NVIDIA may be preparing a new version of the GeForce RTX 5070, and this time the card could arrive with a larger GPU die than the one used in current models.
According to a recent hardware leak, a new GeForce RTX 5070 variant is expected to use the GB203 graphics die instead of the GB205 die typically associated with the RTX 5070. If accurate, this would be another example of NVIDIA using larger silicon in lower-tier graphics cards while disabling parts of the chip to match the intended product specifications.
This approach is not unusual for NVIDIA. The company has previously used different GPU dies across variants of the same graphics card, especially when it makes sense for manufacturing efficiency and chip yield management. A similar situation has already been seen with some GeForce RTX 5060 models. While standard RTX 5060 cards are based on the GB206 die, certain newer versions reportedly use the larger GB205 die, with parts of the chip disabled to deliver the same performance target.
Now, the same strategy may be coming to the RTX 5070.
The GB203 die is currently associated with higher-end Blackwell graphics cards such as the GeForce RTX 5070 Ti and RTX 5080. It is significantly larger than GB205, measuring around 378 mm² compared to the GB205 die’s 263 mm². That makes GB203 roughly 43% larger.
However, if NVIDIA uses GB203 for a standard GeForce RTX 5070, the chip would need to be heavily cut down. The full GB203 configuration reportedly includes 10,752 CUDA cores, while the RTX 5070 is designed around 6,144 CUDA cores. To match that specification, NVIDIA would need to disable a large portion of the chip, bringing the shader count down to the same level as the current RTX 5070.
Memory configuration would also need adjustment. The GB203 die supports a wider memory interface, but the RTX 5070 is expected to retain a 192-bit memory bus with 12 GB of GDDR7 memory. To make that work, NVIDIA would likely disable two of the eight available memory channels on GB203.
The reason for this possible change has not been officially confirmed. However, one likely explanation is chip binning. In GPU manufacturing, not every large die is good enough to become a higher-end product. Some GB203 chips may not meet the requirements for an RTX 5070 Ti or RTX 5080, but they may still function well enough when cut down for an RTX 5070. Instead of discarding those chips, NVIDIA can repurpose them into a lower-tier model.
For buyers, this does not necessarily mean a major performance difference. If NVIDIA configures the GB203-based RTX 5070 to match the existing RTX 5070 specifications, gaming performance should remain largely the same. The main differences may appear in power behavior, thermals, manufacturing variation, or board design, depending on how partners implement the new variant.
The leak also suggests that a new NVIDIA graphics driver could arrive soon, possibly with early support for this revised RTX 5070 model. However, there is no official confirmation that the driver is directly connected to the rumored GB203-based RTX 5070.
If the report proves accurate, the GeForce RTX 5070 lineup may soon become more diverse, with some cards using the original GB205 die and others built around a cut-down GB203 chip. This would give NVIDIA more flexibility in managing Blackwell GPU production while keeping the RTX 5070 positioned as a 12 GB GDDR7 graphics card for high-performance gaming PCs.
