After all the rumours, the wild speculation, and the little teases derived from leaks, NVIDIA’s GeForce RTX 5090 Founders Edition (FE) is finally here in the lab. Like all new generations of NVIDIA GPUs, the Blackwell GPU running under the hood of this GPU is top-of-the-line, targeted at the most powerful, overkill, and certainly, the priciest high-end desktop PCs on the market.
As powerful as the card, though, there are some caveats, both expected and surprising, that are mired with the prowess that comes with it.
Specifications
New Look, New Software, New Features
The RTX 5090 FE feels a lot like a deja vu scene; you know NVIDIA once used the dual front-fan design before and after two generations of the Dual Axis flowthrough, the GPU brand has finally come full circle, with some tweaks and improvements to the overall cooling solutions, naturally.
First, it’s now called Double Flow Through. Like the Dual Axis Flowthrough, the cooler shroud of the RTX 5090 FE is practically all fins, but this time around, the card exhausts the air through the card and to the top, rather than having one fan exhausting air out through the back of the card.
This change in airflow is only possible because of the Blackwell GPU and more importantly, its diminutive size. Well, diminutive compared to Ada Lovelace. It is a square piece, literally, centred on the card. It is because of this that NVIDIA is able to apply the new vapour-chamber pipes and cooling shroud that, again, allows the RTX 5090 FE to push the air through on both ends.
Another (welcome) byproduct of the new GPU die size? The RTX 5090 FE is no longer a chonky boy. It’s a 2-slot card again, which in turn makes it a card that would fit in even small form factor (SFF) desktop PCs. What a shame then, Malaysia isn’t an official recipient of Founders Edition cards. Seriously though, putting it side-by-side with the RTX 4090 gives a sense of the length NVIDIA went to shave off all that weight.
Glancing through the specs, we’re looking at 32GB of GDDR7 memory running at a screaming 28 Gbps, with a 512-bit memory bus. Additionally, the RTX 5090 FE has a boost clock of 2.41GHz, which is slightly slower than the RTX 4090.
Then there’s the software that’s been baked into Blackwell as well. There’s new generation RT and Tensor cores, built from the ground up to enable the RTX 5090 FE to take full advantage of Neural Shaders. Of course, and as predicted by earlier rumours building up to its announcement, DLSS4 and Multiframe Generation is a real thing.
Regarding DLSS4, there are one or several issues, depending on how you look at it, and it’s virtually identical to the problem that presented itself in the early stages of DLSS3 when it first arrived on the scene. There’s currently only a handful of titles that will be able to run DLSS4. I say “will” and not can, because the features aren’t currently available for said titles, at least at the publication of this review. I’ll save the detailed explanation of DLSS4 and the accompanying technologies for a separate article.
Having learned its lesson from the RTX 40 Series, the card’s new 12VHPWR connector has been repositioned and reseated into a diagonal angle, identical to the design that NVIDIA adopted when it launched the RTX 3090 and 3080 Founders Edition cards years ago. I speak for myself in this case but the new connector’s angle is a much appreciated adjustment and a clear sign that NVIDIA listened to the unwilling victims of dislodged and burned out PCIe Gen5 connectors.
Testbench
The card I am comparing the RTX 5090 FE to will be the RTX 4090 FE, a card that, for the past two years, has been the de facto GPU for hardware testing. To point out the other obvious point, it is also the control card that the new Blackwell GPU needs to beat. As you’ll see in the next part, the card does prove its mettle but not without some caveats.
Benchmarks, Temperature, And Power Consumption
NVIDIA promised us that the RTX 5090 would have double the performance of the RTX 4090 and to a measurable degree, the brand kept its promise. By that, I mean that it does in some parts and usually in what can be described as “controlled environments”.
Synthetically, the simple truth of the matter is that the RTX 5090 slaps. Against the RTX 4090, it’s not a stretch to say that it whips its predecessor into submission with a very comfortable lead. I mean, with the Fire Strike test, it’s more or less a hair’s breadth ahead of the Ada Lovelace flagship but hey, a lead’s still a lead.
On the subject of gaming, the performance of the RTX 5090 is…interesting, to say the least. I don’t know if it is the GPU or the press drivers provided by NVIDIA, but for some titles, there is a framerate barrier that the card just won’t break through.
In my case, those titles are Deus Ex Mankind Divided (DXMD) and Shadow of the Tomb Raider. Regardless of the resolution I run the RTX 5090 on, it seemingly caps itself at a certain level. For DXMD, that cap is set at 163 fps, while for Shadow of the Tomb Raider, that cap sets in at a little over 200 fps.
For another matter, some titles don’t seem to want to play nice with the RTX 5090. Control is one such title: whether I’m running it on a 1:1 render or half render, the game suffers a catastrophic crash after just a minute or two of gameplay. Worst, the DLSS option is not available for the Blackwell GPU. Remember how I said that some titles tend to work in favour of the GPU in “controlled” environments? Cyberpunk 2077 is one such game but only if and when the new Transformer model is utilised. Available only via DLSS 4, maxing out the setting allows the game to scream past the 250 fps barrier at 4K in Balanced Mode. Switch it to Performance mode, and the average framerates jump again, this time beyond the 300 fps marker.
But with regards to its overall performance, even without the aid of DLSS 4, the card still manages to outperform its predecessor by several frames, both across the board and the resolutions.
Moving on, there is another physical cost to the RTX 5090 maintaining its composure and like all GPUs, that price is power, literally. Off the wall, the card pulls an average 565W of power when running on a full load, which is really close to its advertised 575W TDP. If I’m being honest, it all feels a little excessive, even if it does deliver on performance.
I suppose the other grand thing about the card is that while the thermals certainly seem to run hotter than the RTX 4090 in a constant 20°C room, it’s still not as terrible as I had imagined. On a full load, the RTX 5090 peaks at 75°C and doesn’t appear to budge but again, I would imagine it would go higher if the ambient temperature was also higher. Oh, and the magic of NVIDIA’s Double Flow Through cooling solution is just as impressive: all that heat is literally at the centre of the card, with the fins and shroud’s sides being cool to the touch.
That said, this is a top-tier card that, in Jensen’s words, is made for a high-end system, meaning that whoever it is that purchases this card is likely going to have deep pockets and won’t be worried about haemorrhaging their wallets.
Conclusion
There is no easy for me to say this but as powerful as the NVIDIA GeForce RTX 5090 FE is, unless you have deep, deep pockets, the US$2,000 price tags puts it out of the hand of the gaming masses. And because we don’t get Founders Edition cards, it’s a guarantee that any and all AIB partner variations of the GPU coming into Malaysia will retail north of RM10,000 (looking at you, ASUS).
As I mentioned earlier, this is a high-octane machine, made of equally, ridiculously power-hungry rigs, fitted with a top-of-the-line CPU and quite possibly, enough RGB to make you wonder if you actually love sleeping. Yes, a lot of its capabilities is derived from the software and baked-in hardware-driven features like AI upscaling and ray-tracing, but that’s just been the name of the game for the past seven years at this point. Ultimately, if you’re looking to get invested in the RTX 50 Series without haemorrhaging your wallet, I would recommend that you wait for the other cards; the RTX 5080 will retail at half the price, and just in case you’ve forgotten, the RTX 5070 offers RTX 4090-level performance for 1/5th the price of the GPU when it first launched two years ago.
Photography by John Law.
Follow us on Instagram, Facebook, Twitter or Telegram for more updates and breaking news.
