We continually do not stop seeing new processors on the market, but they have a design process that is conditioned by the type of manufacturing node used and its rules. This leads large functions to compete for customers, who jump from manufacturer to generation. This has led Qualcomm and NVIDIA to sign with TSMC for the use of the 5nm node of this type.
The next big node for manufacturing different types of processors is what we call the "5nm generation" in which TSMC and Samsung under their 5nm nodes and Intel under their 7nm node are going to compete. But for the moment of the big foundries, the one that has the advantage in this technological race is TSMC.
NVIDIA and Qualcomm to use TSMC's 5nm node
Both companies stopped designing their processors, each one on its own land, for TSMC's 7nm node, Samsung being the foundry selected in the case of both for the newest products that both have on the market. For example, in the case of NVIDIA, we know that the entire GeForce Ampere, or known as RTX 3000 range is manufactured using Samsung's 8 nm node instead of TSMC's 7 nm.
But foundries have been in a war for a long time and it is TSMC that is taking the lead in what we call the 5 nm generation, and if a few days ago we received rumors that the next generation of NVIDIA GPUs under the code name "Lovelace" will use TSMC's 5nm node, we now know from Digitimes that NVIDIA together with Qualcomm would have signed the use of TSMC's 5nm node for their next generation of processors.
This node is already fully operational, but for the manufacture of low-power SoCs for Apple's PostPC devices and its new M1 for Macintosh, which is still a derivative of the A14X. So Qualcomm should soon deploy its Snapdragon SoCs for PostPC devices under TSMC's 5nm node version for low-power processors. Let's not forget that these are the first processors to be mass-produced for a new node today.
Many designers, few manufacturers
The fact that Qualcomm and NVIDIA leave Samsung to return to TSMC causes that there are too many participants using the 5 nm node of this foundry, since not only do we have Apple already using the node and the two already mentioned, but we also have to keep in mind that AMD will also make the jump to 5 nm and that Apple will not only use the version of the node for low consumption but will compete in resources with the version of the node for high-performance processors with its SoCs for its Macintosh.
Such a number of clients for the 5 nm node in combination with the fact that each new node is increasingly expensive to deploy, causes the cost of wafers to be higher and higher and with it the production of processors, which also affect design costs. So TSMC is going to take advantage of it to rent its production to the highest bidder, which will cause the initial prices of the processors under this node to be really high.
We do not know how the deployment will be, but everything indicates that the key year is 2022, both for the launch of Zen 4 and RDNA 3 by AMD, and that of the GeForce Lovelace by NVIDIA. Obviously, the first processors that we are going to see are the high-end ones, since this is where they can get higher margins for the launch. Although in the case of the Zen 4, because the chiplets with CPU cores are smaller, they have more numbers to be launched before the rest of the processors.
We really should see TSMC's 5nm node being used for NVIDIA processors and the rest of the foundry's clients between 2022 and 2023. Although you don't rule out changes in the agreements either, such as the fact that Intel may enter into licensing its node 7 nm to third parties.
0 Comments