Verdict
The RTX 5090 produces some absolutely outstanding performance figures. In our testing, it netted an incredible 90 fps on average at 4K across all of our five titles, and it does that without DLSS or Frame Generation thrown into the mix. Similarly, AI performance goes above and beyond, and it’s one of the few cards that have seen a significant hardware uptick compared to last gen.
The problem, if you hadn’t already guessed, is stock, and although the RTX 5090 was initially debuted at $2,000 you can now only find them a lot higher than that.
-
The best gaming GPU on the planet right now -
Available in twin slot designs, and relatively cool -
Massive upgrade on last gen
-
Stock is practically non-existent -
MFG tech can induce latency for some -
Bottlenecks at 1080p
Key Features
-
Review Price: £2000 -
32GB of GDDR7
The RTX 5090 pushes the limits of memory capacity, with 32GB of GDDR7. Not only is this new standard wildly fast, delivering a huge 1.8 TB/s of bandwidth, but it has as much memory as most modern-day gaming PCs actually have. Just on your GPU instead -
GB202-300 GPU Core
One of the biggest GPUs on record, the die alone measures 750 mm² and claps with 92.2 billion transistors and 21,760 CUDA cores. In comparison, the RTX 5080 has just 10,752, and the RTX 4090, just 16,384 -
3 GHz Nearly Unlocked
Our test unit’s GPU max clock reached a phenomenal 2,902 MHz under load, drawing 865.3 W from the wall—that’s huge—and it still managed to keep itself cool at just 76°C while it did it
Introduction
This, right now, is the greatest graphics card Nvidia has ever released. At least from the point of view of pure performance. It has absolutely no equal, and even its predecessor, the RTX 4090, pales in comparison to this goliath.
The spec bump alone from a gen-on-gen standpoint is just astronomical; CUDA cores alone have shot up by 32.8%.
Compared to the “improvements” with the RTX 5080 compared to the 4080 Super, well, it’s an increase of a factor of six, and you can tell, because this thing just demolishes almost every test you throw at it.
But there’s a catch, because of course there is. This is the RTX 50 series after all, and even ignoring some of the ROP issues plaguing certain manufacturers, the reality is that stock isn’t available. Even at launch, it was incredibly limited, and although the MSRP was initially pitched at $2000, some resellers now have these listed at nearly 85% higher than that.
It’s not exactly good value per frame at its retail price, but at that mark-up, it makes for an abysmal purchase. So yes, it is by far the best graphics card you can get today, at least in spirit; it’s just a shame you can’t actually get it.
Specs
The hardware here is utterly wild. Unlike with the RTX 5080 and the lower-specced cards, the 5090 has seen an incredible increase in the overall number of cores available to it. This is in no small part thanks to a massively supersized GPU die leveraging 92.2 billion transistors with an area of 750 mm². In contrast, the RTX 4090, running off a similar 5nm process (albeit slightly less optimised), comes in at 608.5 mm² and 76.3 billion transistors. That is a wild increase.
Top-line specs for the 5090 include 21,760 CUDA cores, 680 TMUs, 176 ROPs, 170 ray tracing cores, and a whopping 680 Tensor cores.
| RTX 5090 – 32GB | RTX 5080 – 16GB | |||||
| GPU | GB202-300 | GB203-400 | ||||
| INTERFACE | PCIe 5.0 x16 | PCIe 5.0 x16 | ||||
| DIE SIZE | 750 mm2 | 378 mm2 | ||||
| LITHOGRAPHY | TSMC 4NP | TSMC 4NP | ||||
| TRANSISTORS | 92.2 Billion | 45.6 Billion | ||||
| CORES | 21,760 | 10,752 | ||||
| BOOST CLOCK ADVERTISED | 2,407 MHz | 2,620 MHz | ||||
| BOOST CLOCK RECORDED | 2,902 MHz | 2,835 MHz | ||||
| MEMORY | 32GB GDDR7 | 16GB GDDR7 | ||||
| MEMORY BUS | 512-bit | 256-bit | ||||
| MEMORY BANDWIDTH | 1,792 GB/s | 960 GB/s | ||||
| TGP | 575W | 360 W | ||||
Across the board, that’s an increase of 32.8% compared to every core count on the RTX 4090. No small figure for sure. On top of that, Nvidia has paired that Blackwell-powered processor with 32GB of Samsung’s latest GDDR7 memory on a 512-bit bus, delivering an astonishing 1,792 GB/s of total memory bandwidth.
