kelvin/hpctools
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

first commit

Browse Source
This commit is contained in:
root
2026-04-26 13:25:50 +08:00
commit 6e46c06edd
20 changed files with 17274 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
pkgs/cuda-13.0/include Symbolic link
View File

@@ -0,0 +1 @@
targets/x86_64-linux/include

1
pkgs/cuda-13.0/lib64 Symbolic link
View File

@@ -0,0 +1 @@
targets/x86_64-linux/lib

504
pkgs/cuda-13.0/nvml/doc/nvml_changelog.txt Normal file
View File

@@ -0,0 +1,504 @@
/*! @page KnownIssues Known issues in the current version of NVML library
*
* This is a list of known NVML issues in the current driver:
* - NVML Field Values from #251 - #273 (Power Smoothing, Clock Event Reason, and Sync Power Balancing related field values) have changed between 13.0 and 13.0U1/v580TRD2.
* - Any application that is using these field IDs must be recompiled using the NVML header file from CUDA 13.0 Update 1 in order to continue working correctly with NVIDIA drivers v580 TRD2 and beyond.
* - On systems where GPUs are NUMA nodes, the accuracy of FB memory utilization provided by NVML depends on the memory accounting of the operating system.
* This is because FB memory is managed by the operating system instead of the NVIDIA GPU driver.
* Typically, pages allocated from FB memory are not released even after the process terminates to enhance performance. In scenarios where
* the operating system is under memory pressure, it may resort to utilizing FB memory. Such actions can result in discrepancies in the accuracy of memory reporting.
* - On Linux GPU Reset can't be triggered when there is pending GPU Operation Mode (GOM) change
* - On Linux GPU Reset may not successfully change pending ECC mode. A full reboot may be required to enable the mode change.
* - \ref nvmlAccountingStats supports only one process per GPU at a time (CUDA proxy server counts as one process).
* - \ref nvmlAccountingStats_t.time reports time and utilization values starting from cuInit till process termination. Next driver versions might change this behavior slightly and account process only from cuCtxCreate till cuCtxDestroy.
* - On GPUs from Fermi family current P0 clocks (reported by \ref nvmlDeviceGetClockInfo) can differ from max clocks by few MHz.
*/
/*! @page Changelog Change log of NVML library
* This chapter list changes in API and bug fixes that were introduced to the library
* \section changelog32 Changes between NVML v575 and v580 ===
* - Fixed bug with NVML_FI_PWR_SMOOTHING_* Field Value numbering, which was different than the v570 values.
* - Adjusted NVML_FI_DEV_CLOCKS_EVENT_REASON_* and NVML_FI_DEV_POWER_SYNC_BALANCING_* field value numbering to resolve overlap with NVML_FI_PWR_SMOOTHING_* field values.
*
* - Added \ref nvmlDeviceGetSramUniqueUncorrectedEccErrorCounts to get the counts of SRAM unique uncorrected ECC errors.
* - Deprecated Applications Clocks APIs, which will be removed in CUDA 14.0:
* - \ref nvmlDeviceSetApplicationsClocks
* - \ref nvmlDeviceGetApplicationsClock
* - \ref nvmlDeviceGetDefaultApplicationsClock
* - \ref nvmlDeviceResetApplicationsClocks
* - Deprecated \ref nvmlDeviceGetViolationStatus, which will be removed in CUDA 14.0
* - Added \ref nvmlDeviceGetNvLinkInfo to query device NVLINK info.
* - Added \ref nvmlDeviceGetPdi to retrieve the device GPU PDI.
* - Added Multi-GPU mode NVLINK Encryption \ref NVML_CC_SYSTEM_MULTIGPU_NVLE
* - Added V2 struct to \ref nvmlDeviceGetNvLinkInfo to query NVLINK Firmware info.
* - Added \ref nvmlDeviceReadWritePRM_v1 to retrieve GPU PRM register contents
* - Added \ref nvmlDeviceGetAddressingMode to retrieve the addressing mode for the device.
* - Added \ref nvmlDeviceGetRepairStatus to get ECC status info.
* - Added \ref nvmlDeviceGetGpuInstanceProfileInfoByIdV which allows for MIG GPU instance profile info to be queried with profileId instead of profile name.
* - Updated nvmlGpuFabricInfoV_t to v3 to include a new Health Summary field, and new Incorrect Configuration statuses.
- nvmlGpuFabricInfo_v2_t is deprecated and will be removed in a future release
* - Added \ref nvmlDeviceGetPowerMizerMode_v1 to query the current and supported power mizer modes on Maxwell and newer gpus. Power mizer mode provides a hint to the driver as to how to manage the performance of the GPU.
* - Added \ref nvmlDeviceSetPowerMizerMode_v1 to set the power mizer mode on Maxwell and newer gpus.
* - Added new Incorrect Configuration Statuses to nvmlGpuFabricInfoV_t
* - NVML_GPU_FABRIC_HEALTH_MASK_INCORRECT_CONFIGURATION_INCOMPATIBLE_GPU_FW
* - NVML_GPU_FABRIC_HEALTH_MASK_INCORRECT_CONFIGURATION_INVALID_LOCATION
* - Added \ref nvmlDeviceSetHostname_v1 and \ref nvmlDeviceGetHostname_v1 to allow custom GPU hostname configuration.
* \section changelog31 Changes between NVML v570 and v575 ===
*
* - Added \ref nvmlSystemEventSetCreate to create a system event set.
* - Added \ref nvmlSystemEventSetFree to free a system event set.
* - Added \ref nvmlSystemRegisterEvents to register system events on a system event set.
* - Added \ref nvmlSystemEventSetWait to wait for system event notification and obtain system event data.
* - Added \ref nvmlGpuInstanceGetCreatableVgpus to query the currently creatable vGPU types on the user provided GPU Instance
* - Added \ref nvmlVgpuTypeGetMaxInstancesPerGpuInstance to query the maximum number of vGPU instances per GPU Instance for the given vGPU type
* - Added \ref nvmlGpuInstanceSetVgpuSchedulerState to set the vGPU scheduler state for the given GPU Instance
* - Added \ref nvmlGpuInstanceGetActiveVgpus to query the currently active vGPU instances on the user provided GPU Instance
* - Added \ref nvmlGpuInstanceGetVgpuSchedulerState to query the vGPU software scheduler state for the given GPU Instance.
* - Added \ref nvmlGpuInstanceGetVgpuSchedulerLog to query the vGPU software scheduler logs for the given GPU Instance.
* - Added \ref nvmlGpuInstanceGetVgpuTypeCreatablePlacements to query the creatable vGPU placement IDs of the vGPU type within a GPU instance
* - Added \ref nvmlGpuInstanceSetVgpuHeterogeneousMode to enable or disable vGPU heterogenous mode for the GPU Instance.
* - Added \ref nvmlGpuInstanceGetVgpuHeterogeneousMode to query the vGPU heterogenous mode for the GPU Instance.
* - Updated \ref nvmlDeviceGetVgpuCapabilities to report whether GPU supports timesliced vGPU on MIG and whether MIG timesliced mode is enabled or not vGPU capabilities.
* - Updated \ref nvmlDeviceSetVgpuCapabilities to set the MIG timesliced mode vGPU capability of a device.
* - Updated \ref nvmlDeviceSetVgpuHeterogeneousMode to return \ref NVML_ERROR_NOT_SUPPORTED when in MIG mode.
* - Updated \ref nvmlDeviceGetVgpuHeterogeneousMode to return \ref NVML_ERROR_NOT_SUPPORTED when in MIG mode.
* - Updated \ref nvmlDeviceGetVgpuTypeCreatablePlacements to return \ref NVML_ERROR_NOT_SUPPORTED when in MIG mode.
* - Updated \ref nvmlDeviceGetVgpuSchedulerLog to return \ref NVML_ERROR_NOT_SUPPORTED when in MIG mode.
* - Updated \ref nvmlDeviceSetVgpuSchedulerState to return \ref NVML_ERROR_NOT_SUPPORTED when in MIG mode.
* - Updated \ref nvmlDeviceGetVgpuSchedulerState to return \ref NVML_ERROR_NOT_SUPPORTED when in MIG mode.
* - Added 3 new NVML_FI_DEV_C2C_LINK_ERROR fieldIds
* - \ref NVML_FI_DEV_C2C_LINK_ERROR_INTR
* - \ref NVML_FI_DEV_C2C_LINK_ERROR_REPLAY
* - \ref NVML_FI_DEV_C2C_LINK_ERROR_REPLAY_B2B
* - Added new NVML_FI_DEV_C2C_LINK_POWER_STATE fieldId
* - Added new CTXSW GPM Metrics
* - Added \ref nvmlDeviceGetHandleByUUIDV that supports both the ASCII and binary format UUID to retrieve the device handle.
* - Added 2 new NVML_FI_DEV_POWER_SYNC_BALANCING fieldIds
* - \ref NVML_FI_DEV_POWER_SYNC_BALANCING_FREQ
* - \ref NVML_FI_DEV_POWER_SYNC_BALANCING_AF
* - Added 5 new Clock Event Reason Counters fieldIds
* - \ref NVML_FI_DEV_CLOCKS_EVENT_REASON_SW_POWER_CAP
* - \ref NVML_FI_DEV_CLOCKS_EVENT_REASON_SYNC_BOOST
* - \ref NVML_FI_DEV_CLOCKS_EVENT_REASON_SW_THERM_SLOWDOWN
* - \ref NVML_FI_DEV_CLOCKS_EVENT_REASON_HW_THERM_SLOWDOWN
* - \ref NVML_FI_DEV_CLOCKS_EVENT_REASON_HW_POWER_BRAKE_SLOWDOWN
* - Updated \ref nvmlDeviceGetMemoryErrorCounter to better account for transient vs. permanent errors
* - Added MIG profiles that can allocate all or none of Decoder, Encoder, JPEG and OFA engines.
*
* \section changelog30 Changes between NVML v565 Update and v570 ===
* - Revert the fix for the issue where PCIe throughput (reported via \ref nvmlDeviceGetPcieThroughput and nvidia-smi -q) is 1000 times bigger than its actual value
