GeForce RTX 5080 FE review — Excellent performance at a reasonable price

You probably saw our analysis and review of the new NVIDIA flagship, the GeForce RTX 5090. In that review, we pointed out that the device delivers a whopping 30% increase in raw power, but when combined with DLSS 4, you can wring upwards of 300% to 400% out of this card. Games that have unstable or even unplayable framerates in the 20s can suddenly be running at over 250 frames per second. While the 5090 dominates the top of the chart, many folks are more comfortable with the price point of the RTX 5080. That’s precisely what we’re taking a look at today. Let’s go.

We’re going to kick off this review with a look at the RTX 5080 compared to the 5090, 5070 Ti, 5070, and the 4080 Super. Heading under the hood, let’s take a look at the chips and count them up. Before we do that, let’s just preface this very, very clearly – you cannot directly compare across generations with just the amount of cores or even speeds. The generational differences between cards (e.g. third-gen RT Cores and fourth-gen Tensor Cores in the 4000 series, vs. fourth-gen RT cores and fifth-gen Tensor Cores in the 5000 series) is always less about the amount of cores, but more about what’s being done with them. Let’s take a look:

As you can see, the 4080 Super is the closest numerical equivalent to the RTX 5080, in terms of fill rate, memory, and more, but that doesn’t tell the whole story. The shift to GDDR7, the next generation of all of these chips, and the various capabilities of the card beyond just pure graphical power are certainly a factor.

Just like the 5090, I have to tip my hat to NVIDIA for the upgrade to the card’s packaging. The 4000 series cards came in a large square box that opened into two triangles, with the card, power cable, and many booklets tucked behind the card. It was as protective as it was huge. The GeForce RTX 5090 and now the 5080 both ship in a smaller but just as effective package. The box is made of stiffened paper fibers, is now 100% plastic free, ink free (the name is die-cut into the surface), meaning that it is, by all indicators, completely recyclable. Also inside the box is a 4x PCIe 8-pin to 1x 16-pin connector, though most likely if you’re purchasing this card, you probably have a PSU with a dedicated 600W 16-pin connector already.

From the unboxing video for the 5090, you can clearly see that the NVIDIA flagship card has been put on a diet. The RTX 5080 is exactly the same size. Whereas the RTX 4090 FE was 304mm length, 137mm height, 61 mm (3-slot), the RTX 5090 and 5080 are both 304mm length, 137mm height and 40mm (2-slot). That makes this card able to be placed inside a small form factor case, if you’re so inclined. What a difference a generation can make, given that this new card has an entire new cooling system, the next generation of all of the internal hardware, and more of it. We all expected a larger card once again, so it’s great to be surprised. In addition, rather than the 1000W recommendation for the 5090, the 5080 drops that to 750.

As I mentioned, the cooling system on the RTX 5000 series has gotten a makeover. The 4000 FE series cards had one fan on top and one fan on bottom, located on opposite corners. The 5090 and 5080 both change to a double flow-through design, with both fans on the bottom – we’ll see if that holds for other Founder Edition cards in the future. It pulls the hot air through the card across the entire chassis for more efficient cooling. Just like the 4000 series cards, these fans go completely idle when the card is not being used heavily, saving power and noise.

The PCB on the card is now split into individual components, with five pipes on each side creating a vapor chamber that extends the full length of the card to maximize the cooling surface. I’m not quite sure what NVIDIA calls each section, but you effectively have heat pipes that extend the full length of the card, with a socket in the center. The PCB that contains all of the processing components are seated inside of this, with a set of fins and the center section to fully enclose those chips, creating a sort of thermal sandwich. Liquid Metal Thermal Interface Material is applied to the chip container to assist in thermal conductivity, resulting in a 4X reduction in temperature.

Around the pipes, NVIDIA has wrapped the large curved fins of the thermal block. These curved fins being wrapped around the pipe help to regulate the temperature and pressure changes that occur in the thermal exchange. Two slits that run the length of the center section of the card provide East-West air venting from the sides. Two 120mm fans are attached to the bottom of the card, pulling air through the pipes and fins then pushing it out the top to be grabbed by the pressure differential air flow in a properly-configured case. This unified 3D vapor chamber allows the heat to be pulled away from components and spread across the fins more uniformly, which should increase cooling efficiency dramatically. The 4000 series was elegant, but this is a whole new ball game.

The RTX 5090 flagship card is well-equipped, but it’s also $1999. The RTX 5080 comes in at half that cost, and as such there are some concessions. The card uses the the Blackwell architecture, meaning new memory and new generations of every chip on the board. While we jump from 24 GB of GDDR7 to 32 GB for the flagship, the 5080 has 16GB. It also drops the bus size from a staggering 512 bit lane to 256, which is in line with the 4080 and 4080 Super. The base clock for the card is 2295 MHz, with a boost clock of 2617 MHz, with the memory at 1875 MHz for a combined effective throughput of 30 Gbps.

