THE graphics cards NVIDIA GeForce RTX 50 series will be based on Blackwell generation GPUs that take advantage of the finesse of TSMC’s 4N manufacturing node.
The news devoted to NVIDIA’s Blackwell B200 chip is interesting, but let’s recognize that these “toys” costing several tens of thousands of dollars each are still quite far from our realities.
On the other hand, if we talk about GeForce, it immediately concerns us more and rumors are accumulating around the next generation of NVIDIA graphics cards… which will share a Blackwell base with the B200.
The TSMC 4N… a 5 nanometer (nm) node
Indeed, you should know that NVIDIA’s new Blackwell architecture is a very broad set which will concern both colossal chips dedicated to the professional world and more “accessible” graphics processors (GPU).
Note the quotes on “accessible”, because we currently have no idea of the prices that NVIDIA will charge for the launch of its future GeForce RTX 50 cards. On the other hand, and the announcement of the B100, B200 and GB200 goes in In this sense, everything suggests that the next graphics cards will be marketed before the end of the year.
Cards whose GPUs will logically be based on these B100, B200 and GB200 with, in particular, the confirmation of the use of the same manufacturing node, the 4N from Taiwanese TSMC.
Higher transistor density
The “4N” node can, however, be misleading. If the “N” is not complicated to understand (it is there for NVIDIA), it is different for 4 since it is in reality only a 5 nm node.
After Intel and Samsung, if TSMC also starts to review the nomenclatures, it will become very complicated to find its little ones. Still, by confirming the use of 4N for the GB202 GPU, Kopite7kimi brings NVIDIA chips dedicated to AI a little closer to our “gaming” GPUs.
Note that 4N allows TSMC to be much more efficient than the N5 node. We are talking about a 6% higher transistor density and an improvement in electrical efficiency of around 22%. In addition, Kopite7kimi emphasizes that this node change is accompanied by an increase in the L1 cache for faster exchanges deep within the chip.
Source : VideoCardz
0