* - Added field values for data related to Power Smoothing
* - Added \ref nvmlDevicePowerSmoothingActivatePresetProfile to activate a specific Preset Profile for Power Smoothing
* - Added \ref nvmlDevicePowerSmoothingSetState to enable/disable the Power Smoothing feature
* - Added \ref nvmlDevicePowerSmoothingUpdatePresetProfileParam to update parameters to preset profiles for Power Smoothing
* - Added new enums for fieldId NVML_FI_DEV_NVLINK_GET_STATE to expose INACTIVE, ACTIVE, and SLEEP state for a link
* - Added \ref nvmlDeviceGetMarginTemperature to retrieve the thermal margin temperature (distance to nearest slowdown threshold).
* - Added \ref nvmlDeviceGetNvlinkSupportedBwModes to get all supported Nvlink Bandwidth modes
* - Added \ref nvmlDeviceGetNvlinkBwMode to get the current Nvlink Bandwidth mode
* - Added \ref nvmlDeviceSetNvlinkBwMode to set the Nvlink Bandwidth mode
* - Added MIG profiles with support for graphics.
* - Added support for new recovery action - NVML_GPU_RECOVERY_ACTION_DRAIN_AND_RESET
* - Deprecated nvml fieldIds NVML_FI_DEV_RESET_STATUS and NVML_FI_DEV_DRAIN_AND_RESET_STATUS. Usee NVML_FI_DEV_GET_GPU_RECOVERY_ACTION instead
* - Added \ref nvmlDeviceGetDramEncryptionMode and \ref nvmlDeviceSetDramEncryptionMode to query and configure DRAM Encryption Mode
* - Added 3 new flags to GPU Fabric Health Mask
* - NVML_GPU_FABRIC_HEALTH_MASK_SHIFT_ROUTE_RECOVERY
* - NVML_GPU_FABRIC_HEALTH_MASK_SHIFT_ROUTE_UNHEALTHY
* - NVML_GPU_FABRIC_HEALTH_MASK_SHIFT_ACCESS_TIMEOUT_RECOVERY
* - Added 4 new GPM metrics
* - NVML_GPM_METRIC_NVENC_0_UTIL
* - NVML_GPM_METRIC_NVENC_1_UTIL
* - NVML_GPM_METRIC_NVENC_2_UTIL
* - NVML_GPM_METRIC_NVENC_3_UTIL
* - Added new counters for Nvlink5
* - \ref NVML_FI_DEV_NVLINK_COUNT_EFFECTIVE_ERRORS to get sum of the number of errors in each Nvlink packet
* - \ref NVML_FI_DEV_NVLINK_COUNT_EFFECTIVE_BER to get Effective BER for effective errors
* - \ref NVML_FI_DEV_NVLINK_COUNT_FEC_HISTORY_0 to 15 to get count of symbol errors that are corrected
* - Swapped the values of field IDs \ref NVML_FI_DEV_IS_MIG_MODE_INDEPENDENT_MIG_QUERY_CAPABLE and \ref NVML_FI_DEV_NVLINK_GET_POWER_THRESHOLD_MAX to fix backwards compatibility with v550.
* - New revision of nvmlPlatformInfo_t -- nvmlPlatformInfo_v2 has been added. In this version the following fields from v1 have been renamed
* - rackGuid to chassisSerialNumber
* - chassisPhysicalSlotNumber to slotNumber
* - computeSlotIndex to trayIndex
* - nodeIndex to hostId
* - nvmlPlatformInfo_v1 is deprecated and will be removed in subsequent releases
*
* \section changelog29 Changes between NVML v560 Update and v565 ===
* - Fixed the ECC error count mismatch between nvidia-smi query output and NVML APIs, \ref nvmlDeviceGetMemoryErrorCounter and \ref nvmlDeviceGetFieldValues.
* - Added new value NVML_CC_SYSTEM_CPU_CAPS_AMD_SNP_VTOM for CC CPU capability reporting
* - Added \ref nvmlDeviceGetCoolerInfo to retrieve a cooler's control signal characteristics and target that cooler cools.
* - Added new value NVML_CC_SYSTEM_CPU_CAPS_AMD_SEV_SNP for CC CPU capability reporting
* - Added \ref nvmlDeviceGetFanSpeedRPM to report the intended operating speed in rotations per minute (RPM) of the device's specified fan.
* - Added \ref nvmlDeviceGetPerformanceModes to retrieve a performance modes string with all the performance modes defined for this device along with their associated GPU Clock and Memory Clock values.
* - Added \ref nvmlDeviceGetCurrentClockFreqs to retrieve a string with the associated GPU Clock and Memory Clock values for the current pstate.
* - Added \ref nvmlNvlinkVersion_t enum to define NvLink Version
* - Added \ref nvmlDeviceGetPlatformInfo to retrieve the platform information of a device
* - Added new event type nvmlEventTypeGpuUnavailableError
* - Removed support for \p nvmlDeviceGetNvLinkCrcLaneErrorCounter \p nvmlDeviceGetNvLinkEccLaneErrorCounter \p nvmlDeviceGetNvLinkErrorCounter on Blackwell
* - Removed support for fieldIds \ref NVML_FI_DEV_NVLINK_ERROR_DL_REPLAY \ref NVML_FI_DEV_NVLINK_ERROR_DL_RECOVERY \ref NVML_FI_DEV_NVLINK_ERROR_DL_CRC on Blackwell
* - Added \ref nvmlVgpuInstanceGetRuntimeStateSize to get the vGPU runtime state size
* - Updated nvmlDeviceGetVgpuTypeSupportedPlacements function to report both Heterogeneous and Homogeneous vGPU placements.
* - Updated nvmlDeviceGetVgpuCapabilities to report the Homogeneous vGPU capability.
* - Added \ref nvmlDeviceWorkloadPowerProfileGetProfilesInfo to retrieve Workload Power Profile Info
* - Added \ref nvmlDeviceWorkloadPowerProfileGetCurrentProfiles to retrieve current Requested and Enforced Workload Power Profiles
* - Added \ref nvmlDeviceWorkloadPowerProfileSetRequestedProfiles to set Requested Workload Power Profiles
* - Added \ref nvmlDeviceWorkloadPowerProfileClearRequestedProfiles to clear Requested Performance Profiles
* - Added new event type nvmlEventTypeGpuRecoveryAction
* - Added new fieldId to query gpu recovery action NVML_FI_DEV_GET_GPU_RECOVERY_ACTION
* - Deprecated fieldIds
* - \ref NVML_FI_DEV_NVLINK_COUNT_VL15_DROPPED to get Number of VL15 MADs dropped on a link in NVLink5
* - \ref NVML_FI_DEV_NVLINK_COUNT_RAW_BER_LANE0 to get BER per lane for lane 0
* - \ref NVML_FI_DEV_NVLINK_COUNT_RAW_BER_LANE1 to get BER per lane for lane 1
* - \ref NVML_FI_DEV_NVLINK_COUNT_RAW_BER to get BER per link. Sum of all the raw errors per lane/Bits received per link
* - \ref NVML_FI_DEV_NVLINK_COUNT_EFFECTIVE_ERRORS to get Sum of the number of errors in each Nvlink packet
* - \ref NVML_FI_DEV_NVLINK_COUNT_EFFECTIVE_BER to get Effective BER for effective errors
*
* \section changelog28 Changes between NVML v555 Update and v560 ===
*
* - Added field values NVML_FI_DEV_PCIE_OUTBOUND_ATOMICS_MASK and NVML_FI_DEV_PCIE_INBOUND_ATOMICS_MASK for nvmlDeviceGetFieldValues.
* - Added field ids NVML_FI_DEV_RESET_STATUS and NVML_FI_DEV_DRAIN_AND_RESET_STATUS which correspond to the nvidia-smi output.
* - Added NVML_DEVICE_ARCH_T23X architecture type.
* - Added \ref nvmlVgpuTypeGetBAR1Info to query the BAR1 information of a vGPU type.
* - Added new event types, nvmlEventTypeSingleBitEccErrorStorm, nvmlEventTypeDramRetirementEvent, nvmlEventTypeDramRetirementFailure, nvmlEventTypeNonFatalPoisonError and nvmlEventTypeFatalPoisonError.
* - Added \ref nvmlSystemGetDriverBranch to query the driver branch information.
*
* \section changelog27 Changes between NVML v550 Update and v555 ===
*
* - Added \ref nvmlDeviceGetClockOffsets to query min, max and current clock offset value on a Maxwell and later GPU for a specified clock.
* - Added \ref nvmlDeviceSetClockOffsets to control clock offset value on a Maxwell and later GPU for a specified clock.
* - Added new fieldIds for Nvlink5 telemetry on Blackwell
* - \ref NVML_FI_DEV_NVLINK_COUNT_XMIT_PACKETS to get Total Tx packets on the link in NVLink5
* - \ref NVML_FI_DEV_NVLINK_COUNT_XMIT_BYTES to get Total Tx bytes on the link in NVLink5
* - \ref NVML_FI_DEV_NVLINK_COUNT_RCV_PACKETS to get Total Rx packets on the link in NVLink5
* - \ref NVML_FI_DEV_NVLINK_COUNT_RCV_BYTES to get Total Rx bytes on the link in NVLink5
* - \ref NVML_FI_DEV_NVLINK_COUNT_VL15_DROPPED to get Number of VL15 MADs dropped on a link in NVLink5
* - \ref NVML_FI_DEV_NVLINK_COUNT_MALFORMED_PACKET_ERRORS to get Number of packets Rx on a link where packets are malformed
* - \ref NVML_FI_DEV_NVLINK_COUNT_BUFFER_OVERRUN_ERRORS to get Number of packets that were discarded on Rx due to buffer overrun
* - \ref NVML_FI_DEV_NVLINK_COUNT_RCV_ERRORS to get Total number of packets with errors Rx on a link
* - \ref NVML_FI_DEV_NVLINK_COUNT_RCV_REMOTE_ERRORS to get Total number of packets Rx - stomp/EBP marker
* - \ref NVML_FI_DEV_NVLINK_COUNT_RCV_GENERAL_ERRORS to get Total number of packets Rx with header mismatch
* - \ref NVML_FI_DEV_NVLINK_COUNT_LOCAL_LINK_INTEGRITY_ERRORS to get Total number of times that the count of local errors exceeded a threshold
* - \ref NVML_FI_DEV_NVLINK_COUNT_XMIT_DISCARDS to get Total number of tx error packets that were discarded
* - \ref NVML_FI_DEV_NVLINK_COUNT_LINK_RECOVERY_SUCCESSFUL_EVENTS to get Number of times link went from Up to recovery, succeeded and link came back up
* - \ref NVML_FI_DEV_NVLINK_COUNT_LINK_RECOVERY_FAILED_EVENTS to get Number of times link went from Up to recovery, failed and link was declared down
* - \ref NVML_FI_DEV_NVLINK_COUNT_LINK_RECOVERY_EVENTS to get Number of times link went from Up to recovery, irrespective of the result
