RTX 5090 vs RTX 4090 tested at same power level: RTX 50 flagship shows efficiency downgrade vs previous gen

[Redaktion]

Nvidia's latest flagship GPU, the RTX 5090, has a 28% higher TDP than its predecessor, the RTX 4090. This 28% increase in TDP, and a healthy increase in CUDA cores, makes the RTX 5090 around 27% faster vs the RTX 4090. However, the gain is much less when both cards use the same amount of power.

https://www.notebookcheck.net/RTX-5090-vs-RTX-4090-tested-at-same-power-level-RTX-50-flagship-shows-efficiency-downgrade-vs-previous-gen.952084.0.html

[Yeshy]

And it seems like it applies down the stack, based on Nvidia's own numbers / compared to 40 Super


I wonder if laptops will basically just be a refresh with MFG. Maybe with better undervolting / v/f curve they can be better?

[Nandiman]

Anyone who is surprised by this new development, please raise your hand. What? No takers?

[Jorb]

Interesting read! Thanks!

And just for your interest as a professional writer, you're using the phrase "begs the question" incorrectly. You should have said "Raises the question".

Begging the question is a specific logical fallacy where an argument assumes the truth of its conclusion without evidence.

[Reg Watson]

Yep - that's right. This is how we know people aren't native born speakers.
Like in "Inglorious Basterds" when the allied soldier incorrectly have the wrong hand gesture.

[Not_Brian]

Yep - that's right. This is how we know people aren't native born speakers.
Like in "Inglorious Basterds" when the allied soldier incorrectly have the wrong hand gesture.

Plenty of native speakers misuse "begging the question".  In fact most of them do, in my experience.  There aren't a lot of logicians or linguists in the general population.

Back to the topic at hand, the problem I have with the article is that it says in the opening that the 5090 outperforms the 4090 by 27% while using 28% more power, then states that the gain is "much less" when their power consumption is equal.  However, it then goes on to say that, when you do that, the 5090 outperforms the 4090 by 17% when it uses 0% more power.  So the 5090's gain when using the same power is in fact much more (and is actually a gain, relative to power consumption).  Unless I'm terribly confused.

[Aran]

Yep - that's right. This is how we know people aren't native born speakers.
Like in "Inglorious Basterds" when the allied soldier incorrectly have the wrong hand gesture.

Plenty of native speakers misuse "begging the question".  In fact most of them do, in my experience.  There aren't a lot of logicians or linguists in the general population.

Back to the topic at hand, the problem I have with the article is that it says in the opening that the 5090 outperforms the 4090 by 27% while using 28% more power, then states that the gain is "much less" when their power consumption is equal.  However, it then goes on to say that, when you do that, the 5090 outperforms the 4090 by 17% when it uses 0% more power.  So the 5090's gain when using the same power is in fact much more (and is actually a gain, relative to power consumption).  Unless I'm terribly confused.

One issue with this review is that limiting the power consumption to 450 or 400 watts for the RTX 4090 is not really a "limit." The RTX 4090 DOES NOT exceed 400 watts in most usage scenarios unless you are overclocking or under specific loads. This means that with the same 450W power cap, the RTX 5090 and RTX 4090 could very likely be operating at different power levels.

[Janon]

Its still on the same process node, so expecting efficiency gains is silly.

[Bougie buddy]

Well of course. 5090 needs extra power because it has extra cores made on the same node. If you don't power the extra cores then you don't use the extra cores.

[Blecked]

Weird the thermals weren't tested at the same time? This could've shown the new efficiency of the thermal designs.