Misc. bug: Only using 1 compute core on AMD

[vRobM]

Name and Version

LM Studio 0.3.14
latest vulcan-llama.cpp v1.26.0
CPU Ryzen 7 7800 X3D w/ on die GPU only
{
"name": "llama.cpp-win-x86_64-vulkan-avx2",
"version": "1.26.0",
"domains": [
"llm",
"embedding"
],
"engine": "llama.cpp",
"target_libraries": [
{
"name": "llm_engine_vulkan.node",
"type": "llm_engine",
"version": "0.1.2"
},
{
"name": "liblmstudio_bindings_vulkan.node",
"type": "liblmstudio",
"version": "0.2.26"
}
],
"platform": "win",
"cpu": {
"architecture": "x86_64",
"instruction_set_extensions": [
"AVX2"
]
},
"gpu": {
"framework": "Vulkan"
},
"supported_model_formats": [
"gguf"
],
"vendor_lib_package_names": [
"win-llama-vulkan-vendor-v1"
],
"manifest_version": "4",
"extension_type": "engine"
}

[
{
"modelCompatibilityType": "gguf",
"runtime": {
"hardwareSurveyResult": {
"compatibility": {
"status": "Compatible"
},
"cpuSurveyResult": {
"result": {
"code": "Success",
"message": ""
},
"cpuInfo": {
"name": "",
"architecture": "x86_64",
"supportedInstructionSetExtensions": [
"AVX",
"AVX2"
]
}
},
"memoryInfo": {
"ramCapacity": 67874881536,
"vramCapacity": 0,
"totalMemory": 67874881536
},
"gpuSurveyResult": {
"result": {
"code": "Success",
"message": ""
},
"gpuInfo": [
{
"name": "AMD Radeon(TM) Graphics",
"deviceId": 0,
"totalMemoryCapacityBytes": 34474295296,
"dedicatedMemoryCapacityBytes": 0,
"integrationType": "Integrated",
"detectionPlatform": "Vulkan",
"detectionPlatformVersion": "1.3.283",
"otherInfo": {
"deviceLUIDValid": "true",
"deviceLUID": "b2f4560000000000",
"deviceUUID": "00000000100000000000000000000000",
"driverID": "1",
"driverName": "AMD proprietary driver",
"driverInfo": "25.3.2 (AMD proprietary shader compiler)",
"vendorID": "4098"
}
}
]
}
}
}
}
]

Should more devices/cores be detected?
Should both compute cores be used?
Can the 3D part be used?

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

No response

Command line

Problem description & steps to reproduce

Running maziyarpanahi/gemma-3-1b-it

works, just looks like it's limping one 1 compute core.

First Bad Commit

No response

Relevant log output

[pl752]

I think that Compute_0 and Compute_1 are not separate compute units, but just separate performance metrics, so actually your iGPU is fully utilized

[vRobM]

Okay, thank you.
There are a number of other options as well