An external GPU (eGPU) is a discrete graphics card housed outside the computer in an enclosure with its own power supply and connected through an external PCI Express link, usually Thunderbolt, USB4 or OCuLink. It gives a laptop, mini PC or handheld device desktop-class graphics performance that its integrated GPU or small mobile GPU cannot provide.
In graphics hardware, the eGPU is a compromise between portability and power. It delivers most of the card’s capability, but the narrow external connection costs some performance compared with the same card installed in a desktop PCIe slot.
How an external GPU connects
Inside the enclosure, the graphics card sits in a normal PCIe x16 slot, but only a few lanes are actually connected. A controller tunnels PCIe traffic over the external cable, and the host operating system sees the card as if it were attached internally. The enclosure power supply, typically 350 to 750 W, feeds the card, and many enclosures also charge the laptop and provide USB and Ethernet ports.
The link width is the decisive parameter. A desktop slot offers PCIe x16, while external links usually provide the bandwidth of PCIe x4, and Thunderbolt 3 controllers often deliver even less usable throughput because the link is shared with other protocols.
| Interface | Nominal link speed | PCIe tunneled | Notes |
|---|---|---|---|
| Thunderbolt 3 | 40 Gbps | PCIe 3.0 x4, about 22 Gbps usable | Hot-plug, widest support |
| Thunderbolt 4 and USB4 | 40 Gbps | PCIe 3.0 x4 class | Standardized, eGPU support optional on USB4 |
| Thunderbolt 5 | 80 Gbps | PCIe 4.0 x4 | Roughly double the PCIe bandwidth |
| OCuLink | Up to 64 Gbps | PCIe 4.0 x4 directly | No tunneling overhead, no hot-plug, no power or USB |
| Desktop slot for reference | Up to 512 Gbps | PCIe 4.0 or 5.0 x16 | Full bandwidth |
Performance loss and its causes
An external GPU typically delivers about 70 to 90 percent of the same card’s desktop performance, but the gap varies widely with the game, resolution and link. Three factors explain most of the loss.
- Limited bandwidth: textures, geometry and command buffers must cross a link four times narrower than a desktop slot.
- Added latency: tunneling controllers add delay to every transaction, which hurts high-frame-rate workloads.
- Internal display loopback: when the image is shown on the laptop’s built-in screen, each finished frame travels back across the same link.
The loss is usually smallest at high resolutions such as 4K, where the GPU itself is the bottleneck, and largest at 1080p with high frame rates, where many small transfers saturate the link. Compute workloads that load a model into VRAM once and then run for hours, such as AI inference or 3D rendering, often lose very little.
| Factor | Better case | Worse case |
|---|---|---|
| Display used | External monitor on the eGPU | Laptop internal screen |
| Resolution | 1440p or 4K | 1080p at high frame rates |
| Workload type | Long compute jobs, rendering | Fast-paced gaming, streaming assets |
| Card VRAM | Large enough for the whole scene | Frequent spills to system RAM |
| Host CPU | Recent multi-core CPU | Low-power CPU that limits frame rate |
External GPU compared with other graphics options
Choosing an eGPU makes sense only when compared with the realistic alternatives: an integrated GPU, a gaming laptop with a mobile discrete GPU, or a separate desktop computer.
| Option | Performance | Portability | Main drawback |
|---|---|---|---|
| Integrated GPU | Low to moderate | Excellent | Shares memory and power with CPU |
| Laptop discrete GPU | Moderate to high | Good | Power and thermal limits, fixed hardware |
| External GPU | High at home | Laptop stays light | Cost of enclosure, bandwidth loss |
| Desktop PC | Highest per cost | None | A second machine to maintain |
An eGPU is attractive mainly for users who want one thin laptop for travel and strong graphics at a desk. Because the card can be replaced, the setup also allows upgrades that a laptop’s soldered GPU never permits.
Practical requirements and setup
Before buying, confirm that the host explicitly supports eGPUs. A USB-C port alone is not enough; it must carry Thunderbolt or USB4 with PCIe tunneling, and the firmware must allow it. Apple silicon Macs do not support eGPUs, although older Intel-based Macs did.
- Check the port: Thunderbolt 3, 4 or 5, USB4 with PCIe tunneling, or an OCuLink port.
- Match the enclosure to the card length, slot width and power draw, including transient spikes.
- Use a short certified cable, since passive cables longer than about 0.8 m may drop to lower speeds.
- Install current drivers and select the eGPU as the preferred GPU in the operating system.
- Connect an external monitor directly to the card for the best results.
On Windows, the operating system lets each application be assigned to a high-performance GPU in the graphics settings. On Linux, eGPU support works but often requires manual configuration of the display server and PCIe hot-plug authorization, so setup takes longer than on Windows.
Limitations and common mistakes
The total cost of an enclosure and card can approach the price of a complete desktop, so the eGPU is rarely the cheapest route to performance. Hot-unplugging while applications use the GPU may crash them, and some laptops route Thunderbolt through a chipset that reduces bandwidth further.
A frequent mistake is pairing a top-tier graphics card with a Thunderbolt 3 link and expecting desktop results. A mid-range or upper mid-range card usually gives better value, because the link, not the GPU, becomes the bottleneck first.
Noise and space are often overlooked. Enclosures contain their own power supply and fans, and a large card can make the box as loud as a small desktop. Some compact enclosures accept only dual-slot cards of limited length, which rules out many high-end models.
The short version
An external GPU is a powerful graphics card in its own box. It plugs into a laptop with one cable. Think of it as a docking station with a strong engine inside. On the road, the laptop stays light. At the desk, it gains the power to play demanding games or edit video.
The cable is narrower than the connection inside a desktop computer. As a result, some speed is lost, but most of the card’s strength still arrives.
Where it makes a difference
An architecture student uses a 14-inch ultrabook with Thunderbolt 4 for lectures. At home, rendering a building model in real-time visualization software produces only a few frames per second on the integrated GPU, and final renders take over an hour.
She buys an enclosure with a 500 W power supply and a mid-range desktop graphics card with 12 GB of VRAM, connected to a 1440p monitor. The viewport becomes smooth, and the final render time drops to under ten minutes. When she tests the internal laptop screen instead, the frame rate falls noticeably, so she keeps the external monitor attached to the eGPU for all heavy work.