NVIDIA will most likely be producing RTX 30 cards along with RTX 40 cards, some RTX 20 models might survive into the Ada Lovelace era as well
Here is the thing though; it is even possible that NVIDIA will continue to produce some high-volume RTX 20 series cards in the Lovelace era as well and all of this makes sense. The GPU market has shifted significantly in the last few years and where, once, old generation GPUs quickly lost their value, they now retain it for years to come (primarily because of crypto). The ongoing semiconductor crunch also means that GPUs are getting more expensive as upstream wafer manufacturers and foundries increase quotes (and costs) for the GPU makers. Before we go any further, here is the quote in question: Now here is the thing, keeping multiple production lines across generations isn’t something new. We have seen it before and it should be very familiar to some of our older readers. Intel used to do this across its three product stacks where the mainstream desktop, HEDT desktop and mobility side of things would be on completely different architectures (N-1) to not only optimize the economies of scale but to spread out R&D/retooling costs as well. The same holds true for NVIDIA as well and with their parallel production of RTX 2060 GPUs, they are able to optimize volume for a major market segment. But there might be more to the story as far as this particular hint goes and that is where the second angle comes in. -Colette Kress, NVIDIA CFO
NVIDIA could be getting ready to tap both Samsung and TSMC for high-end GPU production, significantly increasing its GPU supply
Intel and NVIDIA’s historic use of parallel GPU production has always been restricted to a staggered approach where the market segments were spread out across different architectures. This was done simply to optimize costs. All high-volume GPUs essentially came from the same wafer allocation. With NVIDIA getting ready to move to TSMC for their Ada Lovelace series, this approach presents another major advantage. They could continue to produce their high-end RTX 3080 Ti / RTX 3070 Ti GPUs etc at Samsung and at the same time produce RTX 4080 Ti and RTX 4070 Ti GPUs at TSMC. Since these are two separate foundries, producing two product lines will not eat into the same allocation at all! In fact, considering the market exists in a serious under-supply situation, there is no harm in going for this approach. The ridiculous advantage this tactic would give NVIDIA, both in terms of revenue and market share, means that I have a feeling we will see a lot more than just continued RTX 3060 production at Samsung. Keep in mind there are some caveats associated with this bold educated speculation. Firstly, upstream wafer manufacturers are running at a 100% capacity and the semi-industry exists in a zero-sum game - which might limit, to some extent, the doubled allocation theory. That said, the zero-sum does not apply at the micro level when talking about specific products. For example, I doubt any of my readers care whether Samsung fulfills NVIDIA’s allocation by taking away from some smaller players. What is incredibly likely is that Samsung will do everything it can to retain NVIDIA’s account and the company will very likely walk away with doubled allocation sizes. If TSMC, which is choked full with orders, can make space for someone as big as NVIDIA; Samsung certainly can.
