The GeForce RTX 5090 video card (probable name), according to recent data, will still not receive a 512-bit memory bus, but will be limited to 384-bit.
Created by DALL-E
The GB202 GPU, which will power Nvidia's new gaming flagship, will reportedly have a 384-bit bus, just like the current AD102. A 512-bit bus would significantly increase memory bandwidth, but would make the final product more expensive. As we remember, the RTX 4090 is still very expensive.
But the new product will receive GDDR7 memory, so the throughput in any case will be much higher than that of the RTX 4090. If we take Micron memory with an effective frequency of 32 GHz and the same 384-bit bus for calculations, then the throughput will be 1.5 TB/s! And without a 512-bit bus, this is one and a half times more than the RTX 4090.
By the way, the name of the GPU (GB202) suggests that the Blackwell architecture will probably be used by Nvidia both for new specialized accelerators and for gaming video cards. There will probably still be differences in GPU configuration, but now we have two different architectures for different segments: Ada Lovelace and Hopper.
Insider kopite7kimi was once the first to accurately describe the parameters of GeForce RTX 30 cards, and he also described RTX 40 cards more than once. Overall, it is one of the most reliable sources in this segment.
