After upgrading an ESXi 8.0 U3 host to a newer build (e.g., Build 25205845) and updating the NVIDIA vGPU Manager to a newer version, users may encounter the following issues:
Failed to get device checkpoint buffer`.2026-01-01T01:01:01.000001Z VMotionStream [180359907:7283880299073427794] failed to allocate buffer for destination VMX to receive incoming device checkpoint data from stream 1: Timeout. vMotion migration [180359907:7283880299073427794] failed preparing for stream completion: Timeout vMotion migration [180359907:7283880299073427794] failed draining stream completion: Failure Failed waiting for data. Error 195887137. Timeout.
ESXi 8.0 and later
This issue is typically caused by a version mismatch or the use of an incorrect driver package type within the Guest OS. While the VMware hypervisor layer correctly presents the vGPU hardware to the VM, the driver installation fails if the guest software cannot successfully handshake with the host's vGPU Manager. This is often isolated to the 3rd-party software layer within the Guest OS.
To resolve this installation failure, ensure strict alignment between the host and guest NVIDIA components:
Troubleshooting and Support
If the installation continues to fail after verifying version alignment and performing a clean install, the issue is likely rooted in the NVIDIA guest-side driver logic.
Note: Because the VMware hypervisor is successfully presenting the PCI hardware to the virtual machine, the troubleshooting of the internal NVIDIA installer and guest driver initialization falls under 3rd-party vendor support.
Action: Contact NVIDIA Support for further analysis of the guest-side installation logs and driver compatibility.