The first batch of GPUs supporting GDDR7 use 16Gb modules

Recently, the JEDEC Solid State Technology Association announced the release of the JESD239 Graphics Double Data Rate 7 (GDDR7) standard, which promises to deliver bandwidth that doubles that of its predecessor, GDDR6. This development is designed to meet the escalating demands for higher memory bandwidth across various applications, including graphics, gaming, computation, networking, and artificial intelligence. This advancement is also a strategic preparation for the next generation of graphics cards. It has been reported that the GeForce RTX 50 series flagship GPUs will transition from GDDR6X to GDDR7, marking a significant technological leap.

In a recent revelation, it has been noted that the first batch of GPUs supporting GDDR7 will utilize 16Gb (2GB) modules, which represents the minimum capacity outlined in the newly published GDDR7 standard by JEDEC.

The predecessor, the GeForce RTX 30 series, initially incorporated 8Gb (1GB) modules, and it was not until the RTX 3090 Ti that 16Gb modules were introduced. This standard is continued in the current GeForce RTX 40 series products with GDDR6X, ensuring a memory allocation of 24GB, 16GB, 12GB, and 8GB for 384-bit, 256-bit, 192-bit, and 128-bit memory bus widths, respectively.

However, Nvidia might opt for a non-JEDEC standard version of GDDR7 for some products, preferring instead to co-design a proprietary standard with a DRAM manufacturer that aligns with its graphic architecture’s memory bus signal and power management technologies. Previously, with GDDR6X, Nvidia had selected Micron as its partner. With the advent of GDDR7, it is conceivable that Nvidia might continue this collaboration with Micron.

According to Micron‘s previously disclosed product roadmap, GDDR7 is expected to arrive by the end of 2024, with each data I/O interface achieving speeds of up to 32 Gbps, and capacities ranging from 16Gb to 24Gb. This opens the possibility that the initial wave of GDDR7-supported graphics cards may also include models equipped with 24Gb (3GB) modules, thereby offering memory capacities of 24GB, 18GB, and 12GB at 256-bit, 192-bit, and 128-bit memory bus widths, respectively.

Should GPU manufacturers wish to reduce manufacturing costs by narrowing the memory bus width, opting for 24Gb modules could provide a viable solution to compensate for any potential shortfall in memory capacity.