NVIDIA GeForce RTX 50 sticking to similar memory bus specs

It is this time of the year when the first next-gen rumors are starting to take shape.

As per information from Kopite7kimi, gamers shouldn’t anticipate a 512-bit memory bus for the RTX 50 series. Although this credible hardware leaker had hinted at such a scenario previously, it has now been reiterated that the leading RTX 50 GPU, known as GB202, will maintain its 384-bit memory bus configuration once more.

In fact, the entire GB20X GPU family is said to feature similar memory bus specs to AD10X (RTX 40). The only difference is that NVIDIA is said to change the naming for one processor from ’04’ to ’05, as previously revealed.

While the RTX 50 series GPUs are to retain similar if not identical memory bus specs, there will be an upgrade from GDDR6/GDDR6X to GDDR7 specs, now officially confirmed by JEDEC .

The new standard will once again increase the memory bandwidth per module and will allow NVIDIA to explore new capacity options as the new standard enables capacities like 3GB per module. This opens new opportunities for NVIDIA, and one cannot assume that the memory capacity specs will be identical to RTX 40 entirely. NVIDIA’s partner for GDDR6X and GDDR5X launch, Micron, has confirmed that 24Gb (3GB) modules are planned for 2025.

Micron planning GDDR7 with 2GB and 3GB capacity, Source: Micron

The NVIDIA RTX 50 series is now expected to debut by the end of this year. Thus far we have only learned the basic specs of the flagship GB202 GPU which is expected to feature upwards of 24K CUDA cores. The Blackwell GPUs are also said to use TSMC N3 node and will support DisplayPort 2.1.

Source: Kopite7kimi