On July 8th, Samsung Electronics announced the mass production of PM1763, the company's first PCIe based product ®  6.0 Enterprise Solid State Drive (SSD), with a clear core positioning, is an enterprise level storage hardware specially customized by Samsung for Nvidia's next-generation Vera Rubin AI computing platform.

As the amount of data required for AI training and inference continues to grow rapidly, enterprise grade SSDs (eSSDs) that can provide data quickly and reliably are becoming increasingly important for AI infrastructure.

With high-speed data transmission and optimized main control architecture, PM1763 is expected to become a key storage solution for high-performance AI platforms. PM1763 adopts Samsung's 9th generation V-NAND and newly developed 4-nanometer (nm) controller, significantly improving performance and energy efficiency.

It is reported that PM1763 is Samsung's first eSSD equipped with 9th generation V-NAND flash memory and a new PCIe 6.0 controller with a 4-nanometer process. The new controller has undergone architecture optimization, significantly improving data processing efficiency. This product supports NVMe 2.1 technology standards and offers three capacity options: 4TB, 8TB, and 16TB.

Compared with the previous generation product PM1753, the sequential read and write speed of PM1763 has doubled. Taking the 16TB version as an example, its sequential read speed reaches 28400MB per second and sequential write speed reaches 21900MB per second. This speed means that transferring a large 40GB language model only takes 1.4 seconds (based on the Meta Llama-3 70 billion parameter model calculation). During AI training and inference, SSD is responsible for delivering massive amounts of information, including AI model data and key value (KV) cache data, to CPU and GPU. PM1763 effectively ensures the continuous and efficient operation of AI computing accelerators and processors by minimizing data transmission latency.

PM1763 has been specifically optimized for the liquid cooling environment of next-generation AI servers, using a direct to chip (D2C) cooling method that connects the cooling board directly to the components and parts, ensuring no performance degradation during long-term operation under high loads. At the same time, its energy efficiency has increased by 1.8 times compared to previous generation products, which helps significantly reduce the operating costs of data centers.

In terms of security, PM1763 is equipped with Post Quantum Cryptography (PQC) algorithm, which can resist attacks from quantum computers at the source. In addition, the product also supports TDISP (TEE Device Interface Security Protocol), which can protect data transmission channels in virtualized environments and prevent unauthorized external intervention.

*Disclaimer: The above content is reproduced on the WeChat official account of the semiconductor industry circle and does not represent the views and positions of our company. It is only for communication and learning purposes. If you have any questions or objections, please contact us.