Following yesterday’s purported RTX 40 “Super” leak, additional details about the RTX 4070 Super have surfaced. NVIDIA is allegedly prepping a GeForce RTX 4070 Super based on the AD103 GPU (used by the RTX 4080). The source doesn’t mention the core counts or SMs but believes the GPU will be paired with 16GB of GDDR6X memory across a 256-bit bus. The GeForce RTX 4070 features 12GB of GDDR6X memory across a 192-bit bus.
Coming to the shader configuration, an AD103-based core can only go a few ways. The RTX 4080 packs 9,728 cores across 76 SMs, while the RTX 4070 consists of 5,888 cores and 46 SMs. This leaves the RTX 4070 Super with a few viable options. The first and simplest one would be to keep the shader count unchanged and beef up the memory and bandwidth, improving 4K performance.
The second would be to increase the shader count to 7,680 across 60 SMs. Disabling the smaller 8 SM GPC and another 12 SM GPC would achieve this. So would disabling two SMs from each GPC. NVIDIA could also opt for a small shader bump, going from 5,888 (46 SMs) to 6,400 (50 SMs) or 7,168 (56 SMs) cores.
NVIDIA may also launch a cheaper GDDR6 variant of the RTX 4070 with the same shader configuration as the base SKU. This card would slot between the RTX 4060 Ti and 4070 and act as an RTX 4060 Super. At 1080p and 1440p, the performance delta between the two would be within the margin of error.
Source: MegasizeGPU.