That’s an additional 8GB on top of last gen for capacity and 77.8% more speed as a result of the GDDR upgrade.
There are a few more additions here worth mentioning, as it’s also utilising a full-fat PCIe 5.0 x16 connection plus a 575W TDP versus the 4090’s 450W cap. It’s still using the 12VHPWR connector, although it’s a refined version designed to give it a little bit more robustness and hopefully prevent any unwanted cable burnouts (don’t worry, your old 12VHPWR cables are still compatible).
One thing worth noting, however, is that the TDP is sitting pretty at 575W. Nvidia has directly recommended a minimum of a 1000W PSU to run its RTX 5090.
During my time testing, I saw figures run as high as 865W and even higher under some load conditions. I highly recommend you go slightly above that 1000W mark and aim for a 1200W unit instead, purely just for the sake of headroom and comfort.
Test Setup
When it came to really pushing the RTX 5090 to its limits, I wanted to revamp Trusted Reviews’ graphics card benchmarks to showcase just what this card and its surrounding competitors can really deliver in modern titles and settings.
I’ve been testing, benchmarking, and playing around with these GPUs for the last few months, really getting to grips with them and understanding their architectures in all manner of scenarios, so I had a good idea of the games that would really push these things to their limits
To that end, I’m using a combination of both synthetic and in-game benchmarks for the majority of the testing. Five games total across three separate resolutions, with minimum and average frame rates reported.
Each game at each resolution is run three times, to average out results, and I’m testing using the highest possible graphical preset available, with minimal changes, bar a few outliers. Cyberpunk 2077, for instance, I’ve disabled DLSS for its testing, and in Black Myth Wukong I’m only using the “Ultra” graphical preset not the “Extreme” one, with ray tracing additionally disabled. As for Total War: Warhammer 3, I’m using the Mirrors of Madness benchmark to stress the CPU load at the same time.
On top of that, I’ve also run some additional testing for both DLSS and Frame Generation tech at 4K as well. This is utilising Cyberpunk 2077 and Black Myth Wukong again (with ray tracing enabled this time). In those scenarios, I’m testing using DLSS set to “Quality” (67% in Black Myth), and then a separate set of runs with Frame Generation enabled as well. To keep it a fair playing field, and so we can compare against additional older cards, I’m only allowing Cyberpunk to utilise frame-gen at a x2 ratio, not the maximum x4.
Synthetic testing, stems mostly around UL’s 3D Mark and Procyon benchmark suites. For 3D Mark that includes Steel Nomad and Speedway. Procyon is all about AI computational performance for us, and for that, Computer Vision, and Image Generation are the name of the game. I’ve tested all the cards, with either the default APIs, or with Nvidia’s TensorRT.
- CPU: AMD Ryzen 9 9900X
- RAM: 32GB (2x16GB) Team Group T-Create Expert DDR5 @ 6000 C34
- Motherboard: ASUS ROG Strix X870E-E Gaming WiFi
- CPU Cooler: Tryx Panorama 360mm AIO
- Cooling: 10x NZXT RGB Duo 120mm fans
- PSU: 1500W NZXT C1500 2024 80+ Platinum PSU
- SSD 1: 2TB Crucial T705 PCIe 5.0 M.2 SSD
- SSD 2: 4TB Crucial T500 PCIe 4.0 M.2 SSD
- Case: NZXT H9 Elite
As for the test bed, I’ve decided to go with AMD this time around, specifically with the Ryzen 9 9900X.
