Pro 14 Premium PA14250, bug report eGPU, Thunderbolt 4 Port issue: TB 5 eGPU runtime hot-plugging failure

[Additional Information]

1.

The following reports describe the Gigabyte AORUS RTX 5060 Ti AI Box used with MSI Claw 8 AI+ (also Lunar Lake platform using Intel Core Ultra 7 258V):

https://www.reddit.com/r/MSIClaw/comments/1s2crgl/msi_claw_8ai_with_aorus_rtx_5060_ti_ai_box_egpu/

https://www.reddit.com/r/eGPU/comments/1ruhx93/msi_claw_8_ai_egpu_disconnect_usb4_thunderbolt/

Originally, runtime hot-plugging issues also existed there, but BIOS configuration adjustments eventually enabled successful hot-plugging.

Therefore:

it appears likely that Thunderbolt 5 eGPU runtime hot-plugging issues on Lunar Lake platforms are potentially solvable.

2.

Thunderbolt 3 eGPU behavior:

Thunderbolt 3 eGPUs CAN perform runtime hot-plugging successfully.

However:

during hot-plugging,

continuous WHEA Logger 17 warnings appear unless:

* A power adapter is first connected to the opposite-side Thunderbolt port

* Then the Thunderbolt 3 enclosure is connected

* A GPU workload application (commonly CUDA-Z) is launched for approximately one minute

* After WHEA Logger 17 warnings stop appearing, the opposite-side power adapter can then be disconnected

(Testing performed using:

Razer Core X + MSI RTX 4060 Ti Ventus 2X Black 16G OC)

Therefore:

the Dell Pro 14 Premium PA14250 Thunderbolt eGPU runtime hot-plugging issue appears primarily related to Thunderbolt 5 eGPUs rather than Thunderbolt 3 eGPUs.

[Request]

I would appreciate if this issue could be escalated to the BIOS/Firmware engineering team for investigation.

I hope future BIOS updates can improve Thunderbolt 5 eGPU runtime hot-plugging behavior and related MMIO/BAR allocation handling.

Pro 14 Premium PA14250, bug report eGPU, Thunderbolt 4 Port issue: right-side TB 4 Port Fails to Enumerate eGPU During Cold Boot After BIOS FW 2.12.3 Update

(The following content was translated and polished with the assistance of AI)

[Description of the Bug]

After updating to BIOS firmware 2.12.3 (which updated the following components: System BIOS with BiosGuard, Embedded Controller, and Main System TI Port Controller 0), the right-side Thunderbolt 4 port began failing to enumerate a Thunderbolt eGPU during cold boot scenarios (system powered off → eGPU connected to the right-side Thunderbolt 4 port → system powered on).

This issue remains reproducible under BIOS firmware 2.12.4.

Further investigation showed that under BIOS firmware 2.12.3 and 2.12.4, successful eGPU enumeration on the right-side Thunderbolt 4 port requires BOTH of the following conditions:

1. Before the previous shutdown, the NVIDIA GeForce RTX 5060 Ti device must NOT be in a disabled state inside Device Manager.

2. After shutdown:

   * A power adapter must first be connected to the opposite-side (left-side) Thunderbolt 4 port

   * The eGPU must then be connected to the right-side Thunderbolt 4 port

   * The system must then be powered on

Only under these conditions can the eGPU be enumerated successfully during boot.

However, this behavior was NOT observed under BIOS firmware 2.11.1.

Additionally:

The left-side Thunderbolt 4 port does NOT exhibit this issue.

Under BIOS firmware 2.11.1, 2.12.3, and 2.12.4, the left-side port can still enumerate the eGPU normally during cold boot without requiring the two special conditions described above.

[Environment]

Hardware:

PC: Dell Pro 14 Premium PA14250

BIOS Firmware: 2.12.3 and 2.12.4

eGPU: Gigabyte AORUS RTX 5060 Ti AI Box (2 units)

Operating System:

Windows 11 25H2 Build 26200.8457

Additional Environment Details:

* BIOS settings remain default except:

  * "Power On Lid Open" = Disabled

* Registry modification:

  HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\PnP\Pci

  Added:

  HackFlags = 10000

This registry workaround is required because using two AORUS RTX 5060 Ti AI Boxes simultaneously previously caused a Device Serial Number (DSN) conflict.

Without this HackFlags setting, Event Viewer reports Event ID 56 errors.

