Samsung will release the first artificial intelligence accelerator with Baidu

Samsung will release the first artificial intelligence accelerator with Baidu

Baidu, the largest Chinese search engine and South Korean company Samsung Electronics. it has announced the creation of the first Baidu artificial intelligence accelerator chip. Baidu KUNLUN Processor, to work on a wide variety of platforms, from the cloud to its surroundings. Is designed for Companies are set to launch it into mass production early next year.

The Baidu KUNLUN chip is based on the development of Baidu XPU. The proprietary architecture of neural processors for cloud and boundary computing. Utilizing 14-nm process technology and Samsung’s iCube packetization technology.

The new processor provides 512 gigabytes per second (GB / s) of memory bandwidth. And supports 260 teaspoons per second (tops) with a power consumption of 150W. In addition, Chip supports Ernie, a pre-trained neural model for processing natural language data. Which makes logical execution three times faster than traditional GPU / FPGA acceleration models.

artificial intelligence chips

Thanks to its large computing power and low power consumption. The new processor will enable Jodio to perform large-scale tasks with high performance. Including starting large-scale workloads using artificial intelligence, e.g. As such, pedal pedals such as search results ranking, speech recognition, images and natural languages, as well as the development of autonomous systems and deep learning platforms.

As part of the pilot collaboration, Baidu will provide AI with its first AI platform for maximum performance. And Samsung will be designing high-performance computing (HPC) processors in its semiconductor factories for both cloud and peripherals. Will

Because various applications, such as artificial intelligence workloads. Or HPC computing, require consistently high performance. The chip integration technology is becoming increasingly relevant today.

Samsung’s iCube integration technology, which combines logic chip. And high bandwidth (HBM2) memory with one interceptor. It allows even greater density and bandwidth in minimal sizes using various Samsung solutions. ۔

Compared to previous developments, new solutions can significantly increase processor performance. And provide more than 50% increase in computing power/signal integrity. It believed that the introduction of i-CubeTM will mark the beginning of a new era in the diversified computing market.

In addition, Samsung is developing even more advanced packaging technologies. Such as regenerative distribution layer (RDL) technology, as well as integrated cases with 4x and 8 x HBM memory.

Intel acquires Israeli chip maker for AI for $ 2 billion

Intel announced the acquisition of Data Center – Habana Labs, an Israeli developer of deep learning program accelerators for about $ 2 billion. According to Intel, the acquisition strengthens its product portfolio for Artificial Intelligence (AI). And is accelerating its efforts in the new and fast-growing market for Artificial Intelligence. According to Intel, It is expected to reach 25 billion by 2024. The dollar will exceed.

Habana Labs will remain an independent business unit and will continue to lead its current management team. Habana will report to the Intel Data Platform Group, which has a wide range of Intel technology for AI class data centres. As a result, Habana will gain access to Intel A’s resources, including those built in the last three years, with extensive experience in AI software, algorithms and research that will help Habana grow.

Also Read:

Source = Samsung

Leave a Reply

Your email address will not be published. Required fields are marked *