Although potentially an X3D chip may have delivered higher frame-rates, the fact its 3D V-Cache tends to vary in performance levels dependent on the title that’s running it, and whether or not that can take advantage of it, I’ve opted to stay with the standard X version instead, preferring to lean on greater multi-core performance and higher clockspeeds as a result.
Both Intel and AMD’s latest processors are beastly when it comes to the fight to become the best gaming CPU, so you shouldn’t see too much of a difference going from one chip to the other (in fact my own personal testing can back that up as well).
Gaming Performance
- Unbelievable 4K performance
- 1080p leads to major bottlenecks
- Dominant even without DLSS
Nvidia’s RTX 5090 is one of those graphics cards that just defies all expectations. Yes, the price point is high, the power draw transcendent, and its girth (certainly with the Zotac Solid variant I have for testing) unfathomable. And yet, despite those minor foibles (and stock availability, but we’ll get onto that later), it just absolutely decimates the competition at the high-end.
There’s no argument; from a pure raw performance standpoint, this is the greatest graphics card of all time, at least by 2025’s standards.
| AVERAGE ACROSS ALL 5 TITLES | – FPS – |
| 1080p – Average FPS | 140.77 |
| 1080p – Minimum FPS | 73.05 |
| 1440p – Average FPS | 127.7 |
| 1440p – Minimum FPS | 71.71 |
| 4K – Average FPS | 89.13 |
| 4K – Minimum FPS | 57.98 |
Let’s take those 4K numbers as an example. Average frame rate across all five titles at 4K was 89.13 fps; the minimum was 57.98.
Yeah, its minimum frame rate was nearly hitting that fabled 60 frames we all know and love as standard at 4K. And the thing is, our testing suite isn’t kind by design. Both Cyberpunk and Black Myth Wukong are massive frame-rate drains compared to most games, and none of the games on test are utilising DLSS to generate those numbers.
Cyberpunk 2077, for instance, landed at 54.9 fps, and Black Myth at 60.3. Jump into Metro Exodus or Total War: Warhammer 3, and you’re talking 93 fps and 80 fps, respectively. Again, it’s driving near enough 8.3 million pixels that rapidly. It’s just astounding.
At 2560×1440, average frame rates sit comfortably at 127.7 fps, and 1080p sees a lesser performance bump, with average frames across all titles landing at 140.77 fps as well. Those are solid numbers. Yes, that 1080p number is seemingly quite low, but I suspect that’s more likely down to the CPU becoming the bottleneck for the majority of those games.
Gaming Benchmarks
| 1920 x 1080 – GAME TESTING | ||||
| Cyberpunk 2077 | Average Framerate – Runs | 108.9 | 108.0 | 107.9 |
| Cyberpunk 2077 | Minimum Framerate – Runs | 83.6 | 83.5 | 85.2 |
| Cyberpunk 2077 | Average Framerate – Overall Average | 108.2 | ||
| Cyberpunk 2077 | Minimum Framerate – Overall Average | 84.1 | ||
| Black Myth Wukong | Average Framerate – Runs | 111.0 | 111.0 | 112.0 |
| Black Myth Wukong | Minimum Framerate – Runs | 74.0 | 69.0 | 73.0 |
| Black Myth Wukong | Average Framerate – Overall Average | 111.3 | ||
