It is rumored that NVIDIA is considering two specifications for the RTX 4070, and it is also possible that both specifications will be launched.
The NVIDIA RTX 4070 was originally rumored to have 160bit GDDR6 10GB of memory, which was later updated to 192bit GDDR6 12GB, but it’s possible that these are two specs, not spec revisions.
There are currently rumored that there are four different SKUs: 340, 341, 336, and 337. Each has a separate circuit board design, but essentially 340/341 and 336/337 pairs denote the same SKU. Two of the same model may be distinguished as one is the FE version, and the other is the special version designed by the partner manufacturer.
The 340/341 is rumored to be a full AD104 GPU (AD104-400) with 7680 CUDA cores, this model has 192bit GDDR6X 12GB of memory at 21Gbps, which means a maximum bandwidth of 504GB/s.
The 336/337 is a cut-down AD104 GPU (probably AD104-275), with 7168 CUDA cores, with 160bit GDDR6X 10GB memory, the same speed is 21Gbps, and the bandwidth is reduced by 84GB/s to 420GB/s.
Some recent rumors about power consumption have been revised downward. The earliest information is that the RTX 4070 is 300W, but it may be reduced to 285W and 250W.
The NVIDIA RTX 4070 is expected to be announced in October along with the RTX 4090
and RTX 4080, but the actual release time may be at the end of the year.
Source: @kopite7kimi | Via: videocardz