AMD Plans to Launch NPU Acceleration Card

AMD Plans to Launch NPU Acceleration Card

On July 31, according to foreign media reports from CRN, AMD Vice President and General Manager of Client Business Group Rahul Tikoo recently stated that AMD is researching the launch of a standalone NPU acceleration card to help achieve the goal of “AI computing accessible to everyone.”

Currently, AI PCs mainly rely on SoCs with built-in NPUs, while some desktop PCs and professional workstations with higher performance requirements still primarily depend on GPUs as AI accelerators. This leads to high power consumption due to prolonged high-load operation of GPUs, and they must share resources with graphics processing, which also limits the performance of certain AI applications. In contrast, if a standalone NPU acceleration card can provide dedicated computing performance at low power consumption, it will alleviate the burden on GPUs and CPUs, enhancing overall efficiency.

AMD may plan to leverage the technology acquired from Xilinx and expand it into a standalone NPU acceleration card. This product will be optimized for AI inference and will play a role in high-speed computing scenarios such as generative AI and LLM inference.

As Lenovo, Dell, and HP begin to explore PCs equipped with standalone NPUs, the market demand for dedicated AI accelerators is heating up. Currently, this AMD NPU is still in the preliminary evaluation stage, and the company will continue to monitor market developments and, if appropriate, integrate its CPU, GPU, and NPU technologies to gradually advance its AI hardware ecosystem.

Editor: Lin Zi, Chip Intelligence News

Previous Exciting ArticlesH20 has serious security risks, NVIDIA has been summoned!The chip equipment giant emerging from Hong Kong, ChinaWith a total investment of 4.25 billion yuan, 8 major integrated circuit projects signed in ChongqingRumored AMD to increase MI350 series prices by 66.7%!The era of AI inference has arrived, CloudWalk Technology uses “computing power building blocks” architecture to solve localization challengesNew server chip KH-50000 released by Zhaoxin: 3GHz, 96 cores!LiShan’s first GPU officially released, performance catching up with NVIDIA RTX4060!Q2 loss of 2.9 billion USD! Intel cancels factories in Germany and Poland, and will lay off another 20%!Dimensionality reduction attack! The new generation 4GHz C86 processor enters the market, domestic terminals move from “usable” to “easy to use”!Another domestic GPU manufacturer starts IPO guidance, valuation exceeds 10 billion yuanDeep analysis of Alibaba’s Xuantie C930: frequency exceeds 3.4GHz, SPECint2006 score 15.2/GHzPerformance benchmarks against Arm N2! Domestic open-source RISC-V CPU helps reduce chip design costs by 33%!NVIDIA officially announces: CUDA will fully support RISC-V architecture!H20 is unblocked! Jensen Huang visits China again, this time wearing a leather jacket!Plans to raise 8 billion yuan! Moores Threads IPO accepted: cumulative losses exceed 5 billion yuan over 3 years!Plans to raise 3.9 billion yuan! Muxi Co., Ltd. IPO accepted: cumulative GPU sales exceed 25,000 units!

For industry communication and cooperation, please add WeChat: icsmart01Chip Intelligence News official communication group: 221807116

Leave a Comment