This isn't really as impressive an improvement as it would initially appear. I'll explain why by using snippets from the article. First let's use the headline:
"Nvidia GeForce RTX 4090 is 78% faster than RTX 3090 Ti, hits 3.0 GHz in leaked benchmarks"
That 78% is pretty impressive-sounding, but there's something that I picked up on here that makes it actually disappointing. Here's why:
"The leak also shows the card reaching a 3,015 MHz frequency. For comparison, the RTX 3090 Founders Edition boosts to 1,695 MHz."
- Ok, so...
3015 ÷ 1695 = 1.77876.... (aka 78%)
Well, this RTX 4090 just looks like a wickedly overclocked 3090. Of course, to maintain a stable overclock, TDP must have an equally dramatic increase. Did that also happen here?
"Being a triple-slot card, the NVIDIA GeForce RTX 3090 Ti draws power from 1x 16-pin power connector, with power draw rated at 450 W maximum."
- RTX 3090 Ti - TechPowerUp GPU Database
Ok, so what's 78% more than 450W?
450W x 1.78 = 801W
Now, what was said about the TDP of the RTX 4090? Oh yes, that's right:
"According to the leaker, it has a default TDP of 450W, though it is designed for 600W to 800W."
I see a
maximum of 800W. Almost exactly 78% more than the previous generation and most likely the TDP that this leaked card was running with to get these performance numbers. The leaker said that it has a
default TDP of 450W but there's nothing said about what this card was running to get the numbers that were posted. A performance increase of 78% and a maximum TDP increase of also 78% is just too perfect to ignore.
Then of course, to do this, you'd have to increase your cooling equally dramatically and nVidia is claiming that the RTX 4090 will
not fit in a mid-tower case. This sounds like a card that will just barely fit into my gigantic Ultra U12-40670 SuperTower (with regard to its length). Also who knows how many slots it will take up? Maybe four, five or even six? With multi-GPU setups all but dead there is a lot of residual room for a card that big. Wouldn't it be funny if it had 78% more cooling surface area than the 3090?
It just sounds like nVidia took Ampere, overclocked the hell out of it and increased the TDP and cooling to compensate. There doesn't appear to be much of a technological improvement with regard to efficiency. The performance-per-watt appears to be exactly the same. That's not incredibly impressive to me.
DISCLAIMER:
Firstly, this analysis is based on the assumption that the leak is accurate which, as we know, is by no means guaranteed. Secondly, I realise that I've bounced back and forth between the RTX 3090 Founder's Edition and the RTX 3090 Ti which are not
exactly the same card. However, their performance delta is insignificant so they are more or less the same card.
According to TechPowerUp, there is only a paltry 7% difference between the RTX 3090 and RTX 3090 Ti. I believe that Steve Walton considers a 3% difference to be statistically a margin-of-error tie so there's just a 4% real difference between them.