Nvidia’s GeForce RTX 5090 hype train left the station last week. The Big Daddy card wallops everything that’s come before, but the price you pay for pixel-pushing greatness is vast. I expect availability to be slimmer than Man Utd’s chances of making a title run in the next few years, while I won’t be surprised seeing scant few partner cards hitting the retail shelves at over £2,500.
If you’re looking for better value, let me point you in the direction of GeForce RTX 5080. Also available from tomorrow, the $999/£979 starting fee is more palatable, though expect to pay a hefty premium for four-slot partner offerings.
Architecture speaks to RTX 5080 being literally half an RTX 5090 in all metrics that matter, but I know performance is more than 50%, and certainly good enough to game with all the bells, whistles, and frames turned on. If you’re reading this, let me sate curiosity be comparing Nvidia’s two best RTX 50 Series GPUs head-to-head.
| RTX 5080 | RTX 5090 | Ratio | |
|---|---|---|---|
| Released | Jan 2025 | Jan 2025 | – |
| Codename | Blackwell | Blackwell | – |
| GPU | GB203 | GB202 | – |
| Process | TSMC 4N | TSMC 4N | |
| Transistors bn | 45.6 | 92.2 | 0.49 |
| Die size mm² | 378 | 744 | 0.51 |
| CUDA cores | 10,752 | 21,760 | 0.49 |
| Base clock MHz | 2,295 | 2,017 | 1.14 |
| Boost clock MHz | 2,617 | 2,407 | 1.09 |
| FP32 Boost TFLOPS | 56.3 | 104.8 | 0.54 |
| SM count | 84 | 170 | 0.49 |
| RT cores | 84 (4th Gen) | 170 (4th Gen) | 0.49 |
| RT TFLOPS | 170.6 | 317.5 | 0.54 |
| Tensor cores | 336 (5th Gen) | 680 (5th Gen) | 0.49 |
| FP16 Acc TFLOPS | 450.2 | 838 | 0.54 |
| FP4 TOPS | 1,801 | 3,352 | 0.54 |
| ROPS | 112 | 176 | 0.64 |
| Memory GB | 16 | 32 | 0.50 |
| Memory type | GDDR7 | GDDR7 | – |
| Mem. clock Gb/s | 30 | 28 | 1.07 |
| Mem. interface bits | 256 | 512 | 0.50 |
| Mem. bandwidth GB/s | 960 | 1,792 | 0.54 |
| Board power watts | 360 | 575 | 0.63 |
| Launch MSRP $ | 999 | 1,999 | 0.50 |
Tale of the tape underscores that Nvidia’s positioning strategy remains intact from one generation to the next. RTX 4080 (Super) followed an eerily similar trajectory when compared to RTX 4090, so why change a winning formula.
RTX 5080 uses a smaller die known as GB203. Conveniently enough, even at a whopping 45.6 billion transistors, it’s almost exactly 50% the size of outlandish GB202 powering the RTX 5090 behemoth. Nevertheless, unlike said powerhouse card, Nvidia uses the full die for this x80 model. If there is to be an RTX 5080 Super down the track, it won’t use existing GB203 in any meaningful way – there’s no room to accommodate more cores.
Nvidia endows RTX 5080 with higher base and boost shader frequencies, pushing FP32 TFLOPs to just over half RTX 5090, boding well for performance. Going further down the table cements this notion of RTX 5080 being more than exactly half, albeit not by a huge degree.
Giving RTX 5080 a further helping hand, Nvidia uses the fastest memory I’ve seen on a consumer card – 30Gb/s GDDR7 hums along across a 256-bit bus, offering a notional near-1TB/s bandwidth. Lovely stuff.
Elevating frequencies across cores and memory has an inevitable knock-on effect of increasing total board power by more than one would ordinarily think. I’m also sure Nvidia keeps the best frequency/voltage silicon for RTX 5090, meaning RTX 5080 is driven to higher-than-linear wattages. All that said, 360W RTX 5080 isn’t all much hungrier than RTX 4080 Super, so no problems in terms of power delivery or card design.
Performance
I’ve put both graphics cards through their paces using Club386’s trusty Ryzen 9 7950X3D-led test bench, updated to the latest drivers on all fronts.
Our 7950X3D Test PCs
Club386 carefully chooses each component in a test bench to best suit the review at hand. When you view our benchmarks, you’re not just getting an opinion, but the results of rigorous testing carried out using hardware we trust.
Shop Club386 test platform components:
CPU: AMD Ryzen 9 7950X3D
Motherboard: MSI MEG X670E ACE
Cooler: Arctic Liquid Freezer III 420 A-RGB
Memory: 64GB Kingston Fury Beast DDR5
Storage: 2TB WD_Black SN850X NVMe SSD
PSU: be quiet! Dark Power Pro 13 1,300W
Chassis: Fractal Design Torrent Grey
Application and AI
RTX 5080 is keen to state it’s no graphics midget. Cranking out almost two-thirds of RTX 5090, it’s a formidable offering in its own right.
Steel Nomad hits the cards a tad differently. Nevertheless, I’m seeing 60% of Big Boy’s performance, which is exactly what I’d expect in a GPU-bashing test.
