Data Centre Dynamics reviews that firm Baidu has created a separate, impartial semiconductor-focused enterprise, aptly named Kunlun Chip Know-how Firm, valued at round $2 billion final June 2023, that may produce its next-gen Kunlun Core II chips.
Baidu’s Kunlun Core II Chips Enter Quantity Manufacturing, Will Deal with NVIDIA’s A100 in AI
Kunlun Chip Know-how Co. has begun to fabricate their Kunlun Core 2 processor, utilized for synthetic intelligence purposes. The Kunlun Core 2 processor shares the identical microarchitecture as second-generation XPUs. It’s created utilizing 7nm expertise and can supply as much as 3 times the efficiency of Kunlun’s different predecessors.
Three years in the past, Kunlun launched details about the Kunlun K200, Kunlun’s first-generation processor was created for the sting, cloud, and autonomous vehicular purposes. The chip showcases as much as 256 INT8 TOPS execution, about 64 TOPS INT/FP16 execution, and 16 INT/FP32 TOPS execution, topping at 150 watts of energy.
Beneath is a comparability chart of the Baidu Kunlun Core 2, additionally referred to as the Kunlun II, versus the first-gen Baidu Kunlun Core and NVIDIA’s A100 chips. This chart reveals how the brand new Baidu Kunlun II is ready to sustain with NVIDIA’s A100 chipset that sometimes makes use of 19.5 FP32 TFLOPS with 624/1248 INT8 TOPS.
Baidu Kunlun II Comparability Chart
| 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 |
The Kunlun AI was initially created by the mother or father firm Baidu in 2011. They meticulously examined and tried to recreate their similar XPU microarchitecture utilizing many-small-core XPUs using FPGAs. Nevertheless, in 2018, Baidu constructed silicon that was dedicated to the Samsung Foundry’s 14nm fab processes that boosted efficiency round 14LPP.
14LPP (Efficiency boosted version) is the 2nd FinFET era which the efficiency is enhanced as much as 10%. 14LPP is. the only platform for each software designs with the improved efficiency for computing/Community designs and the lowered. energy consumption for Cellular/Shopper designs.
—Samsung SAS Enterprise website
Kunlun’s AI processor makes use of 8 gigabytes of HBM reminiscence that gives 512 gigabytes per second of peak bandwidth speeds. Within the final half of 2020, Wang Haifeng, Baidu’s Chief Know-how Officer, reported that the unique Kunlun Core produced over 20,000 chips and acknowledged the necessity for the corporate to create a bigger scaled deployment technique.
At present, Kunlun’s first-gen chips are used for mother or father firm Baidu’s cloud structured datacenters and utilized for his or her Apolong autonomous car platforms in addition to different AI purposes.
Supply: MyDrivers, Tom’s Hardware