| Black Myth Wukong | Minimum Framerate – Overall Average | 72.0 | ||
| Metro Exodus | Average Framerate – Runs | 160.2 | 158.5 | 153.7 |
| Metro Exodus | Minimum Framerate – Runs | 79.1 | 78.3 | 76.0 |
| Metro Exodus | Average Framerate – Overall Average | 157.4 | ||
| Metro Exodus | Minimum Framerate – Overall Average | 77.8 | ||
| Final Fantasy XIV | Average Framerate – Runs | 242.8 | 237.8 | 240.1 |
| Final Fantasy XIV | Minimum Framerate – Runs | 78.0 | 78.0 | 77.0 |
| Final Fantasy XIV | Average Framerate – Overall Average | 240.2 | ||
| Final Fantasy XIV | Minimum Framerate – Overall Average | 77.7 | ||
| Total War: Warhammer 3 | Average Framerate – Runs | 88.5 | 85.0 | 86.3 |
| Total War: Warhammer 3 | Minimum Framerate – Runs | 56.0 | 52.0 | 53.0 |
| Total War: Warhammer 3 | Average Framerate – Overall Average | 86.6 | ||
| Total War: Warhammer 3 | Minimum Framerate – Overall Average | 53.7 | ||
| 2560 x 1440 – GAME TESTING | ||||
| Cyperpunk 2077 | Average Framerate – Runs | 95.5 | 95.3 | 94.2 |
| Cyperpunk 2077 | Minimum Framerate – Runs | 80.4 | 81.3 | 81.2 |
| Cyperpunk 2077 | Average Framerate – Overall Average | 95.0 | ||
| Cyperpunk 2077 | Minimum Framerate – Overall Average | 81.0 | ||
| Black Myth Wukong | Average Framerate – Runs | 93.0 | 93.0 | 92.0 |
| Black Myth Wukong | Minimum Framerate – Runs | 71.0 | 71.0 | 71.0 |
| Black Myth Wukong | Average Framerate – Overall Average | 92.7 | ||
| Black Myth Wukong | Minimum Framerate – Overall Average | 71.0 | ||
| Metro Exodus | Average Framerate – Runs | 144.2 | 143.6 | 142.8 |
| Metro Exodus | Minimum Framerate – Runs | 77.5 | 76.5 | 75.6 |
| Metro Exodus | Average Framerate – Overall Average | 143.5 | ||
| Metro Exodus | Minimum Framerate – Overall Average | 76.5 | ||
| Final Fantasy XIV | Average Framerate – Runs | 224.2 | 222.7 | 221.2 |
| Final Fantasy XIV | Minimum Framerate – Runs | 75.0 | 76.0 | 76.0 |
| Final Fantasy XIV | Average Framerate – Overall Average | 222.7 | ||
| Final Fantasy XIV | Minimum Framerate – Overall Average | 75.7 | ||
| Total War: Warhammer 3 | Average Framerate – Runs | 84.6 | 84.6 | 84.7 |
| Total War: Warhammer 3 | Minimum Framerate – Runs | 53.0 | 56.0 | 54.0 |
| Total War: Warhammer 3 | Average Framerate – Overall Average | 84.6 | ||
| Total War: Warhammer 3 | Minimum Framerate – Overall Average | 54.3 | ||
| 3840 x 2160 – GAME TESTING | ||||
| Cyperpunk 2077 | Average Framerate – Runs | 54.8 | 55.0 | 55.0 |
| Cyperpunk 2077 | Minimum Framerate – Runs | 48.8 | 48.9 | 48.7 |
| Cyperpunk 2077 | Average Framerate – Overall Average | 54.9 | ||
| Cyperpunk 2077 | Minimum Framerate – Overall Average | 48.8 | ||
| Black Myth Wukong | Average Framerate – Runs | 59.0 | 61.0 | 61.0 |
| Black Myth Wukong | Minimum Framerate – Runs | 47.0 | 48.0 | 48.0 |
| Black Myth Wukong | Average Framerate – Overall Average | 60.3 | ||
| Black Myth Wukong | Minimum Framerate – Overall Average | 47.7 | ||
| Metro Exodus | Average Framerate – Runs | 93.7 | 92.9 | 92.6 |
| Metro Exodus | Minimum Framerate – Runs | 63.0 | 63.3 | 63.1 |
| Metro Exodus | Average Framerate – Overall Average | 93.0 | ||
