Skip to content

cuda.bindings.nvml: Big wave of handwritten API #1524

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
mdboom wants to merge 6 commits into
base: main
Choose a base branch
from
Draft

cuda.bindings.nvml: Big wave of handwritten API #1524

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
338d5b1
cuda.bindings.nvml: Big wave of handwritten API
mdboom Jan 22, 2026
b11d828
Minor changes based on Copilot's comments
mdboom Jan 22, 2026
1299944
Fix tests, having run on more hardware
mdboom Jan 22, 2026
6fc9e77
Further fix tests
mdboom Jan 22, 2026
3566c1d
Fix tests on even more hardware
mdboom Jan 22, 2026
a75e4c3
More robust testing for Windows
mdboom Jan 23, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
9 changes: 1 addition & 8 deletions cuda_bindings/cuda/bindings/_internal/_nvml.pxd
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
#
# SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
#
Expand Down Expand Up @@ -198,7 +198,6 @@ cdef nvmlReturn_t _nvmlDeviceSetDefaultAutoBoostedClocksEnabled(nvmlDevice_t dev
cdef nvmlReturn_t _nvmlDeviceSetDefaultFanSpeed_v2(nvmlDevice_t device, unsigned int fan) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceSetFanControlPolicy(nvmlDevice_t device, unsigned int fan, nvmlFanControlPolicy_t policy) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceSetTemperatureThreshold(nvmlDevice_t device, nvmlTemperatureThresholds_t thresholdType, int* temp) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceSetPowerManagementLimit(nvmlDevice_t device, unsigned int limit) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceSetGpuOperationMode(nvmlDevice_t device, nvmlGpuOperationMode_t mode) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceSetAPIRestriction(nvmlDevice_t device, nvmlRestrictedAPI_t apiType, nvmlEnableState_t isRestricted) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceSetFanSpeed_v2(nvmlDevice_t device, unsigned int fan, unsigned int speed) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
Expand Down Expand Up @@ -337,13 +336,7 @@ cdef nvmlReturn_t _nvmlDeviceGetComputeInstanceId(nvmlDevice_t device, unsigned
cdef nvmlReturn_t _nvmlDeviceGetMaxMigDeviceCount(nvmlDevice_t device, unsigned int* count) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceGetMigDeviceHandleByIndex(nvmlDevice_t device, unsigned int index, nvmlDevice_t* migDevice) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceGetDeviceHandleFromMigDeviceHandle(nvmlDevice_t migDevice, nvmlDevice_t* device) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlGpmSampleGet(nvmlDevice_t device, nvmlGpmSample_t gpmSample) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlGpmMigSampleGet(nvmlDevice_t device, unsigned int gpuInstanceId, nvmlGpmSample_t gpmSample) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlGpmQueryDeviceSupport(nvmlDevice_t device, nvmlGpmSupport_t* gpmSupport) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlGpmQueryIfStreamingEnabled(nvmlDevice_t device, unsigned int* state) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlGpmSetStreamingEnabled(nvmlDevice_t device, unsigned int state) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceGetCapabilities(nvmlDevice_t device, nvmlDeviceCapabilities_t* caps) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDeviceWorkloadPowerProfileClearRequestedProfiles(nvmlDevice_t device, nvmlWorkloadPowerProfileRequestedProfiles_t* requestedProfiles) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDevicePowerSmoothingActivatePresetProfile(nvmlDevice_t device, nvmlPowerSmoothingProfile_t* profile) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDevicePowerSmoothingUpdatePresetProfileParam(nvmlDevice_t device, nvmlPowerSmoothingProfile_t* profile) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
cdef nvmlReturn_t _nvmlDevicePowerSmoothingSetState(nvmlDevice_t device, nvmlPowerSmoothingState_t* state) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil
Expand Down
149 changes: 1 addition & 148 deletions cuda_bindings/cuda/bindings/_internal/_nvml_linux.pyx
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
#
# SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
#
Expand Down Expand Up @@ -247,7 +247,6 @@ cdef void* __nvmlDeviceSetDefaultAutoBoostedClocksEnabled = NULL
cdef void* __nvmlDeviceSetDefaultFanSpeed_v2 = NULL
cdef void* __nvmlDeviceSetFanControlPolicy = NULL
cdef void* __nvmlDeviceSetTemperatureThreshold = NULL
cdef void* __nvmlDeviceSetPowerManagementLimit = NULL
cdef void* __nvmlDeviceSetGpuOperationMode = NULL
cdef void* __nvmlDeviceSetAPIRestriction = NULL
cdef void* __nvmlDeviceSetFanSpeed_v2 = NULL
Expand Down Expand Up @@ -386,13 +385,7 @@ cdef void* __nvmlDeviceGetComputeInstanceId = NULL
cdef void* __nvmlDeviceGetMaxMigDeviceCount = NULL
cdef void* __nvmlDeviceGetMigDeviceHandleByIndex = NULL
cdef void* __nvmlDeviceGetDeviceHandleFromMigDeviceHandle = NULL
cdef void* __nvmlGpmSampleGet = NULL
cdef void* __nvmlGpmMigSampleGet = NULL
cdef void* __nvmlGpmQueryDeviceSupport = NULL
cdef void* __nvmlGpmQueryIfStreamingEnabled = NULL
cdef void* __nvmlGpmSetStreamingEnabled = NULL
cdef void* __nvmlDeviceGetCapabilities = NULL
cdef void* __nvmlDeviceWorkloadPowerProfileClearRequestedProfiles = NULL
cdef void* __nvmlDevicePowerSmoothingActivatePresetProfile = NULL
cdef void* __nvmlDevicePowerSmoothingUpdatePresetProfileParam = NULL
cdef void* __nvmlDevicePowerSmoothingSetState = NULL
Expand Down Expand Up @@ -1735,13 +1728,6 @@ cdef int _init_nvml() except -1 nogil:
handle = load_library()
__nvmlDeviceSetTemperatureThreshold = dlsym(handle, 'nvmlDeviceSetTemperatureThreshold')