* - \ref NVML_FI_DEV_NVLINK_COUNT_RAW_BER_LANE0 to get BER per lane for lane 0
* - \ref NVML_FI_DEV_NVLINK_COUNT_RAW_BER_LANE1 to get BER per lane for lane 1
* - \ref NVML_FI_DEV_NVLINK_COUNT_RAW_BER to get BER per link. Sum of all the raw errors per lane/Bits received per link
* - \ref NVML_FI_DEV_NVLINK_COUNT_EFFECTIVE_ERRORS to get Sum of the number of errors in each Nvlink packet
* - \ref NVML_FI_DEV_NVLINK_COUNT_EFFECTIVE_BER to get Effective BER for effective errors
* - \ref NVML_FI_DEV_NVLINK_COUNT_SYMBOL_ERRORS to get Number of errors in rx symbols
* - \ref NVML_FI_DEV_NVLINK_COUNT_SYMBOL_BER to get BER for symbol errors
* - Added two new field ids NVML_FI_DEV_PCIE_COUNT_TX_BYTES and NVML_FI_DEV_PCIE_COUNT_RX_BYTES for nvmlDeviceGetFieldValues.
* - Added new API nvmlDeviceGetCapabilities with the first capability bit NVML_DEV_CAP_EGM for Extended GPU Memory (EGM) capability.
* - Added multiGpuMode display on CC enabled system via new API nvmlSystemGetConfComputeSettings or "nvidia-smi conf-compute --get-multigpu-mode" or "nvidia-smi conf-compute -mgm".
* - Added new field ID \ref NVML_FI_DEV_NVLINK_GET_POWER_THRESHOLD_MAX to get the Max Nvlink Power Threshold for a device.
* - Deprecated \ref nvmlDeviceGetTemperature and replaced with a new API \ref nvmlDeviceGetTemperatureV to retrieve device temperature.
* - Added new field ID \ref NVML_VGPU_DRIVER_CAP_WARM_UPDATE and NVML_DEVICE_VGPU_CAP_WARM_UPDATE to query whether the driver and the device supports FSR and warm update of vGPU host driver without terminating the running guest VM respectively.
* - Added new field ID \ref NVML_FI_DEV_NVLINK_GET_POWER_THRESHOLD_MIN to get the Min Nvlink Power Threshold for a device.
* - Added new field ID \ref NVML_FI_DEV_NVLINK_GET_POWER_THRESHOLD_UNITS to get the Units of the Nvlink Power Threshold for a device.
* - Added new field ID \ref NVML_FI_DEV_NVLINK_GET_POWER_THRESHOLD_SUPPORTED to get if Nvlink Power Threshold is supported for a device.
*
*
* \section changelog26 Changes between NVML v545 Update and v550 ===
*
* - Added \ref nvmlDeviceGetNumaNodeId to query the NUMA node of a GPU.
* - Fix the issue where PCIe throughput (reported via nvmlDeviceGetPcieThroughput and nvidia-smi -q) is 1000 times bigger than its actual value.
* - Added new GPM metric Id NVML_GPM_METRIC_NVOFA_1_UTIL to \ref nvmlGpmMetricId_t.
* - Added new field ID \ref NVML_FI_DEV_IS_MIG_MODE_INDEPENDENT_MIG_QUERY_CAPABLE, to check MIG query capable device irrespective of MIG mode.
* - Deprecated NVML_P2P_CAPS_INDEX_PROP and added NVML_P2P_CAPS_INDEX_PCI to reflect the same P2P capability.
* - Added \ref nvmlDeviceGetProcessesUtilizationInfo to retrieve the recent utilization and process ID for all running processes.
* - Added new struct \ref nvmlProcessesUtilizationInfo_v1_t, which includes the new utilization of NVJPG and NVOFA.
* - Added \ref nvmlDeviceGetVgpuInstancesUtilizationInfo to retrieve the recent utilization for vGPU instances running on a physical GPU.
* - Added \ref nvmlDeviceGetVgpuProcessesUtilizationInfo to retrieve the recent utilization for processes running on vGPU instances on a physical GPU.
* - Added \ref nvmlDeviceSetVgpuHeterogeneousMode to enable or disable vGPU heterogenous mode for the device.
* - Added \ref nvmlDeviceGetVgpuHeterogeneousMode to query the vGPU heterogenous mode for the device.
* - Added \ref nvmlVgpuInstanceGetPlacementId to query placement ID of the active vGPU instance.
* - Added \ref nvmlDeviceGetVgpuTypeSupportedPlacements to query the supported vGPU placement IDs of a vGPU type.
* - Added \ref nvmlDeviceGetVgpuTypeCreatablePlacements to query the creatable vGPU placement IDs of a vGPU type.
* - Added support to display confidential compute protected memory along with fb & bar1 in nvidia-smi pmon & dmon commands.
* - Added \ref nvmlDeviceGetGpuFabricInfoV to query Gpu Fabric Probe Info for the device.
* - Deprecated \ref nvmlDeviceGetGpuFabricInfo. This function should not be used, and will be removed in a future release. Use \ref nvmlDeviceGetGpuFabricInfoV instead.
* - Modified \ref nvmlDeviceGetGpuInstanceProfileInfo and \ref nvmlDeviceGetGpuInstancePossiblePlacements_v2 to no longer require MIG being enabled
* - Added \ref nvmlSystemSetConfComputeKeyRotationThresholdInfo to set confidential compute key rotation threshold.
* - Added \ref nvmlSystemGetConfComputeKeyRotationThresholdInfo to query confidential compute key rotation threshold detail.
* - Added \ref nvmlDeviceSetVgpuCapabilities to set the desirable vGPU capability of a device.
*
*
* \section changelog25 Changes between NVML v535 Update and v545 ===
*
* - Added a new error code \ref NVML_ERROR_GPU_NOT_FOUND to be returned if no supported GPUS are found during initialization.
* - In \ref nvmlGpuFabricInfo_t \p partitionId has been renamed to \p cliqueId.
* - Added new versioned structs \ref nvmlGpuInstanceProfileInfo_v3_t and \ref nvmlComputeInstanceProfileInfo_v3_t.
* - Added \ref nvmlDeviceGetLastBBXFlushTime for retrieving the timestamp and duration of the latest flush of the BBX object to the inforom storage.
* - Added \ref NVML_POWER_SCOPE_MEMORY to report out power usage for GPU Memory.
* - Added \ref nvmlDeviceGetPciInfoExt which expands \ref nvmlDeviceGetPciInfo_v3 to also report PCI base and sub classcodes.
* - Added new struct \ref nvmlPciInfoExt_v1_t, which is used in \ref nvmlDeviceGetPciInfoExt.
* - Added \ref nvmlDeviceGetRunningProcessDetailList api to get information about Compute, Graphics or MPS-Compute processes running on a GPU with protected memory usage info.
*
*
* \section changelog24 Changes between NVML v530 Update and v535 ===
*
* - Fixed \ref nvmlDeviceGetMemoryErrorCounter and \ref nvmlDeviceGetFieldValues to return correct SRAM volatile total error counts.
* - Added \ref nvmlDeviceGetSramEccErrorStatus to query SRAM ECC error status for the device.
* - Added \ref nvmlDeviceGetModuleId for getting device module id
* - Updated \ref nvmlDeviceGetPowerSource API to report undersized power source.
* - Added \ref nvmlDeviceGetJpgUtilization and \ref nvmlDeviceGetOfaUtilization APIs
* - Added \ref nvmlSystemGetNvlinkBwMode and \ref nvmlSystemSetNvlinkBwMode APIs
* - Added \ref nvmlDeviceSetVgpuSchedulerState to set the vGPU scheduler state.
* - Added new field ID \ref NVML_FI_DEV_IS_RESETLESS_MIG_SUPPORTED for device's resetless MIG capability
* - Added \ref nvmlDeviceGetComputeRunningProcesses_v3 to get information about Compute processes running on a GPU.
* - Added \ref nvmlDeviceGetGraphicsRunningProcesses_v3 to get information about Graphics processes running on a GPU.
* - Added \ref nvmlDeviceGetMPSComputeRunningProcesses_v3 to get information about MPS-Compute processes running on a GPU.
* - Added \ref nvmlDeviceGetRunningProcessDetailList to get information about Compute, Graphics or MPS-Compute processes running on a GPU with protected memory usage info. Currently returns NVML_ERROR_NOT_SUPPORTED. Functionality will be implemented in next release.
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_CORRECTABLE_ERRORS for PCIe correctable errors counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_NAKS_RECEIVED for PCIe NAK Receive counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_RECEIVER_ERROR for PCIe receiver error counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_BAD_TLP for PCIe bad TLP counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_NAKS_SENT for NAK Send counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_BAD_DLLP for PCIe bad DLLP counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_NON_FATAL_ERROR for PCIe non fatal error counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_FATAL_ERROR for PCIe fatal error counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_UNSUPPORTED_REQ for PCIe unsupported request counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_LCRC_ERROR for PCIe LCRC error counter
* - Added new field ID \ref NVML_FI_DEV_PCIE_COUNT_LANE_ERROR for per lane error counter with scope as PCIe lane number.
* - Added \ref nvmlDeviceGetPowerUsage to retrieve current power usage
* - Added \ref nvmlDeviceGetTotalEnergyConsumption to get current energy consumption
* - Added \ref nvmlDeviceSetPowerManagementLimit_v2 to set the power limit
* - Renamed nvmlDeviceCcuGetStreamState to nvmlGpmQueryIfStreamingEnabled and nvmlDeviceCcuSetStreamState to nvmlGpmSetStreamingEnabled.
*
*
* \section changelog23 Changes between NVML v525 Update and v530 ===