One of the nice surprises for the GeForce RTX 5080 is a new rendering format. AV1 has enabled blisteringly fast render speeds in DaVinci Resolve, and now we have AV1 Ultra High Quality. This new format cleans up visual noise on large open spaces like skies and walls. It also adds another notch on the speed belt, though we’ll need to test further to quantify that. While video games are often high speed, there’s plenty of open sky and serene blue water, and this new format will be nice to showcase those cleanly.

Before we get into the numbers, I want to talk about DLSS 4 and “Fake frames”. If you are unfamiliar, GeForce cards since the 2000 series have a technology called DLSS – deep learning super sampling. Newer cards also picked up additional technologies like Deep Learning Anti-Aliasing, Ray Reconstruction, and Super Resolution. We’ve deep dived all of them in the past, but let’s focus on DLSS 4.

DLSS 4 utilizes the advanced power of the combined Blackwell architecture to interpolate up to three additional AI-generated (no, not the thieving kind of AI-generated) frames into the stream, driving framerates into the stratosphere. Games like Cyberpunk 2077 are notoriously brutal on videocards, with the flagship of the 4000 or 5000 series barely able to pop above 30 fps when running without synthetic frames. As you’ll see when you read our benchmarks, DLSS 4 makes a world of difference. You’re welcome to complain about “fake frames” but I’ll be over here playing Cyberpunk 2077 on Psycho settings in 4K at nearly 400 fps.

For those of you concerned about the visual impacts of DLSS 4, I’m certain people will spend exorbitant amount of time and energy looking at still images zoomed in 4000%, but here’s Cyberpunk 2077’s benchmark on every single possible DLSS setting, as well as rasterized performance. As requested, it’s timestamped, so go nuts:

When you DO zoom in, you’ll likely see the VASTLY improved Super Resolution image reconstruction. Standard Super Resolution had a soft and often washed out look, whereas DLSS 4 is near-indistinguishable from a native frame. The new Tensor cores are providing the power for this reconstruction. It’s most evident on mid-distance objects where you might otherwise see shimmering or jagged edges, now replaced with clean edges that interact with objects and the environment properly. Things that would otherwise be obfuscated or blurred badly by motion, such as a plane propeller in Microsoft Flight Simulator 2024, are now clean and sharp as well due to the Transformer Model.

Let’s get into the topic of Multi-Frame Generation. The new Ray Reconstruction Transformer engine pulls a frame directly from the engine. It’s then passed to the Blackwell “Flip Metering” tech to generate up to three additional frames using AI. NVIDIA is stating that it can provide upwards of 8X performance boost over native rendering — something we can test in our benchmarks.

75 games will support DLSS 4 on day one, including Alan Wake II, Dragon Age: The Veilguard, Stalker 2, Diablo IV, God of War: Ragnarök, Cyberpunk 2077, Marvel Rivals, Hogwarts: Legacy, Deep Rock Galactic, Still Wakes the Deep, Remnant II, Indiana Jones and the Great Circle, Senua’s Saga: Hellblade II, Icarus, Hitman: World of Assassination, The Thaumaturge, Deliver Us Mars, The Finals, Star Wars: Outlaws, and much more. Put simply — it’s the most bleeding edge games we’re all playing, not something from the back catalog. You’ll see in our benchmarks that not all of these games have 4X available yet, so we’ll revisit these graphs when those patches arrive.

Whether you use DLSS at all is a personal choice, and one you won’t regret either way as there is more than enough raw power in this card to brute force a solid framerate out of just about anything. The improvements to DLSS and Super Resolution have given us cleaner outputs on these generated frames, though you can still catch ghosting at the edges or between leaves on trees. On a still image it sticks out badly, but in motion and at the right combination of settings, it’s almost imperceptible. Again, this is a matter of personal preference, so test it out and judge for yourself.

One of the things I really appreciate is that NVIDIA is also giving some attention to the 4000 series cards. While they don’t have the hardware to do multi-frame generation, they are getting an update for DLSS 3.5 for single frame gen. We’ll have to test, but NVIDIA is claiming lower memory usage and better performance — upwards of 10% improvement, in fact. It’s a great gift to folks who picked up the 4000 series, and it’s honestly the biggest barrier for most folks who will now ask themselves if they need to upgrade to the 5000 series. For me, the improvements to things outside of games makes it worth it. If you’re a professional rendering video or AI-driven content, the card just delivers in a huge way.

It’s important to note that you enable DLSS 4’s new Transformer Model on a per-game basis currently. It’s not something you’ll find in a game menu for every title. Instead, you’ll need to head into the NVIDIA app, select “Program Settings”, then select the game in question and scroll down to the heading entitled Driver Settings. Under this you’ll find DLSS Override – Frame Generation. Within that option you’ll see “Use the 3D application Setting” and you can select 3X or 4X as a Global option. You’ll find similar options for Super Resolution, RTX, and more — just be mindful that you might want to just let the NVIDIA App tune it as you’ll likely get a more optimized result for the combined hardware you have.

