Nvidia GeForce RTX 5090 Mobile Review

The Nvidia GeForce RTX 5090 launched back in January and easily became the most powerful desktop graphics card on the market. Now, a few months later, Nvidia has launched its Blackwell architecture for laptops, and while these mobile Nvidia GPUs are nowhere near as powerful as their desktop counterparts, they do deliver an incredible gaming experience, especially on high-end gaming laptops with high-refresh displays.

However, just like Nvidia’s latest line of desktop graphics cards, the generational improvement in pure performance isn’t nearly as pronounced as it’s been in previous generations. That’s offset a bit in games that support features like Multi-Frame Generation, of course, but it does mean that anyone who already has a decent gaming laptop doesn’t need to sweat upgrading their setup for this new generation.

Specs and Features

While the GPU in the Razer Blade 16 I tested is called an Nvidia GeForce RTX 5090, it’s a far cry from the desktop version of the GPU with the same name. Just by virtue of space and thermal constraints, the mobile version of the RTX 5090 needs to be much smaller: Instead of the 21,760 CUDA cores in the desktop version of the card, the mobile 5090 is limited to 10,496 cores across 82 Streaming Multiprocessors. That’s less than half the silicon, and places the RTX 5090 mobile closer to the desktop RTX 5080.

But it’s more than just the size of the chip – laptops are a lot more power-constrained than a desktop, too, by virtue of needing to operate on a battery or via a portable power brick. The system I tested the RTX 5090 Mobile with had the GPU limited to 160W, which is a little over a quarter of the power of its desktop counterpart. That has a huge impact on gaming performance, since a general rule of thumb is that more power = better performance.

However, compared to the previous-generation Razer Blade 16 with the Nvidia GeForce RTX 4090, the new mobile chip has been beefed up a bit. The RTX 4090 packed 9,728 CUDA cores across 76 Streaming Multiprocessors, at the same 160W power budget. That’s about a 7.4% increase in the amount of cores, which is a minor increase, but it lines up extremely well with the performance increase you can expect from the RTX 5090.

Keep in mind, though, that with laptop graphics, it’s largely up to the laptop manufacturer how much power and cooling a GPU has access to. There may be some gaming laptops with an RTX 5090 configured to run with as little as 95W. The laptop I tested had it configured for 160W, which is above the maximum listed TGP (total graphics power) specified by Nvidia. The point is, your mileage may vary, as a 95W RTX 5090 is going to be much weaker than one running on 150W.

Each Streaming Multiprocessor is architected a bit differently than the RTX 4090 too. Just like the desktop version, the RTX 5090 Mobile is based on the new Blackwell architecture, which places special emphasis on AI computation – after all this is the same architecture powering most cutting-edge supercomputers right now. The new Tensor Cores, combined with a new AI Management Processor, or AMP, allow the GPU to do much more complex AI workloads, and is the driving force behind multi-frame generation.

Nvidia has also made an effort to improve battery life, which has never been a strong point for gaming laptops. It’s done this by improving Nvidia Optimus, a technology that dynamically switches between discrete and integrated graphics depending on what you’re doing. But Team Green has also implemented something called “Battery Boost”, which should improve gaming performance when you’re not plugged into the wall. Basically, Nvidia is using an algorithm to detect what’s going on in the game you’re playing and adjusting performance accordingly. For example, it’ll lower GPU clocks when you’re in a dialogue scene, and push more performance when you actually need it.

Performance

Just like its desktop counterpart, the Nvidia GeForce RTX 5090 Mobile doesn’t have the strongest generational improvement over its predecessor. On average, I found that the RTX 5090 was about 8% faster than the RTX 4090. The performance gap was a bit better in workloads like 3DMark, but for the most part it doesn’t deliver much better gaming performance, at least before Frame Generation is brought into the equation.

To be clear, testing laptop graphics performance is a bit more complicated than testing a desktop graphics card, especially across generations. The Razer Blade 16 that Nvidia sent me for this architecture review is sporting an AMD Ryzen 9 AI HX 370, while the previous-gen Razer Blade 16 I used for testing the RTX 4090 has an Intel Core i9-13950HX – two very different processors. Plus, the chassis design has changed completely, and while we have a full review on the laptop coming later, the new design is much thinner and lighter, which calls into question thermal constraints.

However, throughout my testing, I didn’t notice the RTX 5090 being thermal- or power-limited in any way, even if the RTX 4090 was able to stretch a little above the 160W power limit – though not by much. So, while my initial results point to the RTX 5090 being on average 8% faster than the RTX 4090, there is enough variability here to take the numbers with a bit of a grain of salt. We’ll have a clearer picture of what this GPU can do as more laptop designs head to market over the next year or so.

In 3DMark, the new RTX 5090 Mobile was 12% faster than the RTX 4090 in Speed Way, and 16% faster in Steel Nomad. These are both DirectX 12 tests, and indicate potential performance more than anything.

When it comes to actual games, the performance differences are much less pronounced. For instance, in Call of Duty: Black Ops 6 at 1600p, the RTX 5090 Mobile system got 107 fps, compared to 106 fps from the RTX 4090. That’s just a 1% improvement and totally within the margin of error.

Things look a little better in Cyberpunk 2077, where the RTX 5090 Mobile manages 67 fps to the 4090’s 62 fps, making for an 8% improvement. That’s decent, and it’s important to note that this is without Frame Generation, which will further increase the 5090’s lead in frame rate.

