NVIDIA's Ada Lovelace generation of GPUs made big waves just weeks ago at GTC when the new architecture was first presented. Today we bring you our comprehensive review of the NVIDIA GeForce RTX 4090 Founders Edition, tomorrow we'll have seven additional reviews of various custom designs from ASUS, Colorful, Gigabyte, MSI, Palit and Zotac.

Unlike Ampere, which saw the RTX 3080 released first, and RTX 3090 later, NVIDIA is starting with the RTX 4090 this time. The new GeForce RTX 4090 is based on the AD102 graphics processor, which is the world's first 5 nanometer GPU, fabricated at TSMC Taiwan. On the RTX 4090, NVIDIA has enabled 16384 GPU cores (+88% vs RTX 3080, +52% vs RTX 3090 Ti)—this alone will achieve a big performance boost. NVIDIA didn't just add "more", they also made their units smarter. While the CUDA Cores haven't really changed since Ampere, the company increased L2 cache significantly, up to 72 MB from 6 MB on the RTX 3090 Ti—a huge increase. The ray tracing cores got several performance improvement features, like shader execution reordering, opacity tests and micro mesh generation (more about these on the Architecture page of this review). Last, but certainly not least is DLSS 3 Frame Generation, which introduces a completely new way of increasing FPS. With Frame Generation, the GPU will automagically generate an additional frame for each frame rendered, based on the movement in each frame—doubling FPS in the process.

For a majority of gamers, the "classic" raster performance is very important though—highest settings, RT off, DLSS off—so we made sure to extensively test this scenario using 25 games at three resolutions. The GeForce RTX 4090 achieves incredible performance results here: +45% vs RTX 3090 Ti. Yup, 45% faster than last generation's flagship—this is probably the largest jump Gen-over-Gen for many years. Compared to RTX 3080 the uplift is 89%—wow!—almost twice as fast. Compared to AMD's flagship, the Radeon RX 6950 XT, the RTX 4090 is 64% faster. Somehow I feel that after RDNA 2, Jensen said to his people "go all out, I want to conclusively beat AMD next time."

All this testing is done at 4K resolution, and that's the only resolution that really makes sense for the RTX 4090. Maybe 1440p, if you want to drive a 144+ Hz monitor at max FPS, but you'll end up a bit CPU-limited in many titles. Interestingly, when CPU limited at 1080p, the RTX 4090 is clearly behind Ampere cards in several games. It seems the new architecture has a somewhat higher CPU overhead, which further drags down the maximum FPS the CPU can achieve. This is more of an interesting curiosity though, not a real issue.

Where RTX 4090 can flex its muscle is with ray tracing enabled. While previously enabling RT at 4K always meant some compromises—either upscaling or reduced settings—the RTX 4090 will give you 60 FPS with RT active in nearly all titles. Taking a closer look at our ray tracing benchmarks we can see that the performance hit from enabling ray tracing is considerably lower than before, thanks to the various technological improvements. Compared to AMD, the ray tracing performance is often 3x as high—AMD has to innovate here with their next-gen, or they'll fall behind too much and NVIDIA will win ray tracing.

Many years ago, the first NVIDIA Founders Edition cards were using conservative cooling, and ran hotter and louder than competing custom designs. Over the generations, NVIDIA focused more and more on improving their FE, and Ada is no exception. Visually, the RTX 4090 is one of the most beautiful graphics cards ever created, no doubt, "sexy" to us tech nerds. I love how NVIDIA uses metal surfaces of various colors and finish to bring them together as an amazing-looking industrial design. It's as if the designers wanted the product to look "exalted," like an iPhone. Under the hood we find a large vapor-chamber cooling solution with six heatpipes and a VRM that's among the best I've ever seen on a graphics card. The 20+3 phase design is clearly fit to run at 600 W without breaking a sweat. NVIDIA didn't just pack a lot of meat into their cooler, they also paired it with excellent fan settings. The card runs at a cool 66°C and emits only 35 dBA at full load. Especially taking into account the performance class, this is "quiet." The card is definitely audible though, and you need good ventilation in your case to exhaust all of the generated heat. Tomorrow we'll be allowed to publish multiple custom design reviews—these will run with even less noise.