| Metro Exodus | Minimum Framerate – Overall Average | 63.1 | ||
| Final Fantasy XIV | Average Framerate – Runs | 158.0 | 157.0 | 157.2 |
| Final Fantasy XIV | Minimum Framerate – Runs | 75.0 | 75.0 | 75.0 |
| Final Fantasy XIV | Average Framerate – Overall Average | 157.4 | ||
| Final Fantasy XIV | Minimum Framerate – Overall Average | 75.0 | ||
| Total War: Warhammer 3 | Average Framerate – Runs | 79.5 | 79.3 | 81.1 |
| Total War: Warhammer 3 | Minimum Framerate – Runs | 55.0 | 56.0 | 55.0 |
| Total War: Warhammer 3 | Average Framerate – Overall Average | 80.0 | ||
| Total War: Warhammer 3 | Minimum Framerate – Overall Average | 55.3 | ||
Ray Tracing, DLSS, & AI Performance
- Fps increases with DLSS
- Impressive AI Image generation skills
Of course, in the land of AI upscaling and frame generation, DLSS 4 really shines, and its new transformational model takes a lot of the blurriness and latency out of the equation that its old convolutional model used to introduce. I certainly don’t suffer with it anymore. Although I will say I am still a big advocate of traditional rasterisation and MSAA, even still.
Nonetheless, with DLSS on quality, Cyberpunk 2077 at 4K saw frame rates increase by 26 fps. Add in frame generation to the mix, on an x2 factor, and that DLSS score nearly doubles, with the RTX 5090 landing an outstanding 146.1 fps.
| GAME | DLSS + FRAME GEN TESTS @ 4K | FPS | ||
| Cyberpunk 2077 | Average Framerate – Runs – DLSS ONLY | 94.5 | 92.9 | 93.2 |
| Cyberpunk 2077 | Minimum Framerate – Runs – DLSS ONLY | 81.9 | 81.2 | 81.8 |
| Cyberpunk 2077 | Average Framerate – Overall Average | 93.5 | ||
| Cyberpunk 2077 | Minimum Framerate – Overall Average | 81.6 | ||
| Cyberpunk 2077 | Average Framerate – Runs – DLSS + FG | 164.0 | 162.0 | 161.5 |
| Cyberpunk 2077 | Minimum Framerate – Runs – DLSS + FG | 146.9 | 145.9 | 145.4 |
| Cyberpunk 2077 | Average Framerate – Overall Average | 162.5 | ||
| Cyberpunk 2077 | Minimum Framerate – Overall Average | 146.1 | ||
| Black Myth Wukong | Average Framerate – Runs – DLSS ONLY | 64.0 | 63.0 | 63.0 |
| Black Myth Wukong | Minimum Framerate – Runs – DLSS ONLY | 45.0 | 46.0 | 45.0 |
| Black Myth Wukong | Average Framerate – Overall Average | 63.3 | ||
| Black Myth Wukong | Minimum Framerate – Overall Average | 45.3 | ||
| Black Myth Wukong | Average Framerate – Runs – DLSS + FG | 104.0 | 104.0 | 104.0 |
| Black Myth Wukong | Minimum Framerate – Runs – DLSS + FG | 89.0 | 89.0 | 88.0 |
| Black Myth Wukong | Average Framerate – Overall Average | 104.0 | ||
| Black Myth Wukong | Minimum Framerate – Overall Average | 88.7 | ||
Black Myth Wukong might look like DLSS is having no effect, but you have to remember that in its standard testing scenario, it doesn’t have ray tracing enabled. Whereas in our DLSS tests, it does. With TSR at 100% and DLSS at 67%, frame rates remain within the margin of error at 4K.
That in of itself is impressive, as ray tracing considerably stymies frame rate when enabled here. Frame-gen also bumps the figures up too, chucking another 39.7 fps on top for the 5090.
