At this point, it is unclear whether Nvidia's GeForce RTX 4060 Ti 8GB and 16GB will rely on the AD104 graphics processor in the same configuration with 4352 CUDA cores, but this is certainly a likely scenario.
The RTX 4060 Ti will utilize the AD106-350-A1 GPU core, and not 104, which is slightly a cut-down version of the full AD106 graphics chip. This should give the card 34 SMs or 4352 CUDA cores, 16/8 GB GDDR6 memory running at 18 Gbps across a 128-bit bus interface, and 288 GB/s of bandwidth.
There's also 32 MB of L2 cache. According to previous leaks, the 16 GB variant utilizes the PG190 SKU 363 PCB, while the 8 GB variant utilizes the PG190 SKU 361 PCB.
I think the 190 PCB might be used for the Founder's Edition model, and reference AIB cards should sport the PG173 board instead.
As for the vanilla GeForce RTX 4060, it will feature AD106 with 3072 CUDA cores and will therefore consume even less power, so expect low-profile and perhaps even single-slot graphics cards based on this GPU.
No, the RTX 4060 will actually sport the
AD107-400-A1 GPU, the
fully enabled AD107 die (PG173 SKU 371 PCB design). The current GeForce RTX 4060 Laptop GPU also comes in the same config.
So based on this, the card should get 3072 CUDA cores and 8 GB of GDDR6 memory running at 18 Gbps memory speeds across a 128-bit bus interface, within a 110W - 115W reference TDP. And obviously, it will also sport 24 MB of cache (L2).