I plugged an Nvidia RTX 5090 into a gaming handheld

Two weeks ago, I showed you how the world’s fastest graphics card works in a small form factor PC. To my surprise, Nvidia’s RTX 5090 Founders Edition delivered the vast majority of its performance even in a 12.7-liter desktop with a five-year-old CPU.

It made me wonder: what if I plugged this card into a handheld gaming PC instead? So I did, and let me tell you: it’s a wonder to behold. It’s enough to make me believe in a rich future where handhelds get more powerful when you dock them at home.

I started with the same $1,999 RTX 5090 FE and 1000-watt power supply from my desktop test, dropping them both onto a $99 Minisforum DEG1. It’s an open-air external GPU that can connect to the Oculink port that’s now shipping in a handful of portable gaming PCs, so long as you bring your own desktop GPU and power supply.

I plugged that Oculink cable into a $1,000 GPD Win Max 2 handheld. And then, with just an AMD Ryzen 8840U mobile CPU and four lanes of PCIe 4.0 bandwidth, rather than the 16 lanes of PCIe 5.0 that Nvidia’s GPU technically supports, my new Franken-desktop spit fire anyhow. I’m talking over 100 frames per second in Cyberpunk 2077 at 4K resolution and Ultra settings. I’m talking about playable frame rates in all the most intensive games at 4K and near-maximum specs.

Across our eight test games, it did run between 7 percent and 47 percent slower than when my colleague Tom Warren paired a 5090 with the fastest gaming CPU money can buy. But it was only 4 to 29 percent slower than that 5090 in my SFF desktop and far faster than when I stuck an RTX 3080 in that desktop. And it was four to 12 times faster than the handheld’s Radeon 780M integrated GPU could manage on its own.

Start with the “5090 handheld” column in my table below:

For example, my GPD Win Max 2 review unit averages just 17fps in Call of Duty: Black Ops 6 at Extreme spec at the handheld’s native 2560 x 1600 screen resolution — and if I set the game to the lowest, most potato graphics possible at a pixelated 640 x 480 resolution, the very best I can get is 113fps. Plug in the RTX 5090, and I get 131fps with the graphics at their highest settings. From potato to ultra with one plug — and a mortgage payment.

Not that you’d want to play games that potent on a handheld’s small internal screen, of course — and I saw the best results plugging in an external monitor, too, because you lose less performance shuffling data back and forth across that Oculink cable. My Call of Duty frame rate went from 131fps on the handheld’s 1600p internal screen to 134fps on an external 4K monitor. Of our test games, only Black Myth: Wukong fell below 60 frames per second at native 4K.

I don’t want to oversell Oculink GPUs too much because they’re not entirely ready for primetime, and they come with some major caveats compared to Thunderbolt and USB-C tech.

First up, Oculink cables are not hot-swappable like USB products. It’s so tempting to imagine playing a game handheld, then plugging it in and resuming with amazing graphics on a big screen, but that’s not the reality. You have to shut down your PC every time you connect and disconnect or risk damage — the first time I tried an Oculink system a year ago, I apparently damaged its port for good. And before you power on your PC, you have to power on the eGPU first. (The Minisforum DEG1 I’m using does attempt to sync with a connected PC and power on and off at the same time, but it hasn’t been foolproof for me; the company says it’s designed to sync with its own mini-PCs.)

Second, current Oculink cables themselves aren’t as robustly designed as many USB products. While they’re shaped like a smaller DisplayPort, with the same kind of locking connector, they seem far easier to accidentally dislodge, bend, or break.

Third, Oculink is not a single-cable docking solution like Thunderbolt or USB4 because it doesn’t provide USB data, audio, or power to a connected laptop: I had to run a separate USB-C power cord to the handhelds I tested. That’s a feature, not a bug, so there’s nothing to get in the way of your GPU bandwidth.

Fourth, there simply aren’t all that many devices with Oculink. GPD is one devotee, and I tested it working with an early sample of the Ayaneo 3, but the companies you’ll find at your local Best Buy — Asus, Lenovo — are currently putting USB4 ports into their handhelds instead of Oculink.

And fifth, Oculink is currently such an afterthought for companies like Nvidia that their drivers have apparently been bugged for years. When I first plugged the 5090 into the GPD Win Max 2 and the Ayaneo 3, they refused to tap into its power at all until I installed a community-developed “error43” patch that worked like magic. Even so, I saw a couple of crashes that I couldn’t pin down.

Today, prospective eGPU buyers have to pick between the reduced bandwidth and compatibility pains of Thunderbolt 4 and USB 4 or the cleaner but less-user-friendly Oculink, but hopefully best-of-both-worlds options won’t take too long to materialize. Asus has already announced the first Thunderbolt 5 eGPU, even if the company doesn’t currently have any smaller, weaker PCs with Thunderbolt 5 that could actually take advantage of it, and perhaps some future 80Gbps USB4 devices could offer the same without needing Intel’s Thunderbolt certification.

Maybe then, I won’t need a desktop at all. I’ll have a handheld gaming PC, and I’ll just plug it into my eGPU when I get home. Maybe this could even be the future of PC gaming for the masses — a future for those who don’t want towering desktops — if the industry can just figure out which cable to use.