The discourse of the latest Nvidia GPU GeForce RTX 4000 series has been circulating, and reportedly manufacturers will rely on the process node 5nm TSMC than the Samsung 8nm used for GPU Turing. However, this is still in the form of rumors, so it's still far from a definite possibility.
But if true, the expected GPU the latest team green will bring much better performance, even the difference can be up to 50%.
Well, reportedly RTX 4000 series will await at the beginning of next summer. On the other hand, NVIDIA still has a lot of PR, including the present series refresh RTX 3000. Some of which will appear the possibility of the series is the entry-level RTX 3050 or high-end version RTX 3090 Ti.
According to Techspot, the next generation will be brought NVIDIA has the codename ada Lovelace, but not much is known about the new architecture and what will be brought green team. The only consensus among people in the industry seems to be that the performance of the raw will double compared to the current generation, while the power consumption will be greater in the range of 400W.
There are rumors that Nvidia will be using the node process 5 nm TSMC, is believed to be much more expensive than the node process 7 nm used AMD for the GPU the RX series 6000 him. A recent report from Digitimes also can strengthen this hope, citing industry sources from the supply chain.
This publication also noted NVIDIA will utilize packaging technology chip-on-wafer-on-substrate (CoWoS) TSMC to GPU H100 future. This Model will be based on the architecture of the Hopper-oriented computing Nvidia and is expected to be a module GPU multi-chip company first. On the other hand, the GPU ada Lovelace will most likely use packaging FC-BGA belongs to the Technology of ASE, as well as memory GDDR6X traditional.
GPU series RTX 3000 manufactured using the node process 8 nm Samsung relatively mature, so a transfer to node 5 nm TSMC said to be quite risky because it will also be used by other companies.
Example, Apple is currently making chipset A14, M1 Pro, and M1 Max on the same node, while AMD is planning to use it for CPU Zen 4 terbaru. At the same time, Bitmain has also securing a portion of the production capacity of 5 nm TSMC to hardware device mining that will be launched in Q1 2022.