* Modern Standby Fast Startup disabled

* NVIDIA driver installed offline (network disconnected during installation)

[Bug Details]

1. Behavior under BIOS firmware 2.11.1 and earlier

Originally, when using Thunderbolt 5 eGPUs on this system, hot plugging without cold boot caused issues:

If the eGPU was connected after the operating system had already booted:

* The system temporarily became unresponsive

* Some devices (WWAN/audio) stopped functioning

* Event Viewer reported "ISH Offload Item Service" errors

* The eGPU failed to enumerate

(This issue appears related to MMIO resource allocation behavior and is described in another report.)

To achieve a workflow similar to hot plugging, the following process was required:

(1) The system still had to be cold booted with the eGPU already attached before power-on.

The eGPU could be connected to either the left-side or right-side Thunderbolt 4 port.

When using two eGPUs simultaneously, both ports could be connected before boot.

Under BIOS firmware 2.11.1 and earlier, the connection order did not matter.

(2) After entering Windows:

If the eGPU needed to be disconnected and later reconnected without issues:

* The NVIDIA GeForce RTX 5060 Ti device itself could NOT be removed using "Safely Remove Hardware and Eject Media"

* Instead, the GPU device first had to be manually disabled inside Device Manager

* Other associated devices could then be safely removed

* This procedure prevented PCIe BAR reallocation issues after reconnection

(The BAR reallocation issue is described in another report.

Briefly:

A 32MB memory region falls outside the Large memory range after reconnection.

This has minimal impact when using a single eGPU, but under dual eGPU configurations the later reconnected GPU experiences severe bandwidth limitations, followed shortly by system hangs and nvlddmkm Event ID 14 errors.)

Under BIOS firmware 2.11.1 and earlier:

As long as the system remained powered on, the eGPU could be disconnected after disabling the GPU device, then later reconnected and re-enabled inside Device Manager to achieve functionality similar to hot plugging.

Additionally:

If the NVIDIA GeForce RTX 5060 Ti device was left disabled before shutdown, simply reconnecting the eGPU before power-on and later re-enabling the GPU device inside Device Manager would still function correctly after boot.

The 32MB memory region also remained inside the Large memory range.

2. Regression under BIOS firmware 2.12.3 and 2.12.4

Under BIOS firmware 2.12.3 and 2.12.4, the right-side Thunderbolt 4 port began failing to enumerate the eGPU during cold boot scenarios.

Details:

(1) Successful enumeration behavior

When cold boot enumeration succeeds, Device Manager properly detects:

* PCI Express Upstream Switch Port

* PCI Express Upstream Switch Port

* NVIDIA GeForce RTX 5060 Ti

* High Definition Audio Controller

* NVIDIA High Definition Audio

Additionally:

When enumeration succeeds, the upstream PCI Express Root Port reserves approximately 32.28GB (33054MB) Large memory range.

If the system boots without an eGPU attached, the corresponding PCI Express Root Port reserves only 32GB (32768MB) Large memory range.

Example:

Successful cold boot enumeration:

Left-side corresponding PCI Express Root Port Large memory range:

[0000002810000000 - 0000003021DFFFFF]

System booted without left-side eGPU attached:

Left-side corresponding PCI Express Root Port Large memory range:

[0000002810000000 - 000000300FFFFFFF]

(2) Failure scenarios on the right-side Thunderbolt 4 port

(2-1)

If the system is powered on with the eGPU connected to the right-side Thunderbolt 4 port WITHOUT first supplying power through the left-side Thunderbolt port:

The following devices fail to enumerate:

* PCI Express Upstream Switch Port

* PCI Express Upstream Switch Port

* NVIDIA GeForce RTX 5060 Ti

* High Definition Audio Controller

* NVIDIA High Definition Audio

Additionally:

The upstream PCI Express Root Port reserves only 32GB (32768MB) Large memory range:

[0000002010000000 - 000000280FFFFFFF]

(2-2)

If the NVIDIA GeForce RTX 5060 Ti device was disabled before the previous shutdown, and after shutdown:

* The left-side Thunderbolt port is powered first

* The eGPU is then connected to the right-side Thunderbolt 4 port

* The system is then powered on

Then:

The following devices CAN be enumerated:

* PCI Express Upstream Switch Port

* PCI Express Upstream Switch Port

* NVIDIA GeForce RTX 5060 Ti (disabled)

* High Definition Audio Controller

* NVIDIA High Definition Audio

However:

The upstream PCI Express Root Port still reserves only 32GB (32768MB) Large memory range:

[0000002010000000 - 000000280FFFFFFF]

Under this condition:

Attempting to re-enable the NVIDIA GeForce RTX 5060 Ti device results in Code 12 error, making the eGPU unusable.

