According to tipster @Kopite7kimi, NVIDIA is now considering two possible specs for its RTX 4070 graphics card.
The RTX 4070 will reportedly be available in 10GB and 160bit GDDR6 memory sizes, with an “update” to 12GB 192bit later, and neither option seems to have been dismissed yet.
Kopite has given two sets of four different SKU specifications, 340/341 and 336/337, one of which may be used for the Founders Edition, while the other is a semi-custom design.
The 340/341 models are based on the full AD104 GPU (AD104-400) core with 7680 FP32 (CUDA) core. This model is said to have 12GB of GDDR6X memory across a 192bit bus and 21Gbps modules, which means the new cards can have a maximum bandwidth of 504 GB/s.
Meanwhile, the 336/337 SKU is said to be based on a streamlined AD104 GPU (possibly AD104-275) core with 56 stream processors, or 7168 CUDA cores.
In addition, the card’s memory size is limited to 10 GB, resulting in a narrower bus width of 160 bit. since both models are shown at 21 Gbps, the 336/337 SKU will have a reduced bandwidth of 420 GB / s.
It’s also worth noting that the new card’s TBP specs have also been lowered. kopite earlier claimed that the total board power was 300W, but he now says it’s 285W or 250W. Although the TBP is different from the TDP specs, it’s still higher than the RTX 3070 (non-Ti) models, so
It looks like the RTX 40 series is really trading higher power for more performance.
Nvidia will be held on September 19-22 GTC online conference, explore cutting-edge technology innovation, when it is likely to bring a new generation of RTX 4090 graphics cards, a small probability will have 4080 and 4070 graphics cards. From the previous information, RTX 4070 and RTX 4080 will arrive at the end of the fourth quarter