global __nvmlDeviceSetPowerManagementLimit
__nvmlDeviceSetPowerManagementLimit = dlsym(RTLD_DEFAULT, 'nvmlDeviceSetPowerManagementLimit')
if __nvmlDeviceSetPowerManagementLimit == NULL:
if handle == NULL:
handle = load_library()
__nvmlDeviceSetPowerManagementLimit = dlsym(handle, 'nvmlDeviceSetPowerManagementLimit')

global __nvmlDeviceSetGpuOperationMode
__nvmlDeviceSetGpuOperationMode = dlsym(RTLD_DEFAULT, 'nvmlDeviceSetGpuOperationMode')
if __nvmlDeviceSetGpuOperationMode == NULL:
Expand Down Expand Up @@ -2708,55 +2694,13 @@ cdef int _init_nvml() except -1 nogil:
handle = load_library()
__nvmlDeviceGetDeviceHandleFromMigDeviceHandle = dlsym(handle, 'nvmlDeviceGetDeviceHandleFromMigDeviceHandle')

global __nvmlGpmSampleGet
__nvmlGpmSampleGet = dlsym(RTLD_DEFAULT, 'nvmlGpmSampleGet')
if __nvmlGpmSampleGet == NULL:
if handle == NULL:
handle = load_library()
__nvmlGpmSampleGet = dlsym(handle, 'nvmlGpmSampleGet')

global __nvmlGpmMigSampleGet
__nvmlGpmMigSampleGet = dlsym(RTLD_DEFAULT, 'nvmlGpmMigSampleGet')
if __nvmlGpmMigSampleGet == NULL:
if handle == NULL:
handle = load_library()
__nvmlGpmMigSampleGet = dlsym(handle, 'nvmlGpmMigSampleGet')

global __nvmlGpmQueryDeviceSupport
__nvmlGpmQueryDeviceSupport = dlsym(RTLD_DEFAULT, 'nvmlGpmQueryDeviceSupport')
if __nvmlGpmQueryDeviceSupport == NULL:
if handle == NULL:
handle = load_library()
__nvmlGpmQueryDeviceSupport = dlsym(handle, 'nvmlGpmQueryDeviceSupport')

global __nvmlGpmQueryIfStreamingEnabled
__nvmlGpmQueryIfStreamingEnabled = dlsym(RTLD_DEFAULT, 'nvmlGpmQueryIfStreamingEnabled')
if __nvmlGpmQueryIfStreamingEnabled == NULL:
if handle == NULL:
handle = load_library()
__nvmlGpmQueryIfStreamingEnabled = dlsym(handle, 'nvmlGpmQueryIfStreamingEnabled')

global __nvmlGpmSetStreamingEnabled
__nvmlGpmSetStreamingEnabled = dlsym(RTLD_DEFAULT, 'nvmlGpmSetStreamingEnabled')
if __nvmlGpmSetStreamingEnabled == NULL:
if handle == NULL:
handle = load_library()
__nvmlGpmSetStreamingEnabled = dlsym(handle, 'nvmlGpmSetStreamingEnabled')

