GeForce RTX 50 series is starting to become clear
Kopite7kimi, known for his closeness to Nvidia, clarified the GB202 GPU belonging to the “Blackwell” architecture in his recent X post. For those who missed it, this GPU is expected to be one of the flagship solutions of the RTX 50 series, and the first leaks were that it would offer a 512-bit bus. However, Kopite7Kimi stated that the 384-bit bus would be preferred instead. It is worth remembering that this trend has been used in Nvidia architectures for about ten years. It seems that tradition will not be broken in the RTX 50 series.
GDDR7 memories and 1.5 TB/s bandwidth
Going back to the RTX 50 “Blackwell” architecture, it is stated that the architecture will offer a significant increase in bandwidth with its GDDR7 memory configuration. Additionally, the roadmap leaked from Micron shows that 32 Gbps speeds will be reached in GDDR7 consumer memory by the end of the year. This value indicates 50 percent higher speeds of 1.5 TB/s compared to the RTX 4090 using 21 Gbps GDDR6X memory.
However, given the launch date, lower GDDR7 frequencies may be available. In short, we are more likely to see these speeds in Super or Ti renewals. The release date of the RTX 50 Blackwell series is still unclear. However, it has been confirmed that the B100 and GB200 Blackwell graphics chips to be used on the corporate side will be released between 2024-2025. In this context, we can say that consumer GPUs will be introduced in a similar time period.