'%20x='0'%20y='0'%20height='100%25'%20width='100%25'%20%0A%20%20%20%20%20%20%20%20%20%20xlink%3Ahref='data:image/jpg;base64,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'%3E%3C/image%3E%3C/svg%3E)
The Nvidia RTX 5080 is a 4K gaming graphics card launched as part of the latest Blackwell lineup. With better rasterization and AI-powered upscaling prowess, the GPU aims to replace the RTX 4080 Super. The latest 80-class offering has been pegged at the same $999 as the last gen. However, the high price tag puts it among some of the most expensive GPUs you can get your hands on.
I have been testing the Founder's Edition variant of the card in my personal setup for a couple of weeks. During this period, I tested gaming, raw capabilities, and AI workloads on the card. Here's my review of the RTX 5080.
The Nvidia RTX 5080 is a rendering powerhouse: But how does it compare to the RTX 4080?
The Nvidia RTX 5080 is filled to the brim with hardware. You get 10,752 CUDA cores, a 5% uplift over the last generation, but the VRAM buffer remains the same at 16 GB.
Trending
However, the memory has been upgraded to GDDR7 and has a speed of 30 Gbps, directly translating to a 30% uplift in total memory bandwidth. This is particularly useful in demanding games at higher resolutions and helps future-proof the GPU.
Below is the specs sheets of the last-gen and the new 80-class video cards.
Looking at the specs, we don't get a massive jump in terms of the underlying hardware. Across the board, CUDA, Tensor, and RT core counts have only improved by about 5%.
Nvidia claims the biggest gains this generation is in how the hardware is packaged and in AI. With multi-frame generation, you can now get up to 4x more frames, making even the most demanding games buttery smooth.
Unboxing experience
Nvidia never lets gamers down with its Founder's Edition GPU packaging, and Blackwell is no exception. It comes in a minimalist crate-like box that looks inspired by in-game elements from Apex Legends. Moreover, the off-white colorway of the package and the subtle branding add a premium look.
The "little is more" vibe continues to how the card is presented as well. The main package just contained the card with ample cushioning to protect it from any jolts or moves during shipping. Cables and additional paperwork were stuffed away in a separate small box. All of this was well-protected inside a rather large carton.
The RTX 5080 Founder's Edition is an engineering masterpiece
While the design language of the new Founder's Edition card hasn't vastly changed, the GPU has dramatically shrunk in size from the last generation. 50 series cards are two slots wide, which is a welcome move given the large 3+ slot cards we have been getting over the past few generations.
At the core of the slimline design is a completely revamped cooling unit. Nvidia now utilizes two large 140mm fans on the rear, both of which are completely blow-through.
This helps remove sufficient heat from the heatsink to the point where massive blocks of metal aren't required anymore. However, this also means that most heat is escaping through the front of the card, which could directly warm up your CPU and RAM.
The card, however, feels extremely premium to the touch. The black and metal finish makes it well-suited for no-RGB builds. It doesn't stand out in your setup, which is pretty ironic given the performance it's capable of.
Performance tests of the RTX 5080
The Nvidia RTX 5080 was paired with the AMD Ryzen 9 9900X for our tests. We used 32 GB of DDR5 RAM and a high-end Asus X870 motherboard for all results. Our testbench is detailed as follows:
- CPU: AMD Ryzen 9 9900X
- Motherboard: Asus ROG X870-A Gaming WiFi
- RAM: 2x G.Skill Trident Z DDR5-6000 16 GB
- Storage: 1x Gigabyte Gen 4 NVMe 1 TB, 1x Patriot P300 M.2 PCIe Gen 3 x4 128GB
- Cooler: Cooler Master Atmos 240mm liquid cooler
- Storage: Cooler Master MWE 1050W 80+ Gold
- GPU: Nvidia RTX 5080 16 GB Founder's Edition
Synthetic benchmarks
3DMark Fire Strike is our first test. While DirectX 11 has fallen out of fashion, this benchmark tests raw rasterization and rendering horsepower of a GPU. Nvidia's Direct3D 11 drivers are pretty mature. Hence, it gives us a baseline to compare older GPUs against.
Moving on to 3DMark Time Spy, a DirectX 12 test, we can expect larger gains on newer GPUs as they have hardware that makes use of rendering technologies introduced with the tech stack. However, do note that any card released after 2018 has enough capabilities to pull through the test with flying colors.
Ray tracing is where the new Blackwell and Ada Lovelace cards truly shine. With more RT cores and improved CUDA and Tensor hardware, the 50 series pulls off a massive gain in 3DMark's ray tracing feature test, delivering almost double performance since our last champion, the RTX 4080 laptop GPU.
3DMark Port Royal is a test of real-time ray tracing, the kind that's used in video games. In this benchmark, the 5080 almost matches the RTX 4090 in terms of the total score.
3DMark Speed Way combines DirectX 12 Ultimate tech stack with ray tracing for benchmarking high-end gaming PCs. The RTX 5080 takes a massive lead over the last champion, the Radeon RX 7900 XTX, in this benchmark while costing about the same in MSRP.