global __nvmlDeviceGetCapabilities
__nvmlDeviceGetCapabilities = dlsym(RTLD_DEFAULT, 'nvmlDeviceGetCapabilities')
if __nvmlDeviceGetCapabilities == NULL:
if handle == NULL:
handle = load_library()
__nvmlDeviceGetCapabilities = dlsym(handle, 'nvmlDeviceGetCapabilities')

global __nvmlDeviceWorkloadPowerProfileClearRequestedProfiles
__nvmlDeviceWorkloadPowerProfileClearRequestedProfiles = dlsym(RTLD_DEFAULT, 'nvmlDeviceWorkloadPowerProfileClearRequestedProfiles')
if __nvmlDeviceWorkloadPowerProfileClearRequestedProfiles == NULL:
if handle == NULL:
handle = load_library()
__nvmlDeviceWorkloadPowerProfileClearRequestedProfiles = dlsym(handle, 'nvmlDeviceWorkloadPowerProfileClearRequestedProfiles')

global __nvmlDevicePowerSmoothingActivatePresetProfile
__nvmlDevicePowerSmoothingActivatePresetProfile = dlsym(RTLD_DEFAULT, 'nvmlDevicePowerSmoothingActivatePresetProfile')
if __nvmlDevicePowerSmoothingActivatePresetProfile == NULL:
Expand Down Expand Up @@ -3459,9 +3403,6 @@ cpdef dict _inspect_function_pointers():
global __nvmlDeviceSetTemperatureThreshold
data["__nvmlDeviceSetTemperatureThreshold"] = <intptr_t>__nvmlDeviceSetTemperatureThreshold

global __nvmlDeviceSetPowerManagementLimit
data["__nvmlDeviceSetPowerManagementLimit"] = <intptr_t>__nvmlDeviceSetPowerManagementLimit

global __nvmlDeviceSetGpuOperationMode
data["__nvmlDeviceSetGpuOperationMode"] = <intptr_t>__nvmlDeviceSetGpuOperationMode

Expand Down Expand Up @@ -3876,27 +3817,9 @@ cpdef dict _inspect_function_pointers():
global __nvmlDeviceGetDeviceHandleFromMigDeviceHandle
data["__nvmlDeviceGetDeviceHandleFromMigDeviceHandle"] = <intptr_t>__nvmlDeviceGetDeviceHandleFromMigDeviceHandle

global __nvmlGpmSampleGet
data["__nvmlGpmSampleGet"] = <intptr_t>__nvmlGpmSampleGet

global __nvmlGpmMigSampleGet
data["__nvmlGpmMigSampleGet"] = <intptr_t>__nvmlGpmMigSampleGet

global __nvmlGpmQueryDeviceSupport
data["__nvmlGpmQueryDeviceSupport"] = <intptr_t>__nvmlGpmQueryDeviceSupport

global __nvmlGpmQueryIfStreamingEnabled
data["__nvmlGpmQueryIfStreamingEnabled"] = <intptr_t>__nvmlGpmQueryIfStreamingEnabled

global __nvmlGpmSetStreamingEnabled
data["__nvmlGpmSetStreamingEnabled"] = <intptr_t>__nvmlGpmSetStreamingEnabled

global __nvmlDeviceGetCapabilities
data["__nvmlDeviceGetCapabilities"] = <intptr_t>__nvmlDeviceGetCapabilities

global __nvmlDeviceWorkloadPowerProfileClearRequestedProfiles
data["__nvmlDeviceWorkloadPowerProfileClearRequestedProfiles"] = <intptr_t>__nvmlDeviceWorkloadPowerProfileClearRequestedProfiles

global __nvmlDevicePowerSmoothingActivatePresetProfile
data["__nvmlDevicePowerSmoothingActivatePresetProfile"] = <intptr_t>__nvmlDevicePowerSmoothingActivatePresetProfile

Expand Down Expand Up @@ -5833,16 +5756,6 @@ cdef nvmlReturn_t _nvmlDeviceSetTemperatureThreshold(nvmlDevice_t device, nvmlTe
device, thresholdType, temp)


cdef nvmlReturn_t _nvmlDeviceSetPowerManagementLimit(nvmlDevice_t device, unsigned int limit) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlDeviceSetPowerManagementLimit
_check_or_init_nvml()
if __nvmlDeviceSetPowerManagementLimit == NULL:
with gil:
raise FunctionNotFoundError("function nvmlDeviceSetPowerManagementLimit is not found")
return (<nvmlReturn_t (*)(nvmlDevice_t, unsigned int) noexcept nogil>__nvmlDeviceSetPowerManagementLimit)(
device, limit)