This differs from BIOS firmware 2.11.1 behavior.

Under BIOS firmware 2.11.1:

Even if the NVIDIA GeForce RTX 5060 Ti device was disabled before shutdown, the right-side Thunderbolt 4 eGPU could still enumerate correctly during cold boot.

The upstream PCI Express Root Port would still reserve approximately 32.28GB (33054MB) Large memory range.

Re-enabling the GPU device inside Device Manager would then function normally.

Under BIOS firmware 2.12.3 and 2.12.4:

The left-side Thunderbolt port still behaves correctly.

If the GPU device is disabled before shutdown, then cold booted with the left-side eGPU attached:

* The upstream PCI Express Root Port still reserves approximately 32.28GB (33054MB) Large memory range

* Re-enabling the GPU device inside Device Manager functions normally

(2-3)

If before shutdown:

* The NVIDIA GeForce RTX 5060 Ti device remained enabled

* The eGPU was removed using "Safely Remove Hardware and Eject Media"

Then after shutdown:

* The left-side Thunderbolt port is powered first

* The eGPU is connected to the right-side Thunderbolt 4 port

* The system is powered on

Then:

The following devices enumerate correctly:

* PCI Express Upstream Switch Port

* PCI Express Upstream Switch Port

* NVIDIA GeForce RTX 5060 Ti

* High Definition Audio Controller

* NVIDIA High Definition Audio

Additionally:

The upstream PCI Express Root Port reserves approximately 32.28GB (33054MB) Large memory range:

[0000002010000000 - 0000002821DFFFFF]

Under this condition, the GPU functions normally.

Contact Dell GHN (Get Help Now) chat technical support Monday through Friday. Click the blue "Get Help Now" on the right to start a private live chat session. Share the private Service Tag with them so that they can verify the ownership and warranty status. This will also generate a unique Technical Support case for your unique Service Tag. Provide link to this thread so that they can see the troubleshooting.

(The following section was originally intended to be the final paragraph of the bug report titled:

"Pro 14 Premium PA14250 - eGPU / Thunderbolt 4 Port issue: right-side TB 4 Port Fails to Enumerate eGPU During Cold Boot After BIOS FW 2.12.3 Update"

This section mainly discusses the user-experience impact of the issue. However, it appears to have been omitted when Dell forum staff merged the two original bug report threads into a single thread for administrative reasons.

------

By the way:

After the Dell forum staff merged the two original bug report threads into a single thread, I also lost the ability to edit — or even delete — the content of my own original posts (for reasons unknown to me).

I still believe the two bugs discussed in the merged thread are fundamentally different issues from a user-experience perspective.

"right-side TB 4 Port Fails to Enumerate eGPU During Cold Boot After BIOS FW 2.12.3" is a newly introduced regression that started around May 2026. This behavior previously worked correctly, and the issue has a more direct and frustrating impact on day-to-day usability.

In contrast, "TB 5 eGPU runtime hot-plugging failure" appears to be a much longer-standing issue.

While I understand there may have been administrative or organizational reasons for merging the two reports, the merge also introduces a potential side effect: if the merged thread link is provided directly to an AI system for analysis, the AI may incorrectly interpret the thread as discussing only the "TB 5 eGPU runtime hot-plugging failure" issue, while overlooking the separate "right-side TB 4 Port Fails to Enumerate eGPU During Cold Boot After BIOS FW 2.12.3" regression entirely.

In practice, I have found that the full text content must be provided directly to the AI in order for it to correctly recognize that the thread actually reports two distinct categories of bugs.)

------

[Practical Impact]

Under BIOS firmware 2.11.1, although true hot plugging was not functional, the workflow was still practical.

The Dell Pro 14 Premium PA14250 is lightweight and portable, making it convenient for daily office and meeting usage.

When eGPU acceleration was needed for LLM inference:

* Both eGPUs only needed to be connected before that boot session

* Before disconnecting the eGPUs, the GPU devices simply needed to be disabled inside Device Manager

* Later reconnection only required reconnecting the eGPUs and re-enabling the GPU devices

With dual RTX 5060 Ti 16GB Thunderbolt eGPUs, quantized models under 32GB can run successfully.

Although Thunderbolt 4 bandwidth limitations reduce scaling efficiency, actual testing using Ollama with the Gemma4 26B A4B Q8_0 (27GB) model still achieves approximately 60 tokens/s, making the setup practical for real-world workloads.