I tested Metro Exodus: Enhanced Edition with no upscaling, to give an idea of what to expect from pure ray tracing performance. In this game, the RTX 5090 gets 61 fps at 1600p with the Extreme Preset, compared to 56 fps from the RTX 4090, that’s a 9% improvement, lining up with the increase in CUDA cores almost exactly.

Red Dead Redemption 2 is a weird case, as it puts the RTX 5090 behind the RTX 4090 at 1600p with DLSS set to Balanced. The newer GPU gets 89 fps, compared to 93 from the RTX 4090, which is about a 4% drop. However, this is likely due to the CPU not being able to keep up.

Total War: Warhammer 3 is a good way to gauge pure rasterization performance, as it eschews any ray tracing or upscaling technologies. In this game, the RTX 5090 is about 6% faster than the 4090. Though this may be another case where the CPU is holding the 5090 back.

In Assassins Creed Mirage, the GeForce RTX 5090 is able to soar with 107 fps at 1600p with everything maxed and with DLSS set to Balanced. However, that’s only 4% faster than the RTX 4090 with the same settings.

Then, in Black Myth: Wukong, a game that’s extremely GPU-heavy, the GeForce RTX 5090 gets 63 fps on average, which is only 5% faster than the 4090’s 60 fps average. This is one of the games that showed the strongest uplift for the RTX 5090 on desktop, so it's especially damning here.

Even for creators, the Nvidia GeForce RTX 5090 for laptops isn’t a huge improvement. In Blender, the RTX 5090 Mobile is only 9% faster than the RTX 4090 in the Monster benchmark. Blender performance usually gets better over time, but it’s something to keep in mind.

The main reason the desktop version of the RTX 5090 had a large lead over its predecessor was because it had a much higher power budget – 575W vs 450W. With the RTX 5090 and RTX 4090 having the same 160W power limit, we see similar results to other RTX 5000 graphics cards. Chips like the RTX 5070 Ti and RTX 5080 had very modest improvements over their predecessors, which also had similar power requirements. At least when it comes to traditional gaming performance, Blackwell seems to be an architecture that is extremely dependent on power, so it shouldn’t be surprising that laptop performance doesn’t scale as well as Nvidia’s desktop cards.

DLSS 4 For Laptops

The biggest difference between the RTX 5090 and its predecessor is DLSS 4, and specifically multi-frame generation, or MFG. DLSS 4 does also include improvements that are available on previous-generation GPUs, like moving AI upscaling from a convolutional neural network (CNN) to a Transformer network. This greatly improves accuracy and image quality, whether you’re using the RTX 5090 or an older RTX graphics card.

However, multi-frame generation really is the star of the show, particularly on the Razer Blade 16 with its 240Hz display.

Just like the previous version of frame generation that debuted with DLSS 3, MFG uses the Tensor Cores to generate frames between natively rendered frames. It then uses the new AI Management Processor to pace out the frames to minimize latency and other weirdness. Frame pacing is a task that has traditionally been handled by the CPU, so the AMP is especially needed to cut down on system latency when the GPU is magically creating three frames in between each rendered one.

For a high refresh display, though, this is an awesome feature. It doesn’t improve latency or anything, but it does make the gameplay look a lot smoother, which is really all you need in single-player games like Assassins Creed Shadows. Just don’t expect it to magically improve your response times if you enable it in an online shooter like Marvel Rivals or Fragpunk.

Nvidia claims that MFG can triple frame rates, but it’s important to keep in mind that there’s a catch: you need to already have a good frame rate to get a good experience from frame generation. I wouldn’t recommend any kind of frame generation unless you’re already getting 50-60 fps without it. Below that, you’re basically asking for weird artifacts to show up as the algorithm struggles to track objects between frames.

If you do already have a good frame rate though, MFG really does boost frame rates by a huge margin, even if it does come with a slight cost to latency. For instance, in Cyberpunk 2077 with the Ray Tracing Overdrive preset and with DLSS set to balanced with no frame generation, the Razer Blade 16 managed an average of 51 fps at 1600p. That came with latency around 47ms, which isn’t great, but still more than playable. When I turned frame generation to 2x, that frame rate jumps all the way up to 96 fps, with the latency going up to 53ms. That’s a huge improvement to framerate, and the latency cost really isn’t that high. The latency cost is a bit greater with MFG, as turning it up to 4x increases average latency to 59ms – though the framerate jumps to 170 fps.

It’s a similar story in Star Wars Outlaws. At 1600p with DLSS set to balanced, the RTX 5090 Mobile gets around 61 fps on average, with an average latency around 36ms. With 2x frame generation the frame rate goes up to 115 fps, with latency also increasing to 45ms. The jump up to 4x frame generation only brings the latency up to 48ms, but bumps the frame rate up to 210 fps, which is getting extremely close to the laptop’s native refresh rate of 240 Hz.

Realistically, the Nvidia GeForce RTX 5090 is going to primarily be found in high-end gaming laptops with high resolution displays with high refresh rates. That’s exactly where multi frame generation shines, and it’s awesome to finally be able to take advantage of these super-fast laptop screens. Still, if you’re someone who is sensitive to latency or if you play competitive games, it might not be worth enabling.

Jackie Thomas is the Hardware and Buying Guides Editor at IGN and the PC components queen. You can follow her @Jackiecobra