Nvidia only uses GDDR7 on some RTX 50 series graphics cards

At the GTC 2024 conference held at the San Jose Convention Center in California, Jensen Huang, the founder, and CEO of Nvidia, introduced products based on the Blackwell architecture for data centers, which in turn has gradually shifted the focus towards gaming graphics cards also under the Blackwell architecture. Although there has been considerable speculation regarding the GeForce RTX 50 series, it’s expected that these products won’t make their debut until the end of 2024.

Samsung GDDR7 memory bandwidth

According to Tweaktown, the GeForce RTX 50 series will undoubtedly feature new memory technologies, though not all new-generation graphics cards will do so. Nvidia plans to support GDDR7 memory only on its higher-end GB20x series, specifically the GB202, GB203, and GB205 models, which span the product line from the RTX 5070 to the RTX 5090. These models are rumored to support 16Gb (2GB) GDDR7 modules, with a potential for 24Gb (3GB) modules as well, achieving speeds of 28 Gbps, which is below the standard 32 Gbps.

The lower-tier GB206 and GB207 models will continue to use the current generation of memory, though it’s unclear whether this will be GDDR6 or GDDR6X. GDDR6X memory is currently available in 16Gb (2GB) modules with speeds up to 23 Gbps, sufficiently meeting the demands of mid-range and entry-level products. Even standard GDDR6 has seen speed increases to 20 Gbps, with potential for further enhancements in the future, making it more than adequate for entry-level graphics cards.

As per previous statements, the GB20x series, including the GB202, GB203, GB205, GB206, and GB207 chips, will feature memory bus widths of 512/384-bit, 256-bit, 192-bit, 128-bit, and 128-bit, respectively, all manufactured using TSMC’s custom 4NP process.

During this GTC 2024 conference, Samsung showcased the GDDR7 memory prepared for the new generation of GeForce graphics cards.