Thunderbolt 5 fixes many of the issues that held back last-gen Thunderbolt eGPUs. The new interface offers better stability and, most importantly, a higher bandwidth rating, with the bidirectional rating much higher than PCIe 4.0x4 OCuLink. However, bandwidth figures alone don't tell the whole story.

Even with all the improvements that Thunderbolt 5 brings to the table, for performance-focused eGPU setups, there are some trade-offs that make OCuLink a better choice.

Direct PCIe matters more than theoretical bandwidth

Bandwidth numbers are impressive, but direct PCIe is what matters for eGPUs

Thunderbolt 5 may look unbeatable when raw bandwidth numbers are considered, especially compared to previous generations' throughput. However, these figures don't automatically translate into better eGPU performance.

The thing is, the main advantage of OCuLink isn't apparent from its spec sheet. For reference, PCIe 4.0x4 OCuLink has a rated bandwidth of 64Gbps, while Thunderbolt 5 has 80Gbps bi-directional bandwidth. While this is a notable difference, OCuLink behaves like a direct PCIe connection, unlike Thunderbolt.

So, when the eGPU is connected via OCuLink, the GPU communicates with the system over dedicated PCIe lanes, without any additional layers between them. This keeps data transfer stable and latency low, and for gaming workloads, this matters more than peak bandwidth potential.

Despite numerous improvements, Thunderbolt 5 relies on a controller to reroute PCIe traffic. This extra step introduces overhead, and when the GPU is pushed hard, the performance difference becomes noticeable.

Ray tracing performance exposes Thunderbolt 5's limits

A more PCIe-like connection still has the advantage in demanding workloads

One of the improvements brought by Thunderbolt 5 becomes very evident when considering the ray tracing performance of eGPUs. While last-gen options struggled, ray-traced gaming is far more stable with Thunderbolt 5 eGPU setups, but OCuLink still has an edge.

The thing is, for eGPU setups, ray tracing is one of the most demanding workloads. In ray-traced games, there's much higher data exchange between the CPU and GPU than in rasterized games, and in such cases, a more direct, PCIe-like link matters. So, while the eGPU performance with Thunderbolt 5 is much closer than before, in ray-traced titles, there's less consistent frame delivery and lower averages than with an OCuLink setup.

Of course, OCuLink doesn't entirely eliminate the performance penalty of running an external GPU, but it can reduce it more effectively. In some cases, ray-traced performance remains relatively close to that of a desktop configuration, reinforcing OCuLink's superiority when performance is the priority.

High-end GPUs scale better over OCuLink

More powerful GPUs benefit from fewer interface bottlenecks

The performance gap between OCuLink and Thunderbolt 5 eGPUs becomes more apparent as GPU power increases. When compared with mid-range graphics cards, the difference can be modest enough to overlook. However, when moved into upper-mid-range or high-end GPUs, the tunneled interface becomes harder to ignore.

Powerful GPUs require sustained, low-latency communication with the CPU to remain fully utilized. The PCIe-based connection of OCuLink allows these graphics cards to scale more predictably. More specifically, as GPU performance increases, OCuLink tends to lose less performance than PCIe-based eGPU setups.

There's no denying that Thunderbolt 5 significantly narrows the gap compared to previous generations. However, it doesn't eliminate the scaling ceiling entirely, and on high-end GPUs, the performance gains with this interface can start to flatten out sooner. That is, with OCuLink, the eGPU setup can better utilize the higher-end hardware you're paying for.

Convenience is Thunderbolt 5's biggest strength

And also its biggest compromise for eGPU performance

The ease of use is Thunderbolt 5's biggest advantage. Unlike OCuLink, you don't need to turn the whole system off to pair it with a Thunderbolt 5 eGPU, making it feel more like a docking station than an extension hack. In addition to flexibility, the interface's design prioritizes broad compatibility, and the same interface can work for displays, networking, and storage expansion. The same philosophy carries over to the newer USB4 v2 implementation as well.

OCuLink, in comparison, sits at the opposite end of the spectrum. It's not hot-swappable, and there aren't many accessories that can use it. What it offers instead is consistency and the ability to build a setup that's closer to a direct PCIe connection. This advantage, over the convenience and compatibility trade-off, makes OCuLink the right fit for an eGPU build where performance is the priority.