Nvidia CEO Jensen Huang shocked the PC community with his statement that “graphics cards will never get cheaper again”. But is that true? What is behind his statement – and what speaks against it.
Nvidia has introduced the first GPUs of the RTX 4000 generation and although the features presented and the performance capacities of the new cards are quite impressive, another topic has determined the conversation: the significant price increases. For the top model RTX 4090, Nvidia is asking a proud 1,949 euros, 450 euros more than for the RTX 3090 two years ago. But those who still hope that high inflation and fluctuating exchange rates are the only reason for the increased prices may be disappointed. Because Jensen Huang, CEO of Nvidia, says very clearly: The prices for graphics chips will no longer fall.
“Moore’s Law is dead – graphics cards remain expensive”
This statement comes from a question and answers session with journalists last week. When asked about the increased prices and the strong reactions online, Huang made it clear that consumers should adjust their expectations. The main reason for this is the sharp rise in production prices.
“First of all, a 12-inch wafer is much more expensive today than it used to be. Not a little bit more expensive, but a whole lot more expensive,” Huang said.
In addition, one has to say goodbye to the old truism that the performance of computer chips makes significant leaps every few years.
“Moore’s Law is dead. And the notion that doing twice the work for the same cost, or doing the same work at half the cost every year and a half, is gone. It’s completely gone,”
PC gamers should also not only compare the cards based on their model numbers but instead compare the performance with the retail prices. Because if you compare the performance of the current cards with cards from previous generations at the same price point, it has increased enormously.
However, it is not so easy to compare the performance of the past with the performance of today. Since technological limits are slowly but surely being reached, manufacturers are no longer just trying to increase performance in the traditional way – more processing units and higher clock rates. Instead, Nvidia’s Deep Learning Super Sampling (DLSS) takes care of that. Reduce the computing power required by rendering the image at a lower resolution. Using artificial intelligence and a neural network, the so-called tensor cores recalculate the image to the native monitor resolution.
With Ada Lovelace, Nvidia is breaking new ground and introducing DLSS frame generation. Whole frames are no longer calculated on the shader units but artificially generated by AI. Not only the fourth generation Tensor Cores are responsible for this, but also the Optical Flow Accelerators of RTX 4000. Of course, this also increases the complexity of building the GPU.
How expensive does a graphics card have to be?
As an outsider, it is difficult to classify Huang’s statements and check their accuracy. It is definitely true that many external factors have made microchips more expensive to manufacture. The corona pandemic is still having an impact on global trade, commodity prices have risen significantly and the current inflation is evident in all areas of life.
It has also been proven that Moore’s law does not apply indefinitely. Boosting semiconductor performance depends in large part on shrinking feature size, and the RTX-4000 cards are manufactured in TSMC’s N4 node (although some experts say it’s actually a modified 5nm process). However, the structure size of semiconductors cannot be infinitely reduced, because from a certain point (approximately 2nm) classical physics no longer applies – from this point quantum mechanics takes effect.
So there are some arguments that support Huang’s statements. However, it is difficult to judge from the outside how strongly these points actually affect the prices of graphics cards. It is well known that Nvidia is struggling with severe sales slumps – also as a result of the significantly weakening crypto market. The fact that the second-largest cryptocurrency, Ethereum, has switched the mining process from GPU-based proof-of-work to proof-of-stake, which is significantly less computationally intensive, certainly doesn’t help Nvidia.
The presentation of AMD’s RX-7000 series on November 3rd can be an indication of how the graphics card market will develop over the next few years. AMD has the opportunity to send a declaration of war against Nvidia with more attractive prices and poach the customers of the competition. Ultimately, functioning competition is the best means of generating fair prices for consumers and preventing usury. And with Intel, a newcomer is coming onto the market whose products may not yet be able to keep up with the performance of AMD and Nvidia, which is precisely why they will be pushing ahead with attractive prices. The greatest power still comes from consumers, because if the RTX 4000 cards become a slow seller, Huang might change his mind about how expensive a graphics card needs to be.