*
* - Fixed a typo in nvmlGpuP2PStatus_t: added a new enum entry for NVML_P2P_STATUS_CHIPSET_NOT_SUPPORTED with the same numeric value as the existing erroneous entry ("NVML_P2P_STATUS_CHIPSET_NOT_SUPPORED")
* - Added \ref nvmlDeviceGetVgpuSchedulerLog to fetch the vGPU software scheduler logs.
* - Added \ref nvmlDeviceGetVgpuSchedulerState to fetch the vGPU software scheduler state.
* - Added \ref nvmlDeviceGetVgpuSchedulerCapabilities to fetch the vGPU software scheduler capabilities.
*
*
* \section changelog22 Changes between NVML v520 Update and v525 ===
*
* - Added \ref nvmlDeviceSetNvLinkDeviceLowPowerThreshold to set the NvLink low power threshold.
* - Added \p nvmlDeviceGetPcieAtomicCaps to report PCIe atomic capabilities.
* - Added \p nvmlDeviceCcuGetStreamState API to report the counter collection unit stream state.
* - Added \p nvmlDeviceCcuSetStreamState API to set the counter collection unit stream state.
* - Removed support for NVML_FI_DEV_LINK_SPEED_MBPS_L{0..} field Ids in Hopper. Replaced with NVML_FI_DEV_NVLINK_GET_SPEED with scope as link Id.
* - Removed support for NVML_FI_DEV_NVLINK_CRC_FLIT_ERROR_COUNT{0..} field Ids in Hopper. Replaced with NVML_FI_DEV_NVLINK_ERROR_DL_CRC with scope as link Id.
* - Removed support for NVML_FI_DEV_NVLINK_REPLAY_ERROR_COUNT_L{0..} field Ids in Hopper. Replaced with NVML_FI_DEV_NVLINK_ERROR_DL_REPLAY with scope as link Id.
* - Removed support for NVML_FI_DEV_NVLINK_RECOVERY_ERROR_COUNT_{0..} field Ids in Hopper. Replaced with NVML_FI_DEV_NVLINK_ERROR_DL_RECOVERY with scope as link Id.
* - Added new field ID \ref NVML_FI_DEV_NVLINK_GET_STATE to get nvlink state
* - Added new field ID \ref NVML_FI_DEV_NVLINK_GET_VERSION to get nvlink version
* - Added new field ID \ref NVML_FI_DEV_C2C_LINK_COUNT to get C2C link count
* - Added new field ID \ref NVML_FI_DEV_C2C_LINK_GET_STATUS to get C2C link status
* - Added new field ID \ref NVML_FI_DEV_C2C_LINK_GET_MAX_BW to get C2C link bandwidth
*
*
* \section changelog21 Changes between NVML v515 Update and v520 ===
*
* - Added \ref nvmlDeviceGetMemClkVfOffset API to report the MemClk VF offset value.
* - Added \ref nvmlDeviceGetMemClkMinMaxVfOffset API to report the Memory clock min and max VF offset that user can set for a specified GPU.
* - Added \ref nvmlDeviceGetGpcClkMinMaxVfOffset API to report the Graphics clock min and max VF offset that user can set for a specified GPU.
* - Added \ref nvmlGpmMetricsGet to calculate GPM metrics from two GPM samples
* - Added \ref nvmlGpmSampleFree to free allocated GPM sample
* - Added \ref nvmlGpmSampleAlloc to allocate a GPM sample
* - Added \ref nvmlGpmSampleGet to retrieve a GPM snapshot
* - Added \ref nvmlGpmQueryDeviceSupport to query whether a device supports GPM
* - Added \ref nvmlDeviceGetFanControlPolicy_v2 API to report the control policy for a specified GPU fan.
* - Added \ref nvmlDeviceSetFanControlPolicy API to set the control policy for a specified GPU fan.
*
*
* \section changelog20 Changes between NVML v510 Update and v515 ===
*
* - Added \ref nvmlDeviceGetMinMaxClockOfPState API to report the min and max clocks of some clock domain for a given PState.
* - Added \ref nvmlDeviceGetSupportedPerformanceStates API to get all supported Performance States (P-States) for the GPU.
* - Added \ref nvmlDeviceGetGpcClkVfOffset API to report the GPCCLK VF offset value.
* - Added \ref nvmlDeviceGetMinMaxFanSpeed API to report the min and max fan speed that user can set for a specified GPU fan.
*
*
* \section changelog19 Changes between NVML v495 Update and v510 ===
*
* - Added \ref nvmlDeviceGetGpuInstanceProfileInfoV and \ref nvmlGpuInstanceGetComputeInstanceProfileInfoV APIs to include the profile name in their output.
* - Added \ref nvmlDeviceGetMemoryBusWidth API to report the GPU's Memory Bus Width.
* - Added \ref nvmlDeviceGetPcieLinkMaxSpeed API to report the GPU's PCIe Max Speed.
* - Added \ref nvmlDeviceGetPowerSource API to report the GPU's power source as AC or battery.
* - Added \ref nvmlDeviceGetNumFans API to report the GPU's number of fans.
* - Added \ref nvmlDeviceGetNumGpuCores API to report the GPU's number of cores.
* - Added \ref nvmlDeviceGetMemoryInfo_v2. The new version accounts separately for system-reserved memory, and includes it in the used memory amount. The previous version of the API reduced the total memory amount by the amount of system-reserved memory.
* - Added \ref nvmlDeviceGetAdaptiveClockInfoStatus API to report the status of adaptive clocking for the GPU.
*
*
* \section changelog18 Changes between NVML v465 Update and v470 ===
*
* - Added new MIG GPU instance profile NVML_GPU_INSTANCE_PROFILE_1_SLICE_REV1.
* - Added \ref nvmlDeviceGetGpuInstancePossiblePlacements_v2. The previous version of the API will not support the profiles with possible placements greater than its total capacity, such as NVML_GPU_INSTANCE_PROFILE_1_SLICE_REV1.
*
*
* \section changelog17 Changes between NVML v460 Update and v465 ===
*
* - Added new NVML_BRAND_* enumeration values for NVIDIA, NVIDIA_RTX, GEFORCE_RTX, QUADRO_RTX and TITAN_RTX
* - Updated \ref nvmlDeviceGetHandleByUUID to make it MIG-aware.
* - Updated \ref nvmlDeviceGetUUID to return MIG UUIDs in the canonical format, 'MIG-UUID'.
* - Updated \ref nvmlDeviceGetHandleByUUID to accept both UUID formats, 'MIG-UUID' and 'MIG-GPU UUID/GID/CID'.
* - The \ref nvmlDeviceSetAPIRestriction and \ref nvmlDeviceGetAPIRestriction APIs would no longer support the ability to toggle root-only requirement for \ref nvmlDeviceSetApplicationsClocks and \ref nvmlDeviceResetApplicationsClocks.
*
*
* \section changelog16 Changes between NVML v450 Update and v460 ===
*
* - Added \ref nvmlDeviceCreateGpuInstanceWithPlacement to allow placement specification when creating a new MIG GPU instance.
*
*
* \section changelog15 Changes between NVML v445 Update and v450 ===
*
* - Updated \ref nvmlDeviceGetFanSpeed and \ref nvmlDeviceGetFanSpeed_v2 for allowing fan speeds greater than 100% to be reported.
* - Added \ref nvmlDeviceGetCpuAffinityWithinScope to determine the closest processor(s) within a NUMA node or socket.
* - Added \ref nvmlDeviceGetMemoryAffinity to determine the closest NUMA node(s) within a NUMA node or socket.
* - Added support to query and disable MIG mode on Windows.
*
*
* \section changelog14 Changes between NVML v418 Update and v445 ===
*
* - Added support for NVIDIA Ampere architecture.
* - Added support for Multi Instance GPU management. Refer "Multi Instance GPU Management" section for details.
*
*
* \section changelog13 Changes between NVML v361 Update and v418
*
* - Support for Volta and Turing architectures, bug fixes, performance improvements, and new features
*
*
* \section changelog12 Changes between NVML v349 Update and v361
*
* - Added \ref nvmlDeviceGetBoardPartNumber to return GPU part numbers
* - Removed support for exclusive thread compute mode (Deprecated in 7.5)
* - Added NVML_CLOCK_VIDEO (encoder/decoder) clock type as a supported clock type for \ref nvmlDeviceGetClockInfo and \ref nvmlDeviceGetMaxClockInfo.
*
*
* \section changelog11 Changes between NVML v346 Update and v349
*
* The following new functionality is exposed on NVIDIA display drivers version 349 Production or later
* - Updated \ref nvmlDeviceGetMemoryInfo to report Used/Free memory under Windows WDDM mode
* - Added \ref nvmlDeviceGetTopologyCommonAncestor to find the common path between two devices
* - Added \ref nvmlDeviceGetTopologyNearestGpus to get a set of GPUs given a path level
* - Added \ref nvmlSystemGetTopologyGpuSet to retrieve a set of GPUs with a given CPU affinity
* - Updated \ref nvmlDeviceGetAccountingPids, \ref nvmlDeviceGetAccountingBufferSize and \ref nvmlDeviceGetAccountingStats to report accounting information for both active and terminated processes. The execution time field in \ref nvmlAccountingStats_t structure is populated only when the process is terminated.
*
*
* \section changelog10 Changes between NVML v340 Update and v346
*
* The following new functionality is exposed on NVIDIA display drivers version 346 Production or later
* - added the public APIs nvmlDeviceGetPcieReplayCounter and nvmlDeviceGetPcieThroughput
* - Discontinued Perl bindings support
* - Added \p nvmlDeviceGetGraphicsRunningProcesses_v2 to get information about Graphics processes running on a GPU.
*
*
* \section changelog9 Changes between NVML v331 Update and v340
*
* The following new functionality is exposed on NVIDIA display drivers version 340 Production or later
