Summary

• eGPUs saw a decline due to bulky connections, but the rise of PC gaming handhelds has sparked renewed interest in compact external GPUs.
• Limited bandwidth of Thunderbolt connections impacted eGPU performance compared to built-in options, but new technologies like OCuLink are changing that.
• With faster options like Thunderbolt 5 and OCuLink-2 coming soon, the future of eGPUs looks promising despite initial setbacks.

The history of external GPUs, or eGPUs, is a convoluted one. Around six years ago, they were all the rage, with Thunderbolt 3 enabling an easy plug-and play solution for this kind of thing. But the market slowed down very significantly after a couple of years, presumably due to a lack of interest in the idea of having a giant box you need to connect to your laptop.

But in the past year or so, eGPUs have had a bit of a resurgence, and it's thanks in big part to PC gaming handhelds. These tiny devices are great for playing games in the palm of your hand, but they lack significant power for more demanding games, so an external GPU can turn these handhelds into bona fide gaming PCs.

The early excitement for external GPUs

And their quick fall from grace

It's hard to say when people first started dreaming about hooking up a desktop graphics card to a laptop, but eGPUs really became a thing in 2013 with the introduction of Intel's Thunderbolt 2. Older data interfaces like the original Thunderbolt and ExpressCard were technologically capable of driving eGPUs but ultimately didn't spark the rise of eGPUs. What Thunderbolt 2 introduced was simply higher data transfer speeds, which eGPUs need for good performance. These first Thunderbolt 2 eGPU chasses were really expensive and relatively primitive, however, and didn't exactly become mainstream.

eGPUs were finally brought into the limelight with the launch of Thunderbolt 3 in 2017 and Thunderbolt 3-powered enclosures like the Razer Core. These enclosures were higher quality than old Thunderbolt 2 models, had double the bandwidth, cost significantly less, and also came out around the same time Nvidia's legendary GTX 10 series did. It was a really excellent time for eGPUs to finally become accessible to the masses, especially since they seemed to fulfill the promise of bringing desktop-level graphics to the masses, which is what everyone wanted.

From 2016 to around 2020, new eGPU products came out like clockwork, and lots of companies tried their hands at the technology: Razer, PowerColor, Blackmagic Design, and Asus, just to name a few. It was a very vibrant and competitive field for years and it seemed that it would stay that way for a while thanks to the launch of Thunderbolt 4 and USB4, which was compatible with Thunderbolt 3. After 2020 though, the flood of new eGPU enclosures and accessories became a drought and it wasn't really getting any better.

eGPUs couldn't deliver on the hype

There were big sacrifices with an eGPU

Conceptually, eGPUs have always been interesting, but at the same time, but they've also been severely hindered by the technology they rely on. Desktop GPUs are very powerful pieces of hardware, but a big component in making them as useful as they are is the high bandwidth at their disposal. Modern GPUs use a PCIe 4.0 x16 connection, which means they have about 31.5GB/s (that's gigabytes, not gigabits), of bandwidth. That's a lot of data that can be quickly transfered from the GPU to the rest of the system.

However, a Thunderbolt 4 connection (much like Thunderbolt 3) only offers 40Gbps (gigabits), which equates to around 5GB/s. And even then, only 32Gbps of that are for the PCIe connection, meaning at most, you get 4GB/s of bandwidth for your GPU. That's eight times less bandwidth than you'd get from an internally-mounted GPU, and that has a huge impact on performance, potentially introducing latency and stutters.

This made it so that, even if you had a laptop, you would probably get more performance out of a built-in laptop GPU than using an external desktop GPU, plus you can't really take the external GPU with you everywhere the same way a laptop can travel with. Even a gaming laptop is easier to carry than a big box.

Because of that, and GPUs getting increasingly more powerful while Thunderbolt remained stagnant, external GPUs just weren't that relevant anymore.

Gaming handhelds changed things

There's a new need for increased power

This was all well and true, but in 2022, things started to change. Valve introduced the Steam Deck, playing off of the huge success of Nintendo's Switch, and responding to requests for additional performance compared to that handheld. This spurred a series of PC gaming handhelds flooding the market. Some long-time players in the market like One Xplayer and GPD kept releasing new hardware, but the introduction of the Asus ROG Ally in 2023 really solidified the PC gaming handheld as a market, and since then, we've seen other hardware like the Lenovo Legion Go and MSI Claw.

The thing about these devices is that they're even more compact than a laptop, portability is more of a factor than ever. Not only do these handhelds have low-power processors, the idea of a discrete GPU inside the hardware just isn't feasible. And while these devices can play games pretty well on a small screen, they run at fairly low settings, so they can look rough when blown up on a large monitor. Thus, the need for additional power when playing at a desk started to increase again.

Asus was the first to really capitalize on this by including the ROG Xg Mobile connector on the ROG Ally. This proprietary connector allows you to use the ROG Xg Mobile series of GPUs with the company's handheld PC, and that connector also offers far more bandwidth, double what's available via Thunderbolt 4. And since then, we've seen other companies release their own external GPUs, with One Xplayer building the OneXGPU, and GPD introducing the GPD G1. Both of these also support Thunderbolt/USB4, but they also add OCuLink, a non-proprietary connector that offers similar benefits to Asus' ROG Xg Mobile.

However, things are a bit different now. Most of these eGPU solutions come as an all-in-one package, meaning the GPU and the enclosure are one and the same. That means you're not able to upgrade the GPU without buying a whole new unit. At the same time, these units are significantly more portable, too, mostly because they're based on laptop GPUs rather than desktop ones. That may seem like a downgrade, but considering that GPUs will still be limited by the bandwidth of the connection, it wouldn't make sense to invest in a higher-end card when you can't really use its performance adequately.

External GPUs are getting faster

The future of eGPUs is promising

eGPUs are making a bit of a comeback, though it's still early to say whether this will be a long-lasting success or not. That being said, things are looking up in some ways. The rise of OCuLink GPUs shows that there's interest in an external GPU that makes fewer compromises, and with double the bandwidth of Thunderbolt 4, it's a big step up.

Not only that, but Thunderbolt 5 is starting to hit the market, and that too doubles the bandwidth from Thunderbolt 4, thanks to the transition to PCIe 4.0. What's more, OCuLink-2 is also based on PCIe 4.0 and it should offer double the bandwidth of Thunderbolt 5 when it starts showing up in more products, going up to 16GB/s. This will significantly lessen the bandwidth constraints of external GPUs and make them more viable than ever.

It's up to the market now to show whether there's enough interest in this concept for more GPUs to come out and actually leverage this technology. I, for one, would be pretty excited to see it.