Lo and behold: NVIDIA appears to be working on yet another variant of the budget Turing GPU, the GTX 1650. This one will be based on the same die as the GTX 1660 and the 1660 Ti, the TU116. As already reported, there are a total of 10 models in the GTX 1650 family, including the Supers, Tis and the mobile variants.
Other than inclusion of the faster GDRR6 memory, the primary specs are unchanged. You get 896 CUDA cores across a 128-bit bandwidth paired with 4GB of GDDR6 memory. The clock speeds will of course vary from OEM to OEM.
The reasons for this are two-fold. Firstly, the GTX 1650 is easily one of NVIDIA’s best-selling graphics cards as indicated time and again by Steam’s Hardware surveys. Secondly, as the Turing GPUs are phased out to make way for the upcoming Ampere lineup, the chipmaker will be keen on getting rid of any excess dies.
These TU116 dies to be used in the GTX 1650s are defective dies, originally meant for the higher-end 16 series cards. This is basically the process of recycling those dies. However, at the same time, it’s important to note that the encoder varies across GTX 1650 models, with some featuring the Volta encoder, some Pascal and the rest boasting the newer Turing encoder. This will make it harder for the mainstream audience to opt for the “best” variant. After all, it won’t be enthusiasts who’ll be looking to buy a GTX 1650. (Source)