Review: Nvidia’s $999 GeForce RTX 5080 falls disappointingly short of the 4090

Nvidia’s GeForce RTX 4080 graphics card was faster than the RTX 3080 card it replaced. But it was also faster than the RTX 3080 Ti, 3090, and 3090 Ti. One of the good things about a new graphics card generation is that the new cards bring the last generation’s inaccessibly expensive high-end performance down to cards that more people can actually afford.

That’s not the case with the new $999 RTX 5080, which beats the previous-generation RTX 4080 Super by a little bit and the older RTX 4080 by a slightly larger bit but doesn’t come close to beating or even replicating the performance of the outgoing 4090.

Nvidia points to its new DLSS Multi-Frame Generation technology as a mitigating factor here, leaning on its AI-generated frames to close the gap that the 5080’s raw rendering performance can’t close on its own. And sure, it’s nice that this card can do that. On paper, the 5080 is also technically a good value compared to the flagship RTX 5090—between 60 and 75 percent of the performance for half the price (though talking about the MSRP of any of these cards at launch is strictly theoretical, given allegedly short supply and the demand from both actual buyers and scalpers looking to make a buck).

But the 5080 really feels a lot more like a 4080 Super Super—meaningfully better than the 4080 but still in the same general performance category. It’s an upgrade, especially for anyone coming from a 30-series or older GPU, but it’s a disappointing break from Nvidia’s past precedent.

What you need to know about the RTX 5080

Of Nvidia’s mid-generation Super refresh for the 40-series last year, the 4080 Super was already the mildest improvement over the original card, with just a few hundred extra CUDA cores and very mild clock speed increases. Its biggest and most important improvement was that it brought the old 4080’s original $1,299 price tag down to a somewhat less offensive (and, again, strictly theoretical) price of $999.

The strategy with the 5080 looks pretty similar. You get a mild increase in core count (up 10.5 percent over the original 4080 and only 5 percent over the 4080 Super), plus the same 16GB of RAM on the same 256-bit memory interface. Switching from GDDR6X to GDDR7 does get you a decent bump to memory bandwidth, though, on the order of just over 30 percent compared to the 4080 and 4080 Super.

As with the 5090, the 12-pin power connector is now slightly recessed and angled so it doesn’t stick out as far in your case.

Credit: Andrew Cunningham

But all of that is incredibly mild and incremental compared to the big across-the-board jumps in core count, memory bus width, memory size, and memory bandwidth that you get when you go from the RTX 4090 to the RTX 5090. The 5080’s core counts and memory bandwidth also stay well below the RTX 4090’s. This has the benefit of not blowing up the 5080’s power requirements—at 360 W, it’s only 40 W higher than the 4080, and in our actual testing, the 5080 didn’t actually consume that much more power on average—but it also means that the 4090 and 5090 continue to stand apart from the rest of the lineup, inaccessible to anyone with less than a couple thousand dollars to spend on a GPU.

As for the physical design of Nvidia’s Founders Edition version of the 5080, it’s identical to that of the 5090, just as the 4080 looked like the 4090. The 5080 does feel lighter than the 5090, which makes some sense—a card with much lower power consumption and half as many CUDA cores doesn’t need the same kind of cooling apparatus. But the dimensions and connectors on both cards are exactly the same.

GPU temperatures measured during benchmark runs.

Credit: Andrew Cunningham

In our testing, the 5080 ran a little warmer than the 4080 did under load, but only by about 5° Celsius. The 5080 Founders Edition runs 11° or 12° cooler than the 5090 Founders Edition, so it should be a bit easier to keep your system cool with one of these installed.

Testbed notes

Our gaming testbed is identical to the one we used for our RTX 5090 review. Compared to reviews from 2023 and 2024, we’re using a faster CPU (the Ryzen 7 9800X3D) and a beefed-up power supply. We’ve also added some newer games to our test suite and discarded a few that weren’t challenging the cards enough.

As with the 5090, we focused on 4K tests for this review. The 5080 is half the price of the 5090, but you could still buy a good 1080p gaming PC or a pair of game consoles for the same money—it ought to be capable of playing modern games smoothly at high settings and resolutions. (If you are curious about 1440p performance, look at the benchmarks we’ve run with DLSS enabled—we run these in “Quality” mode, which at 4K means a 1440p internal resolution.)

Performance and power

In our 5090 review, we observed that its performance improvement over the 4090 generally scaled linearly with its CUDA core count and its power consumption—all three usually increased by between 30 and 35 percent, depending on the game.

The 5080 did a bit better than this, with average frame rates that were generally between 10 and 20 percent faster than the regular 4080 and between 5 and 13 percent faster than the 4080 Super. A combination of cores, marginally higher clock speeds, and memory bandwidth increases help give it a boost.

But especially compared to the 4080 Super, these numbers look less like a generational performance increase and more like what you could get out of a 4080 Super with a particularly good overclock. We point this out not because people are upgrading their GPUs every generation but because people with an older 30-series, 20-series, or even 10-series GPU who have been waiting to upgrade won’t really be rewarded for waiting for this generation.

DLSS Multi-Frame Generation is meant to bolster the card’s performance numbers, and as with the 5090, they do look impressive on paper. As we covered in that review, Multi-Frame Generation (MFG) is exclusive to the 50-series, and it’s an amped-up version of the Frame Generation technology that was introduced in the 40 series. Both versions take two rendered frames and use AI to interpolate extra frames in between, sort of like a supercharged version of TV motion smoothing. The 40-series cards can generate one frame this way; the 50-series cards can generate up to three.

Multi-frame generation looks great on paper, but it’s misleading to compare MFG frame rates to natively rendered frame rates.

Credit: Andrew Cunningham

Hardware Unboxed on YouTube has a good in-depth look at MFG and its trade-offs, which are mostly the same as the trade-offs with regular FG. It’s useful when your rendered frame rate is already fairly high, giving the algorithm plenty of data to use when generating frames and making individual errors or artifacts harder for your eyes to spot. But it’s not a cure-all for low frame rates. It can also increase latency because user input is only being polled at your actual rendered frame rate, and running the Frame Generation algorithm actually reduces that frame rate by a bit.

The RTX 5080 is fast enough that it can hit the frame rates needed to make MFG look good, but it will still be game-dependent and scene-dependent. It’s a good way to augment rendering performance, but contrary to Nvidia’s messaging, it’s still not a substitute for better rendering performance.

Technically an upgrade

The RTX 5080 isn’t bad; it just doesn’t move things forward a whole lot.

Credit: Andrew Cunningham

This is not really the generational leap that GPU buyers were probably hoping for from an RTX 5080. It is the third-fastest GPU on the market, but it’s far short of the 4090, and it performs more like a second refresh of 2022’s RTX 4080.