According to the whistleblower MEGAsizeGPU, Nvidia is planning to launch a new model equipped with 9 GB of video memory based on the existing RTX 5050 to distinguish it from the current version with 8 GB of video memory as standard. The whistleblower has accurately disclosed Nvidia-related product information many times before. The existing RTX 5050 uses 8 GB, 20 Gbps GDDR6 video memory. With a 128-bit video memory width, the entire card video memory bandwidth is 320 GB/s.
It is said that the new model will use 3 GB capacity GDDR7 memory particles, achieving a total of 9 GB memory capacity through a 96-bit memory bus. Although the bit width has been narrowed, with the support of 28 Gbps GDDR7, the total memory bandwidth can be increased to 336 GB/s, an increase of approximately 5%, while the memory capacity has increased by 12.5%.
In terms of core specifications, the 9 GB version is not expected to be adjusted and will still be equipped with the GB207 GPU based on the Blackwell architecture. It is equipped with 2560 CUDA cores and has a total card power consumption of 130 W. There are currently no signs of changes in frequency or core number.
According to reports, Nvidia also plans to make layout adjustments in higher-positioned models. A tweaked GeForce RTX 5060 will likely switch to the larger GB205 core currently used in the RTX 5070, which will be decentralized in the form of reduced specifications. Under this change, the number of CUDA cores of the RTX 5060 is said to be reduced from 6144 to 3840, and the memory bus is reduced from 192 bits to 128 bits. Such changes mean that AIC board manufacturers need to adopt a new PCB design adapted to GB205 and use a standard 8-pin power supply interface.
Sources said that this series of adjustments is related to the supply pressure of low-end models in the RTX 50 series, and many products including RTX 5050 and RTX 5060 are affected. The GeForce RTX 5060 12 GB version that appeared on the earlier roadmap has now been revealed to have been cancelled.