* - Added \ref nvmlDeviceGetSamples to get recent power, utilization and clock samples for the GPU.
* - Added \ref nvmlDeviceGetTemperatureThreshold to retrieve temperature threshold information.
* - Added \ref nvmlDeviceGetBrand to retrieve brand information (e.g. Tesla, Quadro, etc.)
* - Added support for K40d and K80
* - Added nvmlDeviceGetTopology internal API to retrieve path info between PCI devices (remove this for DITA)
* - Added \ref nvmlDeviceGetViolationStatus to get the duration of time during which the device was throttled (lower than requested clocks) due to thermal or power constraints.
* - Added \ref nvmlDeviceGetEncoderUtilization and \ref nvmlDeviceGetDecoderUtilization APIs
* - Added \ref nvmlDeviceGetCpuAffinity to determine the closest processor(s) affinity to a specific GPU
* - Added \ref nvmlDeviceSetCpuAffinity to bind a specific GPU to the closest processor
* - Added \ref nvmlDeviceClearCpuAffinity to unbind a specific GPU
* - Added \ref nvmlDeviceGetBoardId to get a unique boardId for the running system
* - Added \ref nvmlDeviceGetMultiGpuBoard to get whether the device is on a multiGPU board
* - Added \ref nvmlDeviceGetAutoBoostedClocksEnabled and nvmlDeviceSetAutoBoostedClocksEnabled for querying and setting the state of auto boosted clocks on supporting hardware.
* - Added \ref nvmlDeviceSetDefaultAutoBoostedClocksEnabled for setting the default state of auto boosted clocks on supporting hardware.
*
*
* \section changelog8 Changes between NVML v5.319 Update and v331
*
* The following new functionality is exposed on NVIDIA display drivers version 331 Production or later
* - Added \ref nvmlDeviceGetMinorNumber to get the minor number for the device.
* - Added \ref nvmlDeviceGetBAR1MemoryInfo to get BAR1 total, available and used memory size.
* - Added \ref nvmlDeviceGetBridgeChipInfo to get the information related to bridge chip firmware.
* - Added enforced power limit query API \ref nvmlDeviceGetEnforcedPowerLimit
* - Updated \ref nvmlEventSetWait_v2 to return xid event data in case of xid error event.
* - Added support for K8
*
* \section changelog7 Changes between NVML v5.319 RC and v5.319 Update
*
* The following new functionality is exposed on NVIDIA display drivers version 319 Update or later
*
* - Added \ref nvmlDeviceSetAPIRestriction and \ref nvmlDeviceGetAPIRestriction, with initial ability to toggle root-only requirement for \ref nvmlDeviceSetApplicationsClocks and \ref nvmlDeviceResetApplicationsClocks.
*
* \section changelog6 Changes between NVML v4.304 and v5.319 RC
*
* The following new functionality is exposed on NVIDIA display drivers version 319 Production or later
*
* - IMPORTANT: Added _v2 versions of \ref nvmlDeviceGetHandleByIndex_v2 and \ref nvmlDeviceGetCount_v2 that also count devices not accessible by current user
* - IMPORTANT: nvmlDeviceGetHandleByIndex_v2 (default) can also return NVML_ERROR_NO_PERMISSION
* - Added nvmlInit_v2 and nvmlDeviceGetHandleByIndex_v2 that is safer and thus recommended function for initializing the library
* - nvmlInit_v2 lazily initializes only requested devices (queried with nvmlDeviceGetHandle*)
* - nvml.h defines nvmlInit_v2 and nvmlDeviceGetHandleByIndex_v2 as default functions
* - Added \ref nvmlDeviceGetIndex
* - Added \ref NVML_ERROR_GPU_IS_LOST to report GPUs that have fallen off the bus.
* - Note: All NVML device APIs can return this error code, as a GPU can fall off the bus at any time.
* - Added new class of APIs for gathering process statistics (\ref nvmlAccountingStats)
* - Application Clocks are no longer supported on GPU's from Quadro product line
* - Added APIs to support dynamic page retirement. See \ref nvmlDeviceGetRetiredPages and
* \ref nvmlDeviceGetRetiredPagesPendingStatus
* - Renamed nvmlClocksThrottleReasonUserDefinedClocks to nvmlClocksThrottleReasonApplicationsClocksSetting. Old name is deprecated and can be removed in one of the next major releases.
* - Added \ref nvmlDeviceGetDisplayActive and updated documentation to clarify how it differs from \ref nvmlDeviceGetDisplayMode
*
* \section changelog5 Changes between NVML v4.304 RC and v4.304 Production
*
* The following new functionality is exposed on NVIDIA display drivers version 304 Production or later
*
* - Added \ref nvmlDeviceGetGpuOperationMode and \ref nvmlDeviceSetGpuOperationMode
*
* \section changelog4 Changes between NVML v3.295 and v4.304 RC
*
* The following new functionality is exposed on NVIDIA display drivers version 304 RC or later
*
* - Added \ref nvmlDeviceGetInforomConfigurationChecksum and \ref nvmlDeviceValidateInforom
* - Added new error return value for initialization failure due to kernel module not receiving interrupts
* - Added \ref nvmlDeviceSetApplicationsClocks, \ref nvmlDeviceGetApplicationsClock, \ref nvmlDeviceResetApplicationsClocks
* - Added \ref nvmlDeviceGetSupportedMemoryClocks and \ref nvmlDeviceGetSupportedGraphicsClocks
* - Added \ref nvmlDeviceGetPowerManagementLimitConstraints, \ref nvmlDeviceGetPowerManagementDefaultLimit and \ref nvmlDeviceSetPowerManagementLimit
* - Added \ref nvmlDeviceGetInforomImageVersion
* - Expanded \ref nvmlDeviceGetUUID to support all CUDA capable GPUs
* - Deprecated \ref nvmlDeviceGetDetailedEccErrors in favor of \ref nvmlDeviceGetMemoryErrorCounter
* - Added \ref NVML_MEMORY_LOCATION_TEXTURE_MEMORY to support reporting of texture memory error counters
* - Added \ref nvmlDeviceGetCurrentClocksThrottleReasons and \ref nvmlDeviceGetSupportedClocksThrottleReasons
* - \ref NVML_CLOCK_SM is now also reported on supported Kepler devices.
* - Dropped support for GT200 based Tesla brand GPUs: C1060, M1060, S1070
*
* \section changelog3 Changes between NVML v2.285 and v3.295
*
* The following new functionality is exposed on NVIDIA display drivers version 295 or later
*
* - deprecated \ref nvmlDeviceGetHandleBySerial in favor of newly added \ref nvmlDeviceGetHandleByUUID
* - Marked the input parameters of \ref nvmlDeviceGetHandleBySerial, \ref nvmlDeviceGetHandleByUUID and \ref nvmlDeviceGetHandleByPciBusId_v2 as const
* - Added \ref nvmlDeviceOnSameBoard
* - Added \ref nvmlConstants defines
* - Added \ref nvmlDeviceGetMaxPcieLinkGeneration, \ref nvmlDeviceGetMaxPcieLinkWidth, \ref nvmlDeviceGetCurrPcieLinkGeneration,\ref nvmlDeviceGetCurrPcieLinkWidth
* - Format change of \ref nvmlDeviceGetUUID output to match the UUID standard. This function will return a different value.
* - \ref nvmlDeviceGetDetailedEccErrors will report zero for unsupported ECC error counters when a subset of ECC error counters are supported
* \section changelog1 Changes between NVML v1.0 and v2.285
*
* The following new functionality is exposed on NVIDIA display drivers version 285 or later
*
* - Added possibility to query separately current and pending driver model with \p nvmlDeviceGetDriverModel
* - Added API \ref nvmlDeviceGetVbiosVersion function to report VBIOS version.
* - Added pciSubSystemId to \ref nvmlPciInfo_t struct
* - Added API \ref nvmlErrorString function to convert error code to string
* - Updated docs to indicate we support M2075 and C2075
* - Added API \ref nvmlSystemGetHicVersion function to report HIC firmware version
* - Added NVML versioning support
* - Functions that changed API and/or size of structs have appended versioning suffix
* (e.g. nvmlDeviceGetPciInfo_v2). Appropriate C defines have been
* added that map old function names to the newer version of the function
* - Added support for concurrent library usage by multiple libraries
* - Added API \ref nvmlDeviceGetMaxClockInfo function for reporting device's clock limits
* - Added new error code NVML_ERROR_DRIVER_NOT_LOADED used by \ref nvmlInit_v2
* - Extended \ref nvmlPciInfo_t struct with new field: sub system id
* - Added NVML support on Windows guest account
* - Changed format of pciBusId string (to XXXX:XX:XX.X) of \ref nvmlPciInfo_t
* - Parsing of busId in \ref nvmlDeviceGetHandleByPciBusId_v2 is less restrictive. You can pass 0:2:0.0 or 0000:02:00 and other variations
* - Added API for events waiting for GPU events (Linux only) see docs of \ref nvmlEvents
* - Added API \p nvmlDeviceGetComputeRunningProcesses_v2 and \ref nvmlSystemGetProcessName functions for looking up currently running compute applications
* - Deprecated \ref nvmlDeviceGetPowerState in favor of \ref nvmlDeviceGetPerformanceState.
* - Added \ref NVML_FI_DEV_POWER_REQUESTED_LIMIT to report out the power limit requested by the client.
*/