As for our AI tests, UL Procyon’s Computer Vision, scored a neat 4,879 points with Nvidia TensorRT. That’s not actually as high as I’d expect. The RTX 5080, for reference, landed a score of 4272, and the RTX 4080 Super, 3728.
Image generation, mind you, was a night-and-day difference. With Tensor, the 5090 scored 8333 points, finishing the test in just 12 seconds and producing an AI-generated image in 0.75 seconds per image. That just dwarfs the 5080 and the 4080 Super by a huge margin.
Power Consumption & Temperature
Throughout my testing, I utilised a combination of power meters and HWMonitor, plus a room thermometer to give me a good understanding of how these cards run.
My test bed includes 11 120mm fans in a positive airflow setup, with the GPU being fed directly by three intakes to keep things chilled. My RTX 5090 is Zotac’s Solid AIB model, which is a seriously chunky triple-slot design complete with a thick backplate and its own triple fan cooling solution as well.
| POWER & TEMPS | |||
| Ambient Room Temperature | 24.0 °C | ||
| Max GPU Temp During Testing | 76.0 °C | ||
| Max Power Draw During Testing | 865.3 W | ||
Temps were very solid, maxing out at 76 C under load. I never saw it top much higher than that either, even during seriously long gaming sessions. It does, however, tend to superheat your office a tad, so I’d highly recommend either a good window, air conditioning, or just be willing and ready for it to be toasty, even under idle temps. That’s a lot of GPU die to cool after all.
Power draw capped at 865.3W under maximum load; ironically, it was Final Fantasy XIV that drew the highest peak impressively. Again, if you’re looking at this card, you want to get the best PSU you can budget for, preferably at around 1200W or so.
Should you buy it?
You want the best graphics card money can buy, and can wait for stock
60 fps at 4K without gimmicks, without DLSS, no AI upscaling, nothing. It just dominates those frame rates.
Combined with 32GB of VRAM and some serious AI tech backing it all up, it’s a card that’s going to carry you forward for many years to come. The only downside is if you can’t find stock of it or it’s at that crazy $3700 price point.
You don’t have the budget to back it up
Not only do you need the graphics card itself, but you need to ensure you’ve got one of the best monitors you can get to really takeadvantage of it. Combine that with a sizable PSU and a seriously quick CPU to back that all up, and all of a sudden it’s not quite the small $2000 upgrade you first thought it was.
Final Thoughts
There’s no doubt in my mind that the RTX 5090 is one of the best graphics cards out there on the market today.
For pure performance, it has no equal. In fact, the second-best GPU out there is arguably the RTX 4090, and that’s a two-year-old card at this point. It’s just unparalleled; that 32.8% boost to the die size, the hardware on it, and the overall performance that you get at 4K with no upscaling is just outstanding.
But that comes with some fairly major caveats. No matter how you look at it, even at its retail price, the price-to-performance ratio is seriously lacking. If you’re going into this looking for a good value card or trying to convince yourself of that, you’re going to be in for a bad time. Particularly to get the most out of this, you really do need a high refresh rate 4K display to drive it on.
It is by far the best graphics card you can get today, at least in spirit; it’s just a shame you can’t actually get it easily.
How we test
Each and every graphics card that comes in for review gets thrown into its own fixed, dedicated test bench to be benchmarked and compared against other cards in the GPU results suite.
There’s a total of 67 data points we use to analyze each GPU, covering all three key resolutions, along with GPU temperature, system power draw, $ per frame, and many, many more.
We use a combination of synthetic and in-game benchmarks to give us a good understanding of how these cards perform, both in isolation and against each other, before making any judgements, while also considering real-world stock levels and pricing.
- Reviewed in a dedicated test setup
- 67 total data points captured
- In combination of synthetic and in-game benchmarks
Test Data
Full Specs
| Nvidia GeForce RTX 5090 Review |
|---|
