eGPU vs laptop GPU: why Thunderbolt limits external graphics
Engadget explains why external GPUs lose some of a desktop card's power over Thunderbolt, and when an eGPU still makes sense for laptop owners.

Key takeaways
- eGPUs connect via Thunderbolt or USB4, which limits PCIe bandwidth compared with a laptop's internal GPU link.
- Thunderbolt 5's 120 Gbps boost mode does not raise PCIe throughput beyond 64 Gbps, according to Intel.
- The desktop RTX 5090 has roughly double the CUDA cores and memory bandwidth of the laptop version before any cable penalty.
- Enclosures such as Razer's Thunderbolt 5 Core X V2 cost about $350 and exclude the GPU and power supply.
In this article
The appeal and the catch
An external GPU promises the best of both worlds: a thin laptop for travel, and desktop-class graphics at home. According to Engadget, the idea works, but the desktop card's advantage shrinks because its data must travel through Thunderbolt or USB4 rather than a direct internal connection.
Even cards sharing a name with desktop models differ greatly. The RTX 5090 Laptop GPU has 10,496 CUDA cores, 24GB of GDDR7 memory and a power range of 95W to 150W, while the desktop RTX 5090 offers 21,760 CUDA cores, 32GB of GDDR7 and a 575W rating.
Thunderbolt remains the bottleneck

A modern eGPU usually connects through Thunderbolt or USB4 instead of the laptop's internal PCIe bus. Thunderbolt 4 provides 40 Gbps overall and requires 32 Gbps of PCIe throughput. Thunderbolt 5 raises this to 80 Gbps with 64 Gbps of PCIe data. Internal laptop GPUs can use wider multi-lane PCIe links, so they avoid that constraint.
Thunderbolt 5 also has a 120 Gbps Bandwidth Boost mode, but Intel says PCIe throughput stays at 64 Gbps, so the extra headroom does not reach the GPU. USB4 can carry PCIe traffic too, yet a USB-C port alone does not guarantee eGPU support; the laptop, enclosure, drivers and power supply must all be compatible.
Using the laptop's own screen adds a further penalty, because rendered frames may travel back through the same limited cable. Engadget suggests connecting a monitor directly to the eGPU to reduce this loss. There is no universal figure for how much performance is lost, since the GPU, laptop, connection, resolution and game all matter.
Why built-in GPUs are slower but simpler
A laptop GPU starts with less raw hardware, but it avoids most of this plumbing. It works within the machine's cooling and power limits, never sends frames over an external cable, and stays inside the laptop when you leave the desk.
Portability is the weak point of a desktop-card eGPU. The setup needs an enclosure, a graphics card and often an ATX power supply on the desk, and a large card can add size and noise.
What it costs and when it makes sense
Cost is a major factor. Engadget cites Razer's Thunderbolt 5 Core X V2, which takes full-size NVIDIA and AMD cards up to four slots wide and has its own 120mm fan. The enclosure costs roughly $350 but includes neither a GPU nor the required power supply, so those are extra expenses.
The maths improves if a suitable card is spare after a desktop upgrade. Buying a high-end desktop GPU specifically for an eGPU is harder to justify, because you pay for power that a cable may not let you fully use.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



