The continuous evolution of real-time graphics and the entry of graphics cards into other markets has made a continuous evolution necessary not only in architecture, but also in power. The biggest bottleneck you’ve faced other than video RAM? The energy consumption they could use. However, the standard for power connectors for PCI Express has been stable for a while. Well, we are going to explain the reason why we believe that NVIDIA could not jump from RTX 40 to RTX 50, but rather have an intermediate generation prepared, much better optimized than the current one.
The architecture of the RTX 40 can be further optimized
From a power and raw performance perspective no, as more power means more clock speeds and faster memories. On the other hand, if we look at it from an economic point of view, the fact of having to buy a new power supply and seeing a rise in the electricity bill due to changing the graphics card, well, it’s not funny at all. Especially if we have a pretty tight economic situation.
Now, after weeks of its launch and without detracting from the RTX 4090, which deserves it for its high performance, the general conclusion is that the use of the new connector and its high power level has been more a demand from above than not part of the chip design. Why do we think this? Well, because of the information that can be obtained from performance losses after using the power limiter or performing voltage drop exercises.
The graph that you see above these lines tells us about the performance of the RTX 4090 limiting the maximum consumption and as you can see, the performance cut with a power that does not require the use of the new connector is not proportional to the drop in consumption. In any case, this is normal since the power consumed is a quadratic function in any chip. You already know things about mathematics.
The evidence points to the RTX 40 SUPER
The fact is that if we Undervolt the graphics card and setting the voltage to 0.950 mV we obtain between 95% and 99% of the performance, but consuming close to 350 W, something that can be achieved with a double 8-pin connector, then the 12VHPWR justifies itself even less. In any case, the consumption curve of the AD102 chip does not seem to correspond to that of a GPU designed for the high consumption of the RTX 4090 and we are based on a detail that no one is commenting on.
Technically, the RTX 3090 Ti has the same power consumption as the RTX 4090, a chip of the same size and the same memory configuration. The issue is that its heatsink and fans are lower in height. Let’s not forget that inside the chip there are 128 cores, out of 144, in the new generation chip, compared to 84. What do we think? NVIDIA engineers decided to use the 12VHPWR connector at the last minute to achieve more MHz, which translates into a hotter chip and for which they have had to build a more complex and expensive heatsink.
That is why we believe that we will see an RTX 40 SUPER series, consisting of chips with the same architecture and node, but with a more optimized consumption curve that will allow much greater power to be drawn with the consumption delivered by the new connector and that will allow them to continue. holding the scepter of the most powerful graphics card.
Bad timing for NVIDIA?
It cannot be guaranteed that you will never use or bet on a technology, people believe that the 12VHPWR connector is an imposition of the Jen Hsen Huang brand when it really is not. It is a PCI-SIG standard that will be increasingly adopted by power supply manufacturers and we will see it in more computers. So in the end AMD will also fall into the use of said connector.
However, in the same way that it is not the same to design an architecture taking mobile consumption into account, it is possible that with what has happened with the 12VHPWR connector regarding performance and consumption, we have accidentally discovered that such levels of power require rethinking the way of making a GPU in order to get a higher performance per watt using the new 12+4 pin connector.
So we should expect, not a new generation with a more advanced architecture in the short term, but versions of the RTX 40 with a more optimized consumption curve and reaching higher clock speeds using the same base as the current ones. Which will be part of a new family. Obviously we shouldn’t expect them in a few months.