However, under BIOS firmware 2.12.3 and 2.12.4, the workflow becomes significantly less convenient.

Although dual eGPU functionality can still be achieved by:

* Connecting the first eGPU to the left-side Thunderbolt port before boot

* Then connecting the second eGPU to the right-side Thunderbolt port

* Then powering on the system

Before every shutdown, the user must also remember to:

* Reconnect the eGPU to the right-side Thunderbolt 4 port

* Re-enable the right-side NVIDIA GPU device inside Device Manager

* Remove the eGPU using "Safely Remove Hardware and Eject Media"

Only then can the system shut down while preserving proper future enumeration behavior.

As a result, the eGPU hardware must remain physically available before shutdown, which is highly inconvenient.

[Request]

I hope this issue can be escalated to the BIOS/Firmware engineering team for further investigation.

Ideally:

* Proper MMIO resource allocation behavior for the right-side Thunderbolt 4 port during cold boot can be restored

  OR

* The long-standing Thunderbolt 5 eGPU hot plugging issues can be improved in future BIOS/firmware updates.

[Additional Information: Troubleshooting Steps Attempted, But Did Not Help]

Regarding "right-side TB 4 Port Fails to Enumerate eGPU During Cold Boot After BIOS FW 2.12.3", the following actions were performed during the troubleshooting process but did not resolve the issue:

    (1) Reflashing the BIOS Firmware: While running BIOS FW 2.12.3, I attempted to reflash the exact same BIOS FW 2.12.3 version, but the issue persisted.

    (2) BIOS Configuration Adjustments: Modifying the following options yielded no change in behavior:

        Thunderbolt Boot Support: Toggling this option Enabled/Disabled made no difference.

        Enable Thunderbolt Technology Support: Disabling the option, rebooting, re-enabling it, and rebooting again made no difference.

        Enable Pre-Boot DMA Support: Toggling this option Enabled/Disabled made no difference.

        Extend BIOS POST Time: Changing this setting between 10s and 0s made no difference.

    (3) RTC Reset: Performing a Real-Time Clock (RTC) reset did not resolve the issue.

    (4) Windows 11 Security Settings: Disabling "Memory Integrity" (Core Isolation) and "Kernel-mode Hardware-enforced Stack Protection" had no effect. (Since disabling them did not help, they have been re-enabled).

    (5) Clean GPU Driver Reinstallation: Used DDU (Display Driver Uninstaller) to completely wipe old NVIDIA drivers before performing a clean reinstallation, but it had no effect.

    (6) Windows Power Management: Disabling "Link State Power Management (ASPM)" under the PCI Express settings in Windows Power Options had no effect. (Since disabling it did not help, it has been reverted to its original setting).

    (7) Device Manager Uninstallation: Intentionally uninstalled both the upstream PCI Express Root Port corresponding to the right side and the USB4 Host Router inside Device Manager to let Windows rediscover them, but the behavior remained unchanged.

The only time Dell tested/validated an eGPU was back in 2014 for the Dell Alienware AGA (Alienware Graphics Amplifier). eGPU are not sold on the Dell sales site. Dell is not going to purchase 3rd party Thunderbolt 5 eGPU to do testing and validation. You should contact the manufacturer website for support of their Thunderbolt 5 eGPU.

There is also no mention of Thunderbolt 5 compatibility in the online Pro 14 Premium PA14250 Owner's Manual or sales page specifications.z

The Pro 14 Premium PA14250 is natively equipped with Thunderbolt 4 (40 Gbps) ports which offer backward compatibility with Thunderbolt 3 and USB4 devices. Such as connecting to an external display using USB Type-C to DisplayPort adapter or to a Dell dock.

@DELL-ChrisM2​ 

The primary issue is not general “eGPU support”.

The observed regression affects Thunderbolt PCIe device enumeration and PCIe resource allocation behavior on the system’s native Thunderbolt 4 ports after BIOS 2.12.3.

The eGPU merely serves as a reproducible PCIe tunneling test device that exposes the regression.

The issue specifically involves:

• PCIe Large Memory allocation changes
• Runtime PCIe resource rebalance behavior
• Thunderbolt PCIe hierarchy enumeration differences between BIOS versions
• Cold boot enumeration regression on the right-side Thunderbolt 4 port

These behaviors changed between BIOS 2.11.1 and BIOS 2.12.3 despite identical hardware and operating system environments.

@fbbt160_1​ Just to confirm, have you contacted Support through the Get Help Now button? Frontline support more than likely will not be able to address the concern, but they should be able to contact the necessary teams for additional guidance. A case must be created to properly address the concern with Dell Support. 

