Nvidia GeForce RTX 40 series GPUs are 4nm chips
Nvidia will bring the Ada Lovelace architecture this year to replace the existing Ampere architecture products. Recently, it has been reported that Nvidia’s new generation of GeForce RTX 40 series GPUs will be manufactured using a 4nm process, which may bring advantages in the manufacturing process compared to rival AMD’s RDNA 3 architecture.
It wouldn’t be a surprise if Nvidia chose to do so. At GTC 2022, NVIDIA released a new generation of H100 based on the Hopper architecture for the next generation of accelerated computing platforms. This GPU designed for supercomputers has 80 billion transistors and is manufactured on TSMC’s 4nm process using CoWoS 2.5D wafer-level packaging. The 4nm process used by the H100 is a version tailored by TSMC for Nvidia, which is different from the so-called N4 process.
Rumored AMD RDNA 3 architecture Navi 3x series GPU, Navi 31, and Navi 32 will use MCM multi-chip package, GCD (graphics computing chip) and MCD (multi-cache I/O chip) will use two different process technologies, namely TSMC’s 5nm and 6nm processes.
TSMC’s 4nm process and 5nm process belong to the same process node, but they are optimized on the basis of the 5nm process, and they belong to the same category that provides more PPA (power, performance, area) advantages. The same design rules, design infrastructure, SPICE simulation program, and IP are maintained, and mask count, process steps, risk, and cost are reduced.
This means that AMD and NVIDIA are different in the design direction of the new generation of GPUs. AMD adopts a design method that is considered to be better, and Nvidia may adopt a better process. As for which is more efficient at this stage and the actual operation, the effect is better, we will not know until the two parties release new graphics cards.