RTX 5080’s never getting close to RTX 5090 when everything is lit up, but everything ought to be considered in context. Other than what you see above, I’ll let you in on a little secret; only RTX 4090 from the previous generation is able to beat RTX 5080 in this test. Perspective and all that.
Geekbench AI’s half-precision score is more about frequency than cores. That’s why RTX 5080’s pretty close.
Want to run a large language model right on your desktop? RTX 5090 is the best bet, of course, but the ’80 card is no slouch. Any score above 3,000 represents near-instant responses from this text generation model.
Gaming
Synthetics and AI is all fine and dandy, but I bet you’re here to find out how these two heavyweights stack up against one another when gaming. Assassin’s Creed Mirage relies on brute FP32 core power and bags of memory bandwidth, meaning RT and Tensor cores take a backseat.
Performance delta escalates in concert with resolution. Sure, RTX 5090 is 36% faster at 4K but I feel that’s missing the point in this title. A 110fps average and 85fps low 1% is plenty fine. Sometimes you don’t need the absolute best for a great gaming experience.
The same is mostly true in Final Fantasy XIV: Dawntrail. I’d certainly describe RTX 5080 as a 4K gaming card, though the performance aficionado in me can’t help but marvel at the 53% uplift by switching to RTX 5090.
Run with every setting maxed out along with heavy hitting ray tracing, Forza Motorsport brings the pain at 4K. This is one game crying out for multi frame generation… more on that later.
Huge numbers all round. RTX 5090 reminds RTX 5080 who is the boss. There’s something almost rude about 4K240+ on max settings.
Nothing separates the pair at FHD. Heck, even UHD is damn fine, and I believe RTX 5080’s higher ratio of ROPS helps push rasterisation performance higher than what one might expect from a pure spec sheet alone.
An exercise in GPU cruelty, Cyberpunk 2077 run with Ray Tracing Overdrive brings path tracing out to play. Even RTX 5090 resembles Oliver Twist in please, sir, can I have some more fps. Nvidia has a solution to this frame rate pauperism.
DLSS
Adding in DLSS Quality doubles frame rates at 4K. The game’s not totally smooth on either card, to be fair, as the base fps is too low.
Hold your horses, Nelly. Here’s 4x multi frame generation flexing muscles on RTX 50 Series. Numbers are astonishing, though I recommend using this setting at FHD or QHD because it feels smooth owing to the native frame rate being reasonably good. That’s just how upscaling and frame generation works.
| Game | 5080 to 5090 % uplift at 4K |
|---|---|
| Assassin’s Creed Mirage | 36.3 |
| Final Fantasy XIV Dawntrail | 52.7 |
| Forza Motorsport | 47.4 |
| Mount and Blade | 56.1 |
| Rainbow Six Extraction | 45.9 |
| Cyberpunk 2077 DLSS + FG | 59.9 |
It’s sometimes hard to visualise exactly what’s going on when presented with a bunch of graphs. I’ve therefore tabulated RTX 5080 to RTX 5090 percentage gains at 4K in six games, going from a worst-case 36.3% to a best-case 59.9%. Though the architecture discussion paints a lopsided picture in favour of the behemoth, you’re never going to see 100% improvement between cards. Even 90% is impossible in gaming scenarios.
I’m not scoffing at a 50% gain, of course, but I don’t way to pay double for the privilege. In that sense, RTX 5080 is a good card, especially so as 4K frame rates remain healthy in all but the most extreme cases.
If you’ve found this comparison useful, please do read the following head-to-heads.
GeForce RTX 5080 vs. RTX 4080 Super
GeForce RTX 5080 vs. RTX 3080
GeForce RTX 5080 vs. AMD Radeon RX 7900 XTX
Vitals
Evaluated as a system, you also don’t need twice the power, which stands to reason. A gaming load 475W is consistent with RTX 4080 Super and Radeon RX 7900 XTX, by the way.
Ostensibly the same cooler tasked with fewer watts means cooler temperatures.
But about the same noise.
Conclusion
I can successfully argue GeForce RTX 5080 is a more important GPU than RTX 5090 because the starting $999 pricing brings into view of a greater number of prospective buyers.
Built by halving the head honcho’s vital silicon attributes, performance is better than you might expect on paper. The chief card is around 30-60% quicker in most titles, but that’s not an overly useful metric. Perusing results, I see RTX 5080 exhibit very healthy frame rates at 4K in everything other than Cyberpunk 2077 path tracing, suggesting to me it’s a fine card in its own right.
In a perfect world where money is no object, I’d absolutely choose RTX 5090 every day of the week. But money is an issue for most gamers. In that sense, RTX 5080 is a worthy compromise with respect to performance per dollar.
Arguably the best graphics right now for $999/£979, GeForce RTX 5080 stands tall in the face of its bigger, more muscular brother.
GeForce RTX 5080
“Gear up for game-changing experiences with the Nvidia GeForce RTX 5080 and AI-powered DLSS 4.” Read our review.
GeForce RTX 5090
“GeForce RTX 5090 is the absolute pinnacle of PC gaming. If you can afford it, just go right ahead and buy one.” Read our review.
