Get the App
SLTechnology News&Howtos  ›  IT Information  › 

It is reported that Samsung and SK Hynix are working on developing technologies to reduce memory power consumption and improve the efficiency of high-performance computing.

Shulou Source: shulou.com Published: 2023-11-24 18:04:04 10月04日 Update

CTOnews.com July 13 news, with the increasing number of model training tasks, the importance of memory in high-performance computing such as HBM and DDR5 is increasing, and the power consumption of memory is becoming more and more prominent.

According to foreign media Chosun Biz, memory manufacturers such as Samsung Electronics and SK Hynix are working with academia to invest in next-generation technologies to reduce power consumption and improve efficiency in high-performance computing.

According to CTOnews.com, DRAM currently accounts for 40% of the total power consumption of Nvidia A100-based data center platforms, and the power consumption of HBM will increase as the number of layers increases.

Seoul National University has previously launched DRAM Translation Layer technology, which claims to reduce the power consumption of DRAM by 31.6%, which has attracted Samsung's interest.

▲ Picture Source Seoul National University thesis

▲ Samsung is currently working with Seoul National University to further reduce memory power consumption. Based on Compute Express Link technology, they have developed a 16GB DDR5 DRAM with a 12-nanometer process, which reduces power consumption by 23% compared with the previous generation.

▲ source Compute Express Link website SK Hynix is the launch of LPDDR5X, the High-K metal gate (HKMG) process is applied to mobile DRAM. The dielectric constant of High-K material is about five times higher than that of traditional SiON insulating film, which can store five times as much charge at the same area and thickness and help reduce current leakage. By controlling the leakage current, SK Hynix's LPDDR5X speed has been increased by 33%, and the power consumption has been reduced by more than 20% compared with the previous generation.

▲ image source Pexels with the model training task put forward further technical update requirements for memory manufacturers, manufacturers are now competing with each other, resulting in a certain reduction in memory costs, consumers can also benefit from it.

Tags: Power consumption memory technology university national Samsung Lux high performance products tasks manufacturers processes models current papers previous generation into one cooperation Apple Docker Huawei Linux macOS MariaDB Microsoft MySQL NVidia OPPO Reno Huawei Shulou Information macOS Redmi Docker