33
pkgs/cuda-13.0/nvml/doc/nvml_deprecation_and_removal.txt Normal file
View File

@@ -0,0 +1,33 @@
/*! @page DeprecationNotices Deprecation and/or removal notices for the NVML library
* This chap†er lists the NVML functions marked for deprecation and/or removal. Starting from CUDA 13.1 deprecated functions will generate a compiler warning. Removed functions will return the NVML error code NVML_ERROR_DEPRECATED.
*
* \section depNotice1 CUDA 13.0 ===
* The following functions are deprecated starting CUDA 13.0; they will be removed in CUDA 14.0.
*
* - nvmlDeviceSetApplicationsClocks
* - nvmlDeviceGetApplicationsClock
* - nvmlDeviceGetDefaultApplicationsClock
* - nvmlDeviceResetApplicationsClocks
* - nvmlDeviceGetViolationStatus
* - nvmlVgpuInstanceGetLicenseStatus
* - nvmlDeviceResetNvLinkUtilizationCounter
* - nvmlDeviceFreezeNvLinkUtilizationCounter
* - nvmlDeviceGetNvLinkUtilizationCounter
* - nvmlDeviceGetNvLinkUtilizationControl
* - nvmlDeviceSetNvLinkUtilizationControl
* - nvmlDeviceSetMemClkVfOffset
* - nvmlDeviceSetGpcClkVfOffset
* - nvmlDeviceGetGpuFabricInfo
* - nvmlDeviceGetDetailedEccErrors
* - nvmlDeviceGetPowerManagementMode
* - nvmlDeviceGetPowerState
* - nvmlDeviceGetSupportedClocksThrottleReasons
* - nvmlDeviceGetCurrentClocksThrottleReasons
* - nvmlDeviceGetTemperature
* - nvmlDeviceGetHandleBySerial
*
* The following data structures are deprecated starting CUDA 13.0
*
* - nvmlGpuFabricInfo_v2_t
*
*/

