Shattered.io’s 12-step eGPU guide proposes roughly 90 minutes for assembly, configuration and testing. Treat that as a planning estimate, not a stopwatch challenge. The useful objective is a setup that survives repeatable testing, runs applications on the intended GPU and leaves you with enough diagnostic information to troubleshoot it later.
Below is a Windows 11-focused procedure, with a separate Linux checkpoint and several important corrections to overly broad setup advice.
Understand the connection before buying
Thunderbolt and USB4 carry multiple kinds of traffic. Their advertised connection speeds are not equivalent to bandwidth reserved exclusively for the graphics card. Intel’s specifications distinguish total connection bandwidth from PCIe data requirements—a distinction that matters considerably for eGPUs.
| Connection | What to establish before purchase |
|---|---|
| Thunderbolt 5 | Compatible host, enclosure and cable. Intel specifies up to 80 Gbps bidirectional connectivity and a 64 Gbps minimum PC PCIe data requirement. |
| Thunderbolt 4 | External graphics support and enclosure compatibility. Intel specifies 40 Gbps connectivity and a 32 Gbps minimum PC PCIe data requirement. |
| USB4 | The host’s actual capabilities, supported enclosure and external graphics compatibility—not merely the USB-C connector. |
| OCuLink | The exact host and dock interface, supported PCIe generation, lane count and connection procedure. Follow both manufacturers’ documentation. |
Intel describes Thunderbolt 5’s 120 Gbps Bandwidth Boost as a response to heavy display traffic. It is not a promise of a 120 Gbps PCIe connection to an external GPU. Intel lists Thunderbolt 5 support beginning with Windows 11 version 23H2 and Linux kernel 6.7; those are technology-support floors, not guarantees for every GPU and enclosure combination.
There is also a useful Windows-specific qualification to the usual “USB4 might not support PCIe” warning. Microsoft’s Windows USB4 design requirements call for PCIe tunneling on exposed USB4 connectors. However, Microsoft also documents that its connection manager will not create those tunnels when tunneling is disabled. For buyers, the practical conclusion remains unchanged: establish compatibility for the actual laptop and enclosure, including firmware settings.
A prebuilt example—and its limitation
ASUS’s November 6, 2026 announcement confirms an XG Core external graphics unit containing a Radeon RX 9060 XT LP with 16GB GDDR6. Crucially, ASUS specifies USB4 connectivity up to 40 Gbps, not 80 Gbps. It also provides HDMI and DisplayPort outputs and tells buyers to check host connectivity and external GPU support. This is evidence of an announced product, not independent performance testing or proof of retail availability.
That example illustrates the central buying rule: purchase the capabilities documented for the product, not the maximum capabilities associated with its connector family.
Before starting: your compatibility checklist
For a user-assembled enclosure, establish:
- Host support: the laptop manufacturer’s external graphics guidance.
- Physical fit: card length, height, thickness and power-cable clearance.
- Electrical compatibility: the enclosure’s required power supply and the GPU’s required connectors.
- Software support: a suitable operating system and graphics driver.
- Display connection: an appropriate cable for a monitor connected to the graphics card.
- Included equipment: whether the enclosure actually includes its power supply and host cable. Razer, for example, sells the Core X V2 without a graphics card or PSU.
Do not size the PSU from GPU board power alone. Razer specifically requires an additional 230 watts beyond the graphics card’s needs for its Core X V2. That is a model-specific requirement, not a universal eGPU allowance, but it demonstrates why “a little headroom” is an inadequate substitute for the enclosure manual.
The 12-step laptop eGPU setup
1. Identify the laptop’s supported connection
Start with the specification and support documentation for your exact laptop model.
On Windows 11 systems exposing Microsoft’s USB4 settings, the Settings app can list attached USB4 hubs, devices and capabilities. Microsoft documents this functionality beginning with build 22621.1778. Use it as supporting evidence alongside the manufacturer’s specifications—not as a stand-alone eGPU certification test.
Checkpoint: you can identify the supported host interface and a documented route to external graphics.
2. Check firmware and PCIe tunneling
Follow the laptop manufacturer’s firmware-update procedure and review its relevant USB4 or Thunderbolt settings.
Do not indiscriminately select a BIOS option called “No Security.” Microsoft’s documentation establishes that disabled PCIe tunneling prevents tunnel creation, but it does not make disabling security protections a general setup requirement. On a managed work laptop, involve IT rather than changing device-access policy yourself.
3. Record a repeatable baseline
Before installing the eGPU, choose a benchmark or real task you can reproduce. Record its resolution, quality settings, application version and result.
For games, capture average frame rate and frame-time consistency where available. For rendering, use completion time. For local AI, use the same model, configuration and workload each time.
This is a testing recommendation, not a claim about expected gains: the baseline makes your own comparison useful.
4. Assemble the enclosure safely
Switch off the enclosure and disconnect mains power before opening it. Following its manual:
- Install the required PSU, if supplied separately.
- Seat and secure the graphics card.
- Connect the specified enclosure and GPU power cables.
- Keep cables clear of fans.
- Refit the cover.
Razer’s Core X V2 procedure explicitly covers these operations and warns against working with the enclosure connected to a power outlet. Connector arrangements vary, so do not transplant one enclosure’s wiring instructions to another.
5. Connect the external monitor to the card
For the initial test, connect the monitor to an HDMI or DisplayPort output on the installed graphics card, rather than assuming a laptop or dock display output follows the same route. Razer’s enclosure instructions specify this connection for an external monitor.
