Leaker claims two GeForce RTX 4070 specifications are now being considered

NVIDIA has not made a final decision on RTX 4070 specs yet

According to Kopite7kimi, NVIDIA is now considering two possible specifications for its RTX 4070 graphics card. 

One step forward, one step back. Initial rumors claimed RTX 4070 will launch with 10 GB and 160-bit GDDR6 memory configuration, later to be ‘updated’ to 12GB and 192-bit spec. It appears that neither of the options is off the table yet.

Kopite is now providing specs for four different SKUs: 340, 341, 336 and 337. Each variant will have a separate board design, but essentially 340/341 and 336/337 pairs denote the same SKU only one will likely be used for the reference deign (Founders Edition), while the other goes to semi-custom designs by board partners.

The 340/341 is supposedly based on full AD104 GPU (AD104-400) with 7680 FP32 (CUDA) cores. This model is said to feature 12GB of GDDR6X memory across a 192-bit bus and 21 Gbps modules. This means a maximum bandwidth of 504 GB/s.

Meanwhile, 336/337 SKU is supposedly based on a cut-down AD104 GPU (possibly AD104-275) with 56 Streaming Multiprocessors, so 7168 CUDA cores. Furthermore, this card would have limited memory size to 10 GB and as a result narrower bus width to 160-bit. Since both models are listed with 21 Gbps speeds, the 336/337 SKU would get 84 GB/s lower bandwidth.

Whichever variant will be chosen, the TBP specs have also been reduced since previous rumor emerged. Kopite earlier claimed that the total board power is 300W, but now it is said to be 285W or 250W depending on the SKU. Total board power is not the same spec as TDP though, but it should still be higher than RTX 3070 non-Ti.

The RTX 4070 is expected to make a debut in October alongside RTX 4090 and RTX 4080. However, its launch date is rumored to be in late fourth quarter.

Source: @kopite7kimi