At yesterday’s “2019 Flying Eco-Partners Conference”, Baidu Kunlun Chip General Manager Ouyang Jian revealed that its Kunlun AI chip is adapting to domestic slot servers, to do performance tuning work. Kunlun AI chip is based on Baidu’s self-developed XPU neuroprocessor architecture, using Samsung 2.5D I-Cube packaging technology, through the intermediary layer to connect the SoC main chip and two HBM2 high bandwidth memory, unified package on a substrate.
Provides up to 512GB/s memory bandwidth and supports PCIe 4.0 x8, providing 260TOPS (260 trillion operations per second) at 150W. The chip is expected to be mass produced early next year.
Ouyang Jian introduced, and the fit of the flying server, including three aspects: computing power, architecture, technology.
At the new computing level, AI’s demand for computing power is endless, the soar itself is a high-performance 64-bit processor, Kunlun chip based on soaring, can bring new computing power.
At the new architecture level, Flying is a common computing architecture from the cloud to the end, with the same architecture supporting different scenarios. Baidu Kunlun just adopts the XPU architecture, the combination of the two has the opportunity to create a new architecture, Baidu Kunlun called the “universal architecture.”
At the new technology level, the iteration of the information industry, the underlying technology such as chips and operating systems, has always been dominated by foreign giants. But every iteration, is a shuffle opportunity, currently in the era of AI-smart cloud, chip architecture has changed a lot, but also emerged a lot of new operating systems. Baidu Kunlun and domestic OS, plus domestic in-depth learning framework, coupled with the application of a new era, there will be new technology, this technology will be the national production of technology.