Keep the display arrangement unchanged during subsequent comparisons.
6. Connect and power up as documented
Use the enclosure’s supplied or specified host cable and follow its connection sequence. Razer documents plug-and-play operation for compatible Core X V2 hosts once the hardware is installed. That does not establish the correct sequence for every dock.
For OCuLink, use the conservative procedure in Shattered.io’s guide: shut down before connecting or disconnecting, and follow the dock’s prescribed power sequence. Do not assume Thunderbolt-style hot-plug behavior.
Nor should every Windows user install Thunderbolt Control Center automatically. Windows has its own USB4 connection manager; the applicable management software depends on the host implementation.
7. Install the appropriate graphics driver
Install the driver appropriate to the external card and follow any enclosure-specific instructions.
A particularly important exception concerns laptops already containing an NVIDIA discrete GPU. Razer publishes a dedicated Core X V2 driver-installation procedure for that configuration. That is a good reason not to assume every mixed-GPU system follows one universal installation recipe.
Do not make third-party driver removal the first step of every build. Start with the supported installation path and reserve more disruptive cleanup for a diagnosed problem.
8. Assign the application in Windows 11
Microsoft documents per-application graphics preferences under:
Settings → System → Display → Graphics
Select or add the application, open its graphics options, choose the appropriate preference and save. Restart the application afterward. Where Windows displays a GPU name beside a preference, check that name instead of assuming “High performance” necessarily identifies the external card.
Checkpoint: the application is configured for the intended GPU, not merely displaying on an attached screen.
9. Inspect the link under load
For NVIDIA hardware, use:
nvidia-smi --query-gpu=name,pcie.link.gen.max,pcie.link.gen.current,pcie.link.width.current --format=csv
NVIDIA explicitly documents that current PCIe generation and width can fall while the GPU is idle. Repeat the check during a workload before diagnosing a fault. These readings describe the reported PCIe link; do not treat them alone as an end-to-end Thunderbolt bandwidth benchmark.
10. Follow the Linux-specific path, if applicable
Linux is an alternative configuration path, not an extra requirement for Windows users.
Shattered.io describes using boltctl for authorization and lspci for enumeration. Their applicability depends on the distribution, installed services and hardware security configuration; there is no single package-installation command covering every Linux system.
For supported NVIDIA PRIME render-offload configurations, NVIDIA documents:
__NV_PRIME_RENDER_OFFLOAD=1 __GLX_VENDOR_LIBRARY_NAME=nvidia glxinfo -B
This requires glxinfo and a correctly configured driver stack. Inspect the renderer before launching the target workload. With several NVIDIA GPUs, identifying an NVIDIA renderer alone does not prove which card was selected.
Do not change kernels solely on the strength of an unrelated forum configuration.
11. Repeat the baseline and test stability
Repeat the same workload several times with unchanged settings. Then run a sustained workload while observing temperatures, errors and disconnects.
Recommended acceptance criteria are practical:
- The intended application uses the external GPU.
- Performance meets your needs consistently.
- The enclosure remains connected.
- There are no repeatable display failures or driver resets.
- Results remain reproducible after restarting.
Being “within a few percent of a desktop” is not a universal pass criterion. Your laptop, enclosure and workload constitute a different system.
12. Save a diagnostic record
For NVIDIA hardware, capture a telemetry snapshot during the test:
nvidia-smi --query-gpu=name,temperature.gpu,power.draw,utilization.gpu,pcie.link.gen.current,pcie.link.width.current --format=csv
Save the output with the workload settings and driver version. Unsupported NVIDIA metrics may report N/A; this command is not an AMD logger.
Shattered.io’s accompanying scripts need that same scope warning: their detailed CSV telemetry depends on nvidia-smi. Reading AMD PCIe information through Linux sysfs does not make the scripts complete AMD temperature, power and utilization loggers.
Troubleshooting without making matters worse
| Problem | Sensible first action |
|---|---|
| Card does not enumerate | Recheck documented host support, connection and PCIe-tunneling settings. Microsoft confirms disabled tunneling prevents tunnel creation. |
| Application selects the wrong GPU | Revisit Windows per-app graphics preferences and restart the application. |
| Low PCIe reading at idle | Repeat under load before drawing conclusions. |
| Installation fails on a laptop with an NVIDIA dGPU | Use the enclosure maker’s configuration-specific driver procedure. |
| Assembly or power is uncertain | Shut down, disconnect mains power and recheck the enclosure manual before continuing. |
These checks follow Microsoft, NVIDIA and Razer documentation; they are starting points, not diagnoses of every failure.
The real measure of success
An eGPU project should begin with compatibility and end with a repeatable result—not with an expensive card and crossed fingers.
The strongest approach is straightforward: establish the supported connection, follow the enclosure’s electrical and assembly requirements, configure application routing, test under load and retain a record. Ninety minutes may be enough for a cooperative combination. Taking longer to verify the essentials is still faster than troubleshooting a connector-shaped assumption.
References
- eGPU for Laptop: 12 Steps, 90 Min (2026) - shattered.io shattered.io · 2026-11-16T09:22:06+00:00
- Optimus Laptops and Multi-GPU Desktop Systems — NVIDIA Driver Installation Guide docs.nvidia.com
- Razer Core X V2 - Thunderbolt™ 5 eGPU | Razer United States razer.com