As per kopitekimi, a reputed source on NVIDIA Ampere leaks, the GeForce RTX 3050 will be based on the GA107-300 die with 2304 FP32 cores and a TGP of 90W. Although this wasn’t specified, it’s likely that the RTX 3050 will leverage a 160-bit bus in clamshell mode, with 6GB of GDDR6 memory.
NVIDIA could opt for a slimmer 128-bit bus and 4GB of memory, but seeing as both the RTX 3080 has been upgraded to a 320-bit from the usual 256-bit one, it’s very likely that Team will opt for a 160-bit memory interface and a larger VRAM buffer.
| Graphics Card Name | GPU | FP32 Cores | Memory | Memory Bus |
|---|---|---|---|---|
| NVIDIA GeForce RTX 3090 | GA102-300 | 5,248 | 12GB GDDR6X | 384-bit |
| NVIDIA GeForce RTX 3080 | GA102-200 | 4,352 | 10GB GDDR6X | 320-bit |
| NVIDIA GeForce RTX 3070 Ti | GA104-400 | 3,072 | 8GB GDDR6X | 256-bit |
| NVIDIA GeForce RTX 3070 | GA104-300 | 2,944 | 8GB GDDR6 | 256-bit |
| NVIDIA GeForce RTX 3060 Ti | GA106-400 | 3,840 | 6GB/8GB GDDR6 | 256-bit/192-bit |
| NVIDIA GeForce RTX 3060 | GA106-300 | 3,584 | 6GB/8GB GDDR6 | 256-bit/192-bit |
| NVIDIA GeForce RTX 3050 | GA107-300 | 2,304 | 4GB/6GB GDDR6 | 128/165-bit |
Earlier, the same source also reported the specifications for the RTX 3060 and 3050 Ti. Both these budget parts will be based on the GA106 die, featuring 3,840 and 3,584 FP32 cores, respectively. The bus width with either be the same as the RTX 3070 @ 256-bit or a cut-down 192-bit interface. Memory is most likely to stay at 8GB or 6GB in case of the latter.
NVIDIA will likely launch these parts early next-year, with the RTX 3060 Ti slated to launch next month to tackle the budget RDNA 2 GPUs from NVIDIA.
