As the next generation of best graphics cards is brewing, more and more info seems to be leaking about the design and what we might expect from it. Some of the latest info relates to the die structure that Nvidia seems to be preparing for its flagship GeForce chip in the GB202. It’s expected to be what is used in the flagship of the RTX 50 series in the RTX 5090 and that will make it quite the massive die and design according to known leaker kopite7kimi.
Nvidia keeping enterprise and gaming design separate
The Blackwell series of GPUs has already been featured and released in the enterprise world. With the HPC or AI chips such as the B100 and B200 out there in the world looking to power the craze of AI commuting. That opts for a split multi-chip design of computing power, something that AMD has utilized in the top range of its RDNA 3 GPUs like the RX 7900 XTX.
Instead, Nvidia looks to be keeping things simple by just keeping things to one. Even though the GB202 chip is expected to feature twice the number of SMs and cores as the GB203 (a cut-down version next tier die). If this comes to fruition it will put a massive disparity between the two tiers of graphics cards, of the top and second tier making it that more out of reach than the RTX 4090 already is. A closer AI enterprise GPU than a consumer-grade gaming card.
It could be possible Nvidia will split the chip itself in two much like it did with the GA100 and have great intercommunication between the two. But that solution would just keep inflating the price of the graphics card and the 90 series already seems way over the top, pushing that further would make it a lot less consumer GPU.
