Back in June, it was reported that Baidu has successfully spun off its semiconductor design business into an independent company and the company is now called the Kunlun Chip Technology Co. Who values it at around $2 billion?
Kunlun Chip is the wholly-owned subsidiary of Chinese high-tech giant Baidu, and the company has now entered into volume production of its Kunlun II processor for AI applications. Their new chip for AI is based on the 2nd generation XPU microarchitecture and is made of a 7 nm process technology.
Three years ago the first generation Kunlun K200 processor was released and this is designed for the cloud, edge, and autonomous vehicles applications. It offers around 256 INT8 TOPS performance, around 64 TOPS INT/FP16 performance, and 16 INT/FP32 TOPS performance while powered at 150 Watts.
The company has now announced that its latest generation of Kunlun processors will have about 2-3 times higher performance than its predecessor. If this is true then the new chip can provide from 512 to 768 INT8 TOPS, 128 – 192 INT/FP16 TOPS, and 32 – 48 INT/FP32 TOPS throughput. It can easily compete against the likes of Nvidia’s A100 in AI computations.
Baidu Kunlun II’s Relative Performance
| Baidu Kunlun | Baidu Kunlun II | Nvidia A100 | |
| INT8 | 256 TOPS | 512 ~ 768 TOPS | 624/1248* TOPS |
| INT/FP16 | 64 TOPS | 128 ~ 192 TOPS | 312/624* TFLOPS (bfloat16/FP16 tensor) |
| Tensor Float 32 (TF32) | – | – | 156/312* TFLOPS |
| INT/FP32 | 16 TOPS | 32 ~ 48 TOPS | 19.5 TFLOPS |
| FP64 Tensor Core | – | – | 19.5 TFLOPS |
| FP64 | – | – | 9.7 TFLOPS |
*With sparsity
For those who don’t know, the Kunlun project was initiated by Baidu back in 2011. Initially, its many-small-core XPU microarchitecture using FPGAs were emulated however in 2018 the company built dedicated silicon using one of Samsung Foundry’s 14nm fabrication process.
