Error messages

Error surveying hardware

Error surveying hardware

In LM Studio's runtime list, one engine carries a red label instead of "Compatible", and the Hardware page shows 0 GPUs:

Error surveying hardware

If you copy the system info from the Hardware tab, the same thing appears as JSON, quoted in #287:

"hardwareSurveyResult": {
  "compatibility": {
    "status": "Error surveying hardware",
    "message": "Error occured while surveying hardware"
  },

LM Studio's code isn't public, so this page rests on the company's bug tracker. As of October 1, 2026 it had 10 issues quoting the label, spread across Windows and Linux and all three GPU vendors.

What the survey is

LM Studio runs models through runtimes, separate builds of llama.cpp for CUDA (NVIDIA), ROCm (AMD), Vulkan and plain CPU. Before one can be used, it loads its own library and asks the hardware what's there. The log shows what happens when that load fails. A team member quoted this line in #120:

Failed to survey hardware with backend 'llama.cpp-linux-x86_64-nvidia-cuda-avx2', version '1.1.11': LMSCore load lib failed - child process with PID 3937710 exited with code 1

So the label tells you that one runtime died during its first check. The other runtimes are surveyed on their own, which is why the CPU one often still says it works.

AMD cards after a driver update

This is the biggest cluster in 2026. In #1906, AMD's Adrenalin 26.5.1 driver broke the ROCm runtime on RX 9070 XT cards: the survey crashed, the GPU count went to zero and runtime downloads hung at 100%. Users fixed it in two ways. Some rolled the driver back to 26.3.1. Others moved to the beta channel, where LM Studio 0.4.17 (Build 2) shipped with the note "Fixed bug where AMD GPUs were not detected due to driver updates", according to one commenter, and another confirmed both ROCm and Vulkan working after it.

When someone on runtime 2.22.0 still hit it in #2070, LM Studio's rugvedS07 replied on June 19, 2026 that "the fix landed in 2.22.1" and to update to the latest beta runtime. He also asked users to hover over the error under the runtime's triple dot menu, Manage Versions, which shows more detail than the label.

It isn't gone everywhere. One user on ROCm runtime 2.46.0 reported a failure on September 27, 2026, and a Radeon Pro V620 owner was still stuck in June (rugvedS07 pointed out that AMD's Windows ROCm 7 support list doesn't include that card). The command that user ran, lms runtime survey --refresh --json, reruns the survey and prints the status, in their case gpuSurveyUnsuccessful.

Other causes in the threads

In #287 every runtime failed unless LM Studio ran as administrator. The reporter later found why on his machine: his ~/.cache folder was a link to a directory on another drive, and making it a real folder fixed it. Someone with an RTX 5070 Ti in 2026 got no help from running as admin, so it doesn't explain every case.

In #379 an Intel Arc A770 stopped being detected after a Vulkan runtime update. A team member said build 6 of 0.3.9 might fix it, and the reporter said it didn't.

What to try, in order

Update LM Studio and the failing runtime, and restart the app. Check the hover text in Manage Versions. On AMD, compare your Adrenalin version with the ones in #1906. If you link folders like ~/.cache elsewhere, try a real folder. While you sort it out, select the Vulkan or CPU runtime so you can keep loading models.