In this review we've tested NVIDIA's new DLSS 3 frame-generation capability, and I have to say I'm impressed. At first I was highly sceptical and thought it would be like the soap opera interpolation effect on TVs, but no, it works REALLY well. The algorithm takes two frames, measures how things have moved in those two frames and calculates an intermediate frame in which these things moved only half the distance. While this approach is definitely not problem-free, especially when pixel-peeping at stills or slowed down video, in real-time it's nearly impossible to notice any difference. As you run at higher FPS and resolution it becomes even more difficult because the deltas between each frame are getting smaller and smaller. I also feel like we're only seeing the beginning of this technology, and there will be numerous improvements in the future. Adoption rates should be good, because implementing DLSS 3 frame generation is very easy if you already have DLSS 2 support in your game. Another interesting NVIDIA Tech is "Reflex", which reduces the total gaming latency, so you see things earlier on your screen and can react faster, to get more kills or survive for longer.

Thanks to the move from 8 nm Samsung to 5 nm TSMC, efficiency has improved drastically. Energy efficiency is almost doubled compared to cards like RTX 3090 / 3090 Ti, 50% better than AMD Radeon RX 6950 XT. In typical gaming with ray tracing disabled, we measured around 350 W power consumption—much less than the 450 W default board power limit. This is a big change compared to Ampere, which always ran in its power limit. On Ada there's power headroom to spare, and this is part of the reason why NVIDIA could improve RT performance so much. On Ampere when you started putting an RT load on the GPU, the power consumed by the RT cores would be taken away from the shader cores, to ensure the card stayed within its whole power budget. On Ada there's so much power headroom that you can just light up all the RT cores and the shaders can continue to chug along happily at full speed. With RT cores running at maximum load we measured slightly above 450 W power consumption, so the 450 W default limit is spot on. Enabling DLSS will reduce power draw, because of the lower internal resolution.

For enthusiasts and overclockers, NVIDIA allows manual power limit adjustments up to 600 W. This is huge—previously the company was fairly conservative with this range, limiting it to +50 W or less on most cards. There are no fundamental changes to overclocking, it works just like before. With manual OC we were able to break 3 GHz actual clock frequency—very impressive. The performance gains were somewhat limited though at 4.5%, when compared to a frequency increase of 12%.

NVIDIA is pricing the GeForce RTX 4090 Founders Edition at $1600, some of the custom designs from various board partners will match this price point, most will go higher. $1600 really is a ton of money for a graphics card, especially in these times where money is tight. The fact that the mining boom is over will also ensure that these cards won't fly off shelves. Then there's also quite a bit of GeForce 30 inventory in the market, which is probably the reason why NVIDIA is launching their highest-end card first. If I was NVIDIA, with an obligation to make money, I would have picked that price point too. It's competitive enough to beat the price/performance of the RTX 3090 Ti and roughly matches the $950 RTX 3090 and Radeon RX 6950 XT. Of course, the new card does offer additional features like DLSS 3 and better efficiency, which could tempt a lot of people to spend that much on their next GPU instead of going with a 3090/6950 XT. From a buyer's perspective $1600 is probably also too much psychologically. We've seen RTX 2080 Ti launch at $1200, back then people started realizing "above $1k" pricing was possible, then we had the mining crisis where people would buy overpriced cards just to play something, and now NVIDIA is hoping that this will continue. I'm having doubts. In the communities there's overwhelming disagreement with the $1600 price point, and I can't blame people who are trying to pay their gas and power bills first. Still, RTX 4090 is an amazing new product that's almost flawless, and it will definitely be a hit with the people who can afford it. Everybody else? Hold out a bit longer for the RTX 4080, and especially AMD's new Radeon RX 7000 offerings.