@fbbt160_1​ ,

The observed regression affects Thunderbolt PCIe device enumeration and PCIe resource allocation behavior on the system’s native Thunderbolt 4 ports after BIOS 2.12.3.

Please verify that all the included sub-firmwares properly updated the various micro-controllers (including the Thunderbolt ports).

You can do this by re-running the BIOS-Firmware v2.12.3 updater and compare the columns (also scroll down). Then exit to bail-out.

I suggest you also test if VBS (Core Isolation / Memory Integrity) Enabled setting makes a difference or not.

@DELL-Daniel V​ 

Thank you for your kind reply!

I did try using Get Help Now on May 28th. However, the agent and I unfortunately missed each other—the case was closed before I had a chance to respond. Due to the nature of my job, I am unable to reply immediately during certain work hours, which likely caused the ticket to time out and close automatically.

Because of this, I submitted another Get Help Now request on Friday night, focusing specifically on reporting this firmware regression bug. (I have also submitted a separate Service Request). I truly hope this detailed bug report can be successfully escalated to the relevant engineering team!

P.S. Regarding Usability & Long-Term Stability Concerns

    Current Workaround: It is still technically possible to use dual eGPUs on this laptop right now, but the workflow is significantly more cumbersome and inconvenient compared to how seamlessly it worked under BIOS firmware 2.11.1.

    The Bigger Concern (Stability): Beyond the immediate inconvenience, my biggest worry is future stability. I am concerned that subsequent BIOS firmware updates might degrade behavior further or break dual eGPU support entirely. If that happens, the local AI infrastructure and workflow that I spent considerable time and capital setting up will be completely disrupted.

    My Appeal to the Firmware Team: This is yet another reason why I hope the firmware engineering team is aware of this issue. While long-standing legacy issues may not be resolved immediately, it is essential that firmware updates preserve existing functionality whenever possible. Features that functioned correctly in previous releases (such as BIOS 2.11.1) should not be unintentionally broken, degraded, or removed as a result of new updates or bug fixes. Maintaining backward compatibility and preserving proven functionality are critical to ensuring a reliable user experience.

(PS: The above content was translated and polished with the assistance of AI)

@Tesla1856​ 

It is truly encouraging to see such a professional and insightful response!

The suggestions you mentioned regarding VBS (Core Isolation / Memory Integrity)  and ensuring the sub-firmwares are properly updated align perfectly with the initial steps I analyzed during my troubleshooting process with various AI models (ChatGPT, Gemini, and Claude). Unfortunately, subsequent attempts to reflash the BIOS firmware  (while verifying that all sub-firmware components were correctly applied) as well as toggling the Windows 11 VBS settings yielded no improvement. I have detailed these attempted troubleshooting steps that had no effect in one of my earlier replies within this thread. 

I would also like to clarify that it wasn't my personal opinion that this firmware regression bug needs to be escalated to the engineering level. Throughout our deep-dive troubleshooting session, all three AI models independently and consistently concluded that the symptoms point to a firmware-level issue and recommended submitting a formal bug report to Dell for review by the BIOS/Firmware engineering team. The original bug report and the additional technical details provided in my replies are available for independent review, and I welcome any alternative analysis or conclusions based on the same information.

P.S. A Fascinating(?) Side Note on the Workaround

As a side note, discovering the highly specific workaround—where the left-side Thunderbolt 4 port must be supplied with power first in order for the right-side eGPU to successfully enumerate during a cold boot —was actually a breakthrough proposed by one of the AIs. 

After I fed the AI information regarding a late April 2026 thread on the Dell Forum—where other users reported that the PA14250 failed to boot normally from a cold boot when connected to a Thunderbolt 4 dock—the AI hypothesized that the subsequent firmware update deployed to address that docking issue might have inadvertently introduced a side effect on the right port. It theorized that:

    "EC/TI Port Controller 0 update in BIOS 2.12.3 introduced a timing or synchronization bug where PD power negotiation delays or interferes with the PCIe tunneling handshake during POST on the right port."

Based entirely on this deduction, it suggested the power-sequencing workaround, which successfully broke the deadlock. While I cannot confirm whether its underlying architectural hypothesis is the exact root cause, its suggested sequence is currently the only temporary method that allows me to successfully initialize and utilize my dual eGPUs.

(PS: The above content was translated and polished with the assistance of the AI that found solution (Apparently, it is very, very proud of itself!))