NVIDIA is reportedly preparing a new version of the GeForce RTX 5070 desktop graphics card using a much larger processor die. According to hardware leaker hongxing2020 on social media, the company plans to ship select RTX 5070 models equipped with a modified GB203 GPU rather than the standard GB205 chip. This move allows the chipmaker to repurpose imperfect silicon that fails to meet higher tier requirements.
Standard retail models of the GeForce RTX 5070 rely on the GB205 processor, which operates with 6144 CUDA cores across a 192 bit memory bus. A fully enabled GB203 processor packs 10742 CUDA cores and a 256 bit memory bus width. To fit this larger chip into RTX 5070 board designs, NVIDIA must disable roughly 42.8% of the active CUDA cores along with 25% of the memory bus width.
The resulting graphics card will match the official specifications of standard models. Gamers should see identical clock speeds, memory bandwidth, and power limits regardless of which silicon sits under the cooler. Board partners will simply drop the reconfigured dies into existing card layouts.
Repurposing higher tier silicon is a regular practice for NVIDIA. Fabricating massive chips like the GB203 regularly produces defective dies where specific clusters fail factory quality tests. Disabling the faulty sectors lets the company sell those chips in lower tier cards instead of throwing them in the trash.
NVIDIA used the same strategy by deploying salvaged GB205 dies into lower tier models in the RTX 50 series lineup. For PC builders, these stealth silicon swaps usually roll out quietly into retail channels without changes to pricing or official product branding.