87
pkgs/cuda-13.0/nvml/example/Makefile Normal file
View File

@@ -0,0 +1,87 @@
ARCH := $(shell getconf LONG_BIT)
OS := $(shell cat /etc/issue)
ifneq (,$(wildcard /etc/redhat-release))
RHEL_OS := $(shell cat /etc/redhat-release)
endif
# Gets Driver Branch
DRIVER_BRANCH := $(shell nvidia-smi | grep Driver | cut -f 3 -d' ' | cut -f 1 -d '.')
# Location of the CUDA Toolkit
CUDA_PATH ?= "/usr/local/cuda-8.0"
ifeq (${ARCH},$(filter ${ARCH},32 64))
# If correct architecture and libnvidia-ml library is not found
# within the environment, build using the stub library
ifneq (,$(findstring Ubuntu,$(OS)))
DEB := $(shell dpkg -l | grep cuda)
ifneq (,$(findstring cuda, $(DEB)))
NVML_LIB := /usr/lib/nvidia-$(DRIVER_BRANCH)
else
NVML_LIB := /lib${ARCH}
endif
endif
ifneq (,$(findstring SUSE,$(OS)))
RPM := $(shell rpm -qa cuda*)
ifneq (,$(findstring cuda, $(RPM)))
NVML_LIB := /usr/lib${ARCH}
else
NVML_LIB := /lib${ARCH}
endif
endif
ifneq (,$(findstring CentOS,$(RHEL_OS)))
RPM := $(shell rpm -qa cuda*)
ifneq (,$(findstring cuda, $(RPM)))
NVML_LIB := /usr/lib${ARCH}/nvidia
else
NVML_LIB := /lib${ARCH}
endif
endif
ifneq (,$(findstring Red Hat,$(RHEL_OS)))
RPM := $(shell rpm -qa cuda*)
ifneq (,$(findstring cuda, $(RPM)))
NVML_LIB := /usr/lib${ARCH}/nvidia
else
NVML_LIB := /lib${ARCH}
endif
endif
ifneq (,$(findstring Fedora,$(RHEL_OS)))
RPM := $(shell rpm -qa cuda*)
ifneq (,$(findstring cuda, $(RPM)))
NVML_LIB := /usr/lib${ARCH}/nvidia
else
NVML_LIB := /lib${ARCH}
endif
endif
else
NVML_LIB := ../../lib${ARCH}/stubs/
$(info "libnvidia-ml.so.1" not found, using stub library.)
endif
ifneq (${ARCH},$(filter ${ARCH},32 64))
$(error Unknown architecture!)
endif
NVML_LIB += ../lib/
NVML_LIB_L := $(addprefix -L , $(NVML_LIB))
CFLAGS := -I ../../include -I ../include
LDFLAGS := -lnvidia-ml $(NVML_LIB_L)
all: example supportedVgpus
example: example.o
$(CC) $< $(CFLAGS) $(LDFLAGS) -o $@
supportedVgpus: supportedVgpus.o
$(CC) $< $(CFLAGS) $(LDFLAGS) -o $@
clean:
-@rm -f example.o
-@rm -f example
-@rm -f supportedVgpus.o
-@rm -f supportedVgpus

10
pkgs/cuda-13.0/nvml/example/README.txt Normal file
View File

@@ -0,0 +1,10 @@
The NVIDIA GDK provides a simple example program that shows how to build an
NVML client. When running an NVML client while the GDK is installed, be
sure your library path first includes the actual NVML library (installed
with the driver), not the stub library that exists solely for
compilation on systems without an NVIDIA driver available.
If you have installed this example code via your packaging system,
you should first copy it to a user directory before compilation.
The packaging system uninstall feature will not remove this directory if it
contains new files beyond what were installed as part of the GDK.

