU accelerator, very likely FPGA based.
Articles on Kunlun range from being a standalone GPU accelerator, to standalone XPU, to FPGA prototype evolved to accelerator, to 1 part of a heterogeneous AI chip architecture. Keep this in mind because you will find confusion among different stories about what Kunlun is in different articles, including some claiming it is an ASIC. Imo, "many" of these reporters are simply wrong. It is a fact that Intel FPGAs are used by Baidu in their cloud architecture since Kunlun I and we know Baidu primarily uses Kunlun for their cloud services. We also know Kunlun was designed to be versatile, configurable and programmable, which points to it being either directly FPGA based or co-processed with FPGAs. It's also telling that Kunlun is an AI chip that has an SRAM unit which is reminiscent of HBM used for FPGA acceleration in other AI chips.
antiterror13
Brigadier
you meant 1 year?
antiterror13
Brigadier
well, to be fair, nobody expected that and before the crazy Trump adm, the definition of US content is 25% of the origin to the US tech, now it basically even 0.01% still considered US product, really crazy
Since then, nobody would trust US tech again ever, everybody (or most) would try hard to de-Americanization for all advanced technologies to make sure free from a blackmail from the US and able to sell it to China as China would demand that. It is just a matter of time they will succeed
I meant TWO years, really.
They are comparing the first six months of 2021 to the first six months of 2020. So there is a full 12 months gap, i.e., one year.
That means that the 47 percent refers to 12 monts. If you compound 1.47 to the second power, you get 2,16. So, technically you should double about every 22 months give it or take.
I don't know if you remember but this is a geometric progression using the formula (1+i) to the nth power.
- Status