CybserSecurity News

Tag: Rambus

  • Rambus releases DDR5 server PMIC

    Rambus has announced the launch of its new DDR5 RDIMM server memory dedicated PMIC (Power Management Integrated Circuit). This new series of server PMICs provides a complete chipset solution for DDR5 RDIMM memory modules, catering to a broad spectrum of data center applications.

    Sean Fan, Chief Operating Officer at Rambus, stated that advanced data center workloads, such as generative AI, require server RDIMMs with the highest bandwidth and capacity to meet the increasing demands of data pipelines. With the introduction of this next-generation server PMIC series, Rambus has expanded its product portfolio using cutting-edge chip technology, offering customers a comprehensive memory interface chipset that supports multiple generations of DDR5 server platforms.

    The PMIC is a critical component in DDR5 memory architecture, enabling more memory channels, higher-capacity modules, and increased bandwidth. This server PMIC series includes three chips: PMIC5000 (compliant with JEDEC high current specifications), PMIC5010 (compliant with JEDEC low current specifications), and PMIC5020 (compliant with JEDEC extreme current specifications).

    Key features include:

    • Industry-leading DDR5 server PMICs designed to support the highest performance and capacity memory modules required for AI and other advanced workloads.
    • The new PMIC series supports multiple generations of high-performance servers based on DDR5.
    • Provides a complete memory interface chipset for DDR5 server memory modules, including Register Clock Driver (RCD), PMIC, Serial Presence Detect (SPD) hub, and temperature sensor IC.

    Rambus has confirmed that the PMIC5000, PMIC5010, and PMIC5020 server PMIC series chips are now available on the market.

  • Rambus expects to have faster GDDR6 mass production by the end of this year

    Last year, Samsung began offering the industry’s first 16Gb (2GB) GDDR6 memory modules, boasting a speed of 24 Gbps. These modules, fabricated using Samsung’s 10nm-class 1z process technology, encompass innovative circuit design, advanced insulating material (HKMG), and extreme ultraviolet (EUV) technology. They come equipped with dynamic voltage scaling (DVS) technology, which adjusts the working voltage based on performance requirements, enhancing power efficiency by 20%. Their objective is to amplify the graphical performance of next-generation graphics cards, laptops, and gaming consoles. They also find applications in artificial intelligence-based applications and high-performance computing (HPC) systems.

    According to a report by Twitter user Hassan Mujtaba, Rambus expects faster GDDR6 to go into production by the end of this year, reaching speeds of 24 Gbps.

    Historically, GDDR6X has consistently boasted higher speeds compared to GDDR6. Micron began mass production of 24 Gbps GDDR6X last year, with a 16Gb module capacity. Currently, the GDDR6X memory used in the GeForce RTX 4080 is MT61K512M32KPA-24. The rated speed itself should be 24 Gbps, but it is set at 22.4 Gbps.

    With the speed of GDDR6 also reaching 24 Gbps, AMD may reap benefits, potentially appearing in the Radeon RX 7×50 series. It remains uncertain whether Nvidia will adopt it more widely in their products as the GDDR6 speed catches up. However, Micron has stated that the innovations in PAM4 signal transmission technology within GDDR6X make it more energy-efficient than other GDDR6 products.

    Compared to Samsung and Micron, SK Hynix appears to have placed its emphasis on HBM class memory chips, currently offering only GDDR6 with a speed of 20 Gbps.