NVIDIA GeForce RTX 50 “Blackwell” GB200 Gaming GPUs To Retain Same Memory Interface As Ada, 384-bit At Max

NVIDIA’s GeForce RTX 50 “Blackwell” Gaming GPUs under the GB200 series are likely to retain the same memory bus configurations as Ada.

NVIDIA RTX 50 “Blackwell” & RTX 40 “Ada” Gaming GPUs Rumored To Incorporate Similar Memory Bus Configurations, Flagship GB200 Series Chip To Get 384-Bit Bus

The latest info on the next-generation NVIDIA GeForce RTX 50 “Blackwell” GPUs has been dropped by Kopite7kimi and it looks like the GB200 series of GPUs for gaming graphics cards will be getting similar memory bus interfaces as the existing Ada lineup designed for the GeForce RTX 40 Gaming GPU lineup.

Although I still have fantasies about 512 bit, the memory interface configuration of GB20x is not much different from that of AD10x.

— kopite7kimi (@kopite7kimi) March 9, 2024

The NVIDIA GeForce RTX 40 “Ada” GPU lineup currently consists of five SKUs which include the following:

• AD102 (384-bit)
• AD103 (256-bit)
• AD104 (192-bit)
• AD106 (128-bit)
• AD107 (128-bit)

We have seen NVIDIA go as low as 96-bit for its entry-level AD107 GPU SKUs but that’s just a few mobility configurations. The lineup peaks out with the GeForce RTX 4090 and the RTX 6000 Ada, both of which pack a 384-bit bus interface and up to 24/48 GB video memory capacities. Based on what we are seeing from the first generation of GDDR7 dies, it looks possible that NVIDIA is going to retain the same memory configurations for its next-gen GB200 “Blackwell” Gaming GPUs which are going to be featured on the GeForce RTX 50 family. We can expect something like the following:

• GB202 (384-bit)
• GB203 (256-bit)
• GB204 (192-bit)
• GB206 (128-bit)
• GB107 (128-bit)

If NVIDIA goes on to use the standard 16 Gb (2 GB) memory dies clocked at 32 Gbps pin speeds, the lineup should look like the following:

• GB202 – 384-bit / 32 Gbps / 24 GB (Max Memory) / 1536 GB/s (Max Bandwidth)
• GB203 – 256-bit / 32 Gbps / 16 GB (Max Memory) / 1024 GB/s (Max Bandwidth)
• GB204 – 192-bit / 32 Gbps / 12 GB (Max Memory) / 768.0 GB/s (Max Bandwidth)
• GB206 – 128-bit / 32 Gbps / 8 GB (Max Memory) / 512.0 GB/s (Max Bandwidth)
• GB207 – 128-bit / 32 Gbps / 8 GB (Max Memory) / 512.0 GB/s (Max Bandwidth)

The above configurations are just preliminary and NVIDIA is known to use lower memory clocks on its entry-level SKUs. Furthermore, the maximum memory noted above is only based on single-side configurations. The NVIDIA GeForce RTX 4060 Ti 16 GB was introduced with dual-sided DRAM dies so if that’s the case, the green team can offer doubled memory variants of each GPU side-by-side or as standard from the get-go. If NVIDIA sticks with 24 GB as the maximum VRAM, that would be the third flagship variant which features such as capacity with the previous ones being the GeForce RTX 3090 and the existing GeForce RTX 4090.

It will be interesting if NVIDIA doubles or offers a nice bump in the VRAM count for its entry-level variants as that has been the demand of several gamers since the release of the Ada lineup. AMD has also heated up the mainstream space with its higher VRAM offerings so those need to be accounted to and I think 8 GB would be a tad bit low for even 1080p gamers in 2024 given the increasing demand from high-res textures and high-refresh rate gaming monitors that are coming down the road.

Based on previous rumors, the NVIDIA GeForce RTX 50 “Blackwell” Gaming GPUs will be fabricated on the TSMC 3nm process node and feature DisplayPort 2.1 support along with similar FE Founders Edition cooling as the existing lineup but with a few new upgrades. The next-gen family is expected to arrive sometime in late 2024 which will be almost a year since the release of the RTX 40 “SUPER” family and two years since the release of the RTX 40 “Ada” lineup.

NVIDIA GeForce GPU SKUs: