According to Tweaktown, which cites XpeaGPU, NVIDIA has stated that it would incorporate the GDDR7 memory standard with the top three GPU ASICs powering the upcoming “Blackwell” GeForce RTX 50-series graphics cards. By this, we imply that NVIDIA will separate the majority of its SKUs from the top three physical silicon types. These would comprise the GB202, GB203, and GB205, which will power the RTX 4090 and all subsequent models up to the current RTX 4070. These chips are anticipated to be built by NVIDIA on the TSMC 4N foundry node.

The “Blackwell” generation of GPU ASICs will have several variants that adhere to older memory standards like GDDR6 or perhaps GDDR6X. These would be the replacements for the existing AD106 and AD107 ASICs, which power SKUs like the RTX 4060 Ti and lower. The only provider of the chip to NVIDIA, Micron Technology, collaborated with NVIDIA to co-develop the GDDR6X standard. Given that GDDR6X scales up to 23 Gbps and 16 Gbit, NVIDIA can leverage a lot of performance for the bottom end of its product stack. This is especially important given that its GDDR7 implementation will only operate at 28 Gbps, even though chips with 32 Gbps or even 36 Gbps are now on the market. The technology scales up to 20 Gbps even if NVIDIA decides to stick with the traditional GDDR6 for its entry-level mainstream processing.