180
pkgs/cuda-13.0/nvml/example/example.c Normal file
View File

@@ -0,0 +1,180 @@
/***************************************************************************\
|* *|
|* Copyright 2010-2016 NVIDIA Corporation. All rights reserved. *|
|* *|
|* NOTICE TO USER: *|
|* *|
|* This source code is subject to NVIDIA ownership rights under U.S. *|
|* and international Copyright laws. Users and possessors of this *|
|* source code are hereby granted a nonexclusive, royalty-free *|
|* license to use this code in individual and commercial software. *|
|* *|
|* NVIDIA MAKES NO REPRESENTATION ABOUT THE SUITABILITY OF THIS SOURCE *|
|* CODE FOR ANY PURPOSE. IT IS PROVIDED "AS IS" WITHOUT EXPRESS OR *|
|* IMPLIED WARRANTY OF ANY KIND. NVIDIA DISCLAIMS ALL WARRANTIES WITH *|
|* REGARD TO THIS SOURCE CODE, INCLUDING ALL IMPLIED WARRANTIES OF *|
|* MERCHANTABILITY, NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR *|
|* PURPOSE. IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY SPECIAL, *|
|* INDIRECT, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES *|
|* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN *|
|* AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING *|
|* OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOURCE *|
|* CODE. *|
|* *|
|* U.S. Government End Users. This source code is a "commercial item" *|
|* as that term is defined at 48 C.F.R. 2.101 (OCT 1995), consisting *|
|* of "commercial computer software" and "commercial computer software *|
|* documentation" as such terms are used in 48 C.F.R. 12.212 (SEPT 1995) *|
|* and is provided to the U.S. Government only as a commercial end item. *|
|* Consistent with 48 C.F.R.12.212 and 48 C.F.R. 227.7202-1 through *|
|* 227.7202-4 (JUNE 1995), all U.S. Government End Users acquire the *|
|* source code with only those rights set forth herein. *|
|* *|
|* Any use of this source code in individual and commercial software must *|
|* include, in the user documentation and internal comments to the code, *|
|* the above Disclaimer and U.S. Government End Users Notice. *|
|* *|
|* *|
\***************************************************************************/
#include <stdio.h>
#include <nvml.h>
static const char * convertToComputeModeString(nvmlComputeMode_t mode)
{
switch (mode)
{
case NVML_COMPUTEMODE_DEFAULT:
return "Default";
case NVML_COMPUTEMODE_EXCLUSIVE_THREAD:
return "Exclusive_Thread";
case NVML_COMPUTEMODE_PROHIBITED:
return "Prohibited";
case NVML_COMPUTEMODE_EXCLUSIVE_PROCESS:
return "Exclusive Process";
default:
return "Unknown";
}
}
int main(void)
{
nvmlReturn_t result;
unsigned int device_count, i;
// First initialize NVML library
result = nvmlInit();
if (NVML_SUCCESS != result)
{
printf("Failed to initialize NVML: %s\n", nvmlErrorString(result));
printf("Press ENTER to continue...\n");
getchar();
return 1;
}
result = nvmlDeviceGetCount(&device_count);
if (NVML_SUCCESS != result)
{
printf("Failed to query device count: %s\n", nvmlErrorString(result));
goto Error;
}
printf("Found %u device%s\n\n", device_count, device_count != 1 ? "s" : "");
printf("Listing devices:\n");
for (i = 0; i < device_count; i++)
{
nvmlDevice_t device;
char name[NVML_DEVICE_NAME_BUFFER_SIZE];
nvmlPciInfo_t pci;
nvmlComputeMode_t compute_mode;
// Query for device handle to perform operations on a device
// You can also query device handle by other features like:
// nvmlDeviceGetHandleBySerial
// nvmlDeviceGetHandleByPciBusId
result = nvmlDeviceGetHandleByIndex(i, &device);
if (NVML_SUCCESS != result)
{
printf("Failed to get handle for device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
result = nvmlDeviceGetName(device, name, NVML_DEVICE_NAME_BUFFER_SIZE);
if (NVML_SUCCESS != result)
{
printf("Failed to get name of device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
// pci.busId is very useful to know which device physically you're talking to
// Using PCI identifier you can also match nvmlDevice handle to CUDA device.
result = nvmlDeviceGetPciInfo(device, &pci);
if (NVML_SUCCESS != result)
{
printf("Failed to get pci info for device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
printf("%u. %s [%s]\n", i, name, pci.busId);
// This is a simple example on how you can modify GPU's state
result = nvmlDeviceGetComputeMode(device, &compute_mode);
if (NVML_ERROR_NOT_SUPPORTED == result)
printf("\t This is not CUDA capable device\n");
else if (NVML_SUCCESS != result)
{
printf("Failed to get compute mode for device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
else
{
// try to change compute mode
printf("\t Changing device's compute mode from '%s' to '%s'\n",
convertToComputeModeString(compute_mode),
convertToComputeModeString(NVML_COMPUTEMODE_PROHIBITED));
result = nvmlDeviceSetComputeMode(device, NVML_COMPUTEMODE_PROHIBITED);
if (NVML_ERROR_NO_PERMISSION == result)
printf("\t\t Need root privileges to do that: %s\n", nvmlErrorString(result));
else if (NVML_ERROR_NOT_SUPPORTED == result)
printf("\t\t Compute mode prohibited not supported. You might be running on\n"
"\t\t windows in WDDM driver model or on non-CUDA capable GPU\n");
else if (NVML_SUCCESS != result)
{
printf("\t\t Failed to set compute mode for device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
else
{
printf("\t Restoring device's compute mode back to '%s'\n",
convertToComputeModeString(compute_mode));
result = nvmlDeviceSetComputeMode(device, compute_mode);
if (NVML_SUCCESS != result)
{
printf("\t\t Failed to restore compute mode for device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
}
}
}
result = nvmlShutdown();
if (NVML_SUCCESS != result)
printf("Failed to shutdown NVML: %s\n", nvmlErrorString(result));
printf("All done.\n");
printf("Press ENTER to continue...\n");
getchar();
return 0;
Error:
result = nvmlShutdown();
if (NVML_SUCCESS != result)
printf("Failed to shutdown NVML: %s\n", nvmlErrorString(result));
printf("Press ENTER to continue...\n");
getchar();
return 1;
}

160
pkgs/cuda-13.0/nvml/example/supportedVgpus.c Normal file
View File

@@ -0,0 +1,160 @@
/***************************************************************************\
|* *|
|* Copyright 2010-2016 NVIDIA Corporation. All rights reserved. *|
|* *|
|* NOTICE TO USER: *|
|* *|
|* This source code is subject to NVIDIA ownership rights under U.S. *|
|* and international Copyright laws. Users and possessors of this *|
|* source code are hereby granted a nonexclusive, royalty-free *|
|* license to use this code in individual and commercial software. *|
|* *|
|* NVIDIA MAKES NO REPRESENTATION ABOUT THE SUITABILITY OF THIS SOURCE *|
|* CODE FOR ANY PURPOSE. IT IS PROVIDED "AS IS" WITHOUT EXPRESS OR *|
|* IMPLIED WARRANTY OF ANY KIND. NVIDIA DISCLAIMS ALL WARRANTIES WITH *|
|* REGARD TO THIS SOURCE CODE, INCLUDING ALL IMPLIED WARRANTIES OF *|
|* MERCHANTABILITY, NONINFRINGEMENT, AND FITNESS FOR A PARTICULAR *|
|* PURPOSE. IN NO EVENT SHALL NVIDIA BE LIABLE FOR ANY SPECIAL, *|
|* INDIRECT, INCIDENTAL, OR CONSEQUENTIAL DAMAGES, OR ANY DAMAGES *|
|* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN *|
|* AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING *|
|* OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOURCE *|
|* CODE. *|
|* *|
|* U.S. Government End Users. This source code is a "commercial item" *|
|* as that term is defined at 48 C.F.R. 2.101 (OCT 1995), consisting *|
|* of "commercial computer software" and "commercial computer software *|
|* documentation" as such terms are used in 48 C.F.R. 12.212 (SEPT 1995) *|
|* and is provided to the U.S. Government only as a commercial end item. *|