Reflex was introduced back in 2020, providing improvements to mouse latency for cards all the way down to the GeForce GTX 900. The way it works is by optimizing the rendering pipeline between your mouse to the CPU, to the render cue, to the GPU, then to the Display and back reducing overall latency. Reflex 2 aims to do this even more, dropping up to another 50% of the latency in this chain. Frankly, this is just magic as it can turn an otherwise jerky experience into one that is as smooth as silk. If you aren’t excited about DLSS that’s fine, but there’s plenty to be excited about here. Honestly, I can’t do an explanation justice for how frames get shifted in motion, so check out this video from the team:

Testing the effect on latency is something best left to the experts, so I can’t comment on frame warp other than to say that it feels very smooth in my experience. Anecdotally, in shooters like The Finals I saw my latency go from roughly 50ms down to 20ms in Reflex 1 and Reflex 2 down to around 4ms. That’s not a typo – the latency bounced between 2 and 5, but never higher than 5. I wonder if the e-sports community is going to embrace this or ban it as it can be a HUGE advantage in the field. Time will tell.

Onto the benchmarks, but first a note – we are testing pre-launch drivers, so the numbers we’ll be showing will evolve over time. Developers will continue to patch, and NVIDIA will squeeze these cards harder and harder, resulting in better performance overall. As such, please look at the benchmarks as a moment in time. They’re also rather complicated. As you can see in the hour long Cyberpunk 2077 benchmarks above, there are a LOT of options with framerate results ranging from around 30fps at 4K with everything cranked to the maximum all the way up to over 350. As such, our benchmarks will show a sprinkling of those options, but not all of them – that’s an unreadable mess. Not every game will have as extensive an option set as Cyberpunk, but suffice it to say we’ll look at raw power raster output, and then a smattering of options for DLSS 4.

Finding your sweet spot that matches the capabilities of your monitor or TV is key as going further than that is rather pointless. That said, there are plenty of 4K / 240Hz monitors out there, and if you’re willing to spend $1000 on this GPU, what’s another $1000 for a monitor to match it? Anything above 240 is going to be in an elite tier that is so forward-looking that you likely can’t use all that power with today’s hardware – another complication. While there are competition monitors with framerates approaching insane (the upcoming TCL CSOT has a 1,000Hz refresh rate, but isn’t out yet and doesn’t have a price. It’s not gonna be cheap), you’ll have to find the balance of image quality, capability, and desired framerate yourself. I’ll be using Average FPS for all of these with Raster, DLSS On, and then DLSS 4 giving us three data points to compare. Realize that DLSS 4 cranked to ultra performance will not give you the best image quality, so we’ll go with Quality when available. We also don’t quite have the full complement of 4X capabilities on all titles, though all 75 will be available as the card launches publicly. Let’s get to the benchmarks:

In the benchmarks, you can see that not only is this card roughly 30% better in terms of raw hardware numbers, with DLSS 4 it can be upwards of 300% better than anything else on the market. Generational uplift is usually about half this amount, clocking in at around 15% for raw uplift. Yes, the card is expensive, but this is a case where you’re getting far more than you’re paying for.

I want to also note that there’s another game I’m working on with David Burdette – Kingdom Come: Deliverance II. It’s not optimized, and I can’t talk about it until the embargo passes, but I’ll say that it has an “experimental” mode above Ultra settings. It warns you that it’s going to run terribly and that it’s meant for “future hardware”. The RTX 5090 is capable of running the game at 4K / 60 for the most part, though that drops to around 35-40 at 4K for the RTX 5080 Not too shabby for something that you know isn’t optimized and is thrown in for fun. I guess the future is now.

Putting together a review for a video card is a lot of capture and encoding. To that end, I’d be remiss if I didn’t mention the performance on the new encoders. The jump from the 3090 to the 4090 was significant, cutting it more than in half. Videos that once took 30 minutes were coming in at 13 minutes. Now that same video can be done in about 7. Obviously that isn’t going to likely matter much, until you start rendering 4K or 8K video, or a particularly long one. It also seems to be smoother when running in DaVinci Resolve, but that is purely anecdotal. It’s a nice bonus, and I’m glad to see NVIDIA still investing in this area.

We always end our hardware reviews talking about price and warranty. We don’t make assertions on value as that’s different for everyone. That said, this is a $1000 card, cutting the price in half over the flagship card. Obviously to reach that price, concessions have been made, but far less than you’d expect. I suspect we’ll see the delta widen as we see maturation of drivers and game patches, but right now there’s a roughly 15% native performance uplift over the 4080 Super, and at a price that’s at least $500 cheaper. The delta between the 5090 and 5080 is roughly 54% natively, but DLSS 4 is the silver bullet for the 5000 series. As you can see from these benchmarks and videos, you have a VERY wide variety of options to choose from to get precisely the image quality and framerate you want.

While it uncorks some absolutely eye-wateringly high 4K framerates, DLSS 4 will also enable the eventual lower-tier cards to deliver performance well beyond their base capabilities. Throw in the improvements on AI, the new render and encoding engine, Reflex 2, and more and it’s very easy to see the value proposition of the 5080. Obviously these cards will suffer from some artificial scarcity at launch, but be patient – don’t give into scalpers. The wait will be worth it.

 

See below for our list of partners and affiliates: