Chinese semiconductor industry

Status
Not open for further replies.

Tyler

Captain
Registered Member
Please, Log in or Register to view URLs content!
If Baidu Kunlun 2 is really 3x faster than the Kunlun 1, then it is a phenomenal achievement for the Chinese semiconductor industry. With such a performance it will outperform the A-100 in some AI applications. Also, a 200% performance increase in a timespan of 3 years is rare for the modern semi industry.
Local Chinese GPU are needed for bitcoin mining and graphics intensive Chinese video games.
 

krautmeister

Junior Member
Registered Member
Please, Log in or Register to view URLs content!
If Baidu Kunlun 2 is really 3x faster than the Kunlun 1, then it is a phenomenal achievement for the Chinese semiconductor industry. With such a performance it will outperform the A-100 in some AI applications. Also, a 200% performance increase in a timespan of 3 years is rare for the modern semi industry.
Local Chinese GPU are needed for bitcoin mining and graphics intensive Chinese video games.
Baidu primarily uses the Kunlun I for their cloud service and I believe Kunlun II will be primarily used for the same purpose. These are both FPGA based AI optimized chips with low production runs. The Kunlun I only had a total production run of around 20000 chips. The Kunlun 2 is dependent on Intel FPGAs. Intel's FPGAs are at 7nm and industry leading in power and performance. Much of the Kunlun 2 AI chip performance is related to this. If Baidu wanted to focus on performance alone, they wouldn't be using any FPGAs which are "RELATIVELY" slow and expensive. This is clearly because of the specialization for their cloud services. I expect the Kunlun II to serve as Baidu's newest cloud service chip.
 

Tam

Brigadier
Registered Member
Baidu primarily uses the Kunlun I for their cloud service and I believe Kunlun II will be primarily used for the same purpose. These are both FPGA based AI optimized chips with low production runs. The Kunlun I only had a total production run of around 20000 chips. The Kunlun 2 is dependent on Intel FPGAs. Intel's FPGAs are at 7nm and industry leading in power and performance. Much of the Kunlun 2 AI chip performance is related to this. If Baidu wanted to focus on performance alone, they wouldn't be using any FPGAs which are "RELATIVELY" slow and expensive. This is clearly because of the specialization for their cloud services. I expect the Kunlun II to serve as Baidu's newest cloud service chip.

It doesn't sound like they are using FPGAs, which should only be for prototyping. These are straight out ASICs.
 
Status
Not open for further replies.
Top