The DLSS performance is one of the highlighted features of the Nvidia RTX 50 series. With this generation, the company has introduced multi-frame generation, which multiplies framerates by a staggering 2-4x, a gain reflected in 3DMark's DLSS feature test.
The RTX 5080 is undoubtedly one of the fastest cards ever made. However, the gains over the last generation aren't that big. The GPU is still much slower than the 4090, especially in raw rasterization tests. This makes the card lucrative to only those coming from something older — like the 2080 Super or the 3080.
Gaming benchmarks
We tested the 5080 in a variety of video games across resolutions. The graphics card is designed to play the latest titles without any performance hiccups—and it delivered on that promise.
We logged playable framerates across a variety of titles with ray tracing turned on. Moreover, with DLSS 4 and frame generation turned on, we logged about 150-200 FPS consistently in demanding workloads like Cyberpunk 2077 and Marvel's Spider-Man 2.
Frame generation is available in three setups: 2x, 3x, and 4x. We tested each one of them in detail in Cyberpunk 2077. Together with transformer-based upscaling, they delivered 293%, 409%, and 525% better framerates, pushing the final average framerate to an impressive 163 FPS.
The AI powering these GPUs has been baked to perfection by Team Green to deliver smooth experiences in the most demanding workloads.
Check out our RTX 5080 gaming benchmarks article for detailed insights into what kind of framerates you can expect in the latest titles.
Power draw and operating clock speeds
The RTX 5080 is rated at an eye-watering 360W in terms of total power budget. You will need a high-end power supply with PCIe 5.0/5.1 capabilities to be able to sustain the card. Moreover, the slimline design of the Founder's Edition variant also means the GPU does get a bit hot under full load.
Below is a look at the temperatures of the GPU under sustained torture with Furmark 2. We logged 71.5°C. The ambient temperature for the test was 21°C sustained through air-con.
In gaming, the temperature characteristics are far more varied. We tested the card at the highest settings in Cyberpunk 2077 without DLSS to get the worst-case scenario. The highest logged temperature was 65°C, which is very respectable given the two-slot design of the GPU.
Coming to power draw tests, the GPU can quickly scale up to the maximum rating if the workload is demanding enough. Let's start things with our Cyberpunk 2077 stress. At the most demanding scenarios, the 5080 drew 323W. However, the numbers vary across the board, with an average power draw of 235W.
In synthetic workloads, the power draw is more sustained. We logged these numbers with Furmark 2 torture test. The card consistently drew 360W with a short spike to 372W.
While we didn't notice any such spikes in gaming, this outlier proves the requirement of enough power headroom — don't opt for anything less than 750W while building with the RTX 5080.
Buying the RTX 5080 for AI? Watch out
One of the highlighted features of the new 50 series lineup is their AI capabilities. Being an AI enthusiast myself, I was excited to check out what the GPU can do in my work, which generally focuses on fine-tuning and training large transformer-based models.
I generally use an RTX 4060 8 GB GPU for short proof-of-concept training. Coming to the RTX 5080 definitely led to much shorter times in training such models. However, there's a catch.
Right now, not all AI libraries (such as PyTorch, vLLM, Transformers, and Diffusers) are patched for the latest 50 series lineup. While you can get nightly builds for PyTorch from Hugging Face, missing dependencies would mean loading models with libraries like Unlsoth and vLLM isn't possible.
For instance, I had to write training code for BERT from scratch to get some fine-tuning done. Basic inference with Transformers works — you can get LLMs of up to 7B parameters running with 8-bit quantization. However, libraries with reinforcement-learning algorithm implementations like Unsloth simply didn't work for me. This is why I am skipping a dedicated AI benchmarks section for now.
If you're buying the RTX 5080 primarily for AI — watch out! Right now, you are much better off with a 4080 Super instead. Indeed, the 5080 is powerful as in some of my preliminary tests fine-tuning DeBERTa and RoBERTa, I found it rival the A6000 in terms of training. However, the 16 GB VRAM buffer can be a problem if you're doing serious work with large LLMs.
Value and conclusion
For the bottom line, my experience with the RTX 5080 is a mixed bag. While the card truly outshines its performance, the performance gap with the last generation makes it feel more like an RTX 4080 Ti than a truly next-gen product. Nvidia promised high gains with this generation, but those come with the big disclaimer of AI-generated artificial gains.
While DLSS is good, enthusiasts like me can't help but notice its pitfalls. Visual artifacts, while subtle, still haven't disappeared. Moreover, frame generation isn't the dramatic solution to the absence of rendering capabilities — the problem exacerbates in demanding titles like Cyberpunk and Spider-Man 2.
Should you buy the RTX 5080? The answer to that depends. If you already have a powerful 4K-capable GPU like the 4080 or the 3080 Ti, maybe not. Wait for a 5080 Super or Ti refresher to get a more capable pixel-pusher to launch. However, if you are coming from a lower-tier GPU, like the 3070 or a 2080, the upgrade to 5080 can be worthwhile.
Coming to the card's AI capabilities, the reality is that it's the best you can get for less than $1,000. While AMD GPUs might seem more lucrative, the extra drama associated with ROCm simply renders them unworthy. If you are an LLM enthusiast or are a small research lab without substantial funding and can squeeze your work into 16 GB of VRAM, get a 5080.