cdef nvmlReturn_t _nvmlDeviceSetGpuOperationMode(nvmlDevice_t device, nvmlGpuOperationMode_t mode) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlDeviceSetGpuOperationMode
_check_or_init_nvml()
Expand Down Expand Up @@ -7223,56 +7136,6 @@ cdef nvmlReturn_t _nvmlDeviceGetDeviceHandleFromMigDeviceHandle(nvmlDevice_t mig
migDevice, device)


cdef nvmlReturn_t _nvmlGpmSampleGet(nvmlDevice_t device, nvmlGpmSample_t gpmSample) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlGpmSampleGet
_check_or_init_nvml()
if __nvmlGpmSampleGet == NULL:
with gil:
raise FunctionNotFoundError("function nvmlGpmSampleGet is not found")
return (<nvmlReturn_t (*)(nvmlDevice_t, nvmlGpmSample_t) noexcept nogil>__nvmlGpmSampleGet)(
device, gpmSample)


cdef nvmlReturn_t _nvmlGpmMigSampleGet(nvmlDevice_t device, unsigned int gpuInstanceId, nvmlGpmSample_t gpmSample) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlGpmMigSampleGet
_check_or_init_nvml()
if __nvmlGpmMigSampleGet == NULL:
with gil:
raise FunctionNotFoundError("function nvmlGpmMigSampleGet is not found")
return (<nvmlReturn_t (*)(nvmlDevice_t, unsigned int, nvmlGpmSample_t) noexcept nogil>__nvmlGpmMigSampleGet)(
device, gpuInstanceId, gpmSample)


cdef nvmlReturn_t _nvmlGpmQueryDeviceSupport(nvmlDevice_t device, nvmlGpmSupport_t* gpmSupport) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlGpmQueryDeviceSupport
_check_or_init_nvml()
if __nvmlGpmQueryDeviceSupport == NULL:
with gil:
raise FunctionNotFoundError("function nvmlGpmQueryDeviceSupport is not found")
return (<nvmlReturn_t (*)(nvmlDevice_t, nvmlGpmSupport_t*) noexcept nogil>__nvmlGpmQueryDeviceSupport)(
device, gpmSupport)


cdef nvmlReturn_t _nvmlGpmQueryIfStreamingEnabled(nvmlDevice_t device, unsigned int* state) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlGpmQueryIfStreamingEnabled
_check_or_init_nvml()
if __nvmlGpmQueryIfStreamingEnabled == NULL:
with gil:
raise FunctionNotFoundError("function nvmlGpmQueryIfStreamingEnabled is not found")
return (<nvmlReturn_t (*)(nvmlDevice_t, unsigned int*) noexcept nogil>__nvmlGpmQueryIfStreamingEnabled)(
device, state)


cdef nvmlReturn_t _nvmlGpmSetStreamingEnabled(nvmlDevice_t device, unsigned int state) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlGpmSetStreamingEnabled
_check_or_init_nvml()
if __nvmlGpmSetStreamingEnabled == NULL:
with gil:
raise FunctionNotFoundError("function nvmlGpmSetStreamingEnabled is not found")
return (<nvmlReturn_t (*)(nvmlDevice_t, unsigned int) noexcept nogil>__nvmlGpmSetStreamingEnabled)(
device, state)


cdef nvmlReturn_t _nvmlDeviceGetCapabilities(nvmlDevice_t device, nvmlDeviceCapabilities_t* caps) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlDeviceGetCapabilities
_check_or_init_nvml()
Expand All @@ -7283,16 +7146,6 @@ cdef nvmlReturn_t _nvmlDeviceGetCapabilities(nvmlDevice_t device, nvmlDeviceCapa
device, caps)


cdef nvmlReturn_t _nvmlDeviceWorkloadPowerProfileClearRequestedProfiles(nvmlDevice_t device, nvmlWorkloadPowerProfileRequestedProfiles_t* requestedProfiles) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlDeviceWorkloadPowerProfileClearRequestedProfiles
_check_or_init_nvml()
if __nvmlDeviceWorkloadPowerProfileClearRequestedProfiles == NULL:
with gil:
raise FunctionNotFoundError("function nvmlDeviceWorkloadPowerProfileClearRequestedProfiles is not found")
return (<nvmlReturn_t (*)(nvmlDevice_t, nvmlWorkloadPowerProfileRequestedProfiles_t*) noexcept nogil>__nvmlDeviceWorkloadPowerProfileClearRequestedProfiles)(
device, requestedProfiles)


cdef nvmlReturn_t _nvmlDevicePowerSmoothingActivatePresetProfile(nvmlDevice_t device, nvmlPowerSmoothingProfile_t* profile) except?_NVMLRETURN_T_INTERNAL_LOADING_ERROR nogil:
global __nvmlDevicePowerSmoothingActivatePresetProfile
_check_or_init_nvml()
Expand Down
Loading
Loading