You’re neither the first nor the last to think that prices need to come down when it comes to graphics cards. The market is crazy about speculation, there are less and less excuses for those who justify the costs put by the manufacturers, and if you think that the new batch with Ada Lovelace and RDNA 3 will solve the problem… You are wrong may be. In theory, prices are going to be higher, and the reason is simple and complex at the same time: design costs. What will be the RTX 40 and RX 7000 prices?
This morning, we were talking about how NVIDIA intends to cut costs against AMD with a move that Lisa Su has already made: increase the size of the caches and decrease the buses or simply keep them, but never increase them. Although this reason is valid due to obvious architectural details, there would be a second reason in the shadows and of which little is said.
The price of the RTX 40 and RX 7000 compared to design cost
We go first with the final conclusion and then with the arguments so that it is better understood. In principle and except surprise PS5 way where Nvidia and AMD design at a loss, the price of the RTX 40 and RX 7000 It will not be lower than that of the current RTX 30 and RX 6000.
The reason for this is the design cost of TSMC’s 5nm node, which on the other hand will also drive Intel and Samsung in their respective versions down to smaller nanometers. It is believed that the pitch of the transistors FinFET at SHGs (and its versions customized by each of the three designers) it’s for reasons of density and space, but if it’s true it’s also true for the costs.
The comparison table shows a weighting, a more or less precise overview of what the different industry nodes cost or have cost. Keep in mind that this is top secret and you can’t be completely accurate, but IBS calculates this based on their research.
5nm almost doubles its cost compared to 7nm
This is a trend that will continue in the market until the arrival of GAA and High-NA for EUV, as well as better AI algorithms for design. What we can see is that from the current 7nm we’re at with TSMC and Intel (10 in the case of the blues, although they’re technically better) the cost is almost $300 million by design, whereas with the new 5nm this goes up to the 542.2 million.
That is, for every chip made by NVIDIA or AMD, the cost will be similar to these numbers, so it’s no surprise that they use the same skins for as many dies as possible, because then disabling CU or SM is much more simple. cheap.
But if that weren’t enough, it’s estimated that for TSMC’s 3nm and Intel’s 7nm, the costs are somewhere between 500 million and 1,500 million dollarsalways talking about the most advanced chips of the two brands of course.
For this reason, manufacturers today take much longer to launch new products or simply to launch so-called “refried” ones: they must first optimize the design costs and to this must be added the production, distribution and other small additions. Therefore, it is not expected that the prices of new graphics cards will fall, quite the contrary, it would be a miracle if they remained at figures similar to the current ones.
The mining has shown that there are people willing to pay outrageously for a GPU, we will see if when the mining as such is finished and ceases to be profitable if the average gamer goes through the hoops of AMD, NVIDIA and Intel or if on the contrary it switches to consoles and leaves behind the PC market.