SK Hynix plans to mass produce GDDR7 in 2025Q1

In March this year, JEDEC released the JESD239 Graphics Double Data Rate 7 (GDDR7) standard, which offers twice the bandwidth of GDDR6, catering to the growing demand for high memory bandwidth in future graphics, gaming, computing, networking, and artificial intelligence applications. This is in preparation for the next generation of graphics cards, with rumors suggesting that NVIDIA’s GeForce RTX 50 series GPUs will support GDDR7.

According to Anandtech, SK Hynix showcased GDDR7 samples at COMPUTEX 2024, announcing plans to begin mass production in the first quarter of 2025. Compared to its competitors, Samsung and Micron, SK Hynix appears to be the last to reach the large-scale production phase.

Samsung and Micron both aim to achieve mass production within 2024. Samsung’s GDDR7 product page went live on its official website in March, while Micron recently confirmed in a press release that it has provided samples to partners, planning to start shipments in the second half of 2024. Based on available information, SK Hynix’s progress on GDDR7 is approximately a quarter behind Samsung and Micron. Although SK Hynix’s mass production timeline is later, this does not mean Samsung and Micron have complete control. They have merely taken an initial step ahead in the verification process with their partners.

Unlike the NRZ/PAM2 signaling mechanism used in GDDR6 or the PAM4 mechanism co-developed by NVIDIA and Micron for GDDR6X, GDDR7 employs the PAM3 signaling mechanism. NRZ/PAM2 transmits 1 bit of data per cycle, PAM4 transmits 2 bits per cycle, while PAM3 transmits 3 bits of data every two cycles.

Based on the exhibited samples, SK Hynix plans to offer 16Gb (2GB) and 24Gb (3GB) modules, with speeds reaching up to 40Gbps, corresponding to a bandwidth of 160GB/s. Samsung and Micron’s initial GDDR7 chips have speeds of 32Gbps for 16Gb (2GB) modules. SK Hynix’s confidence may lie in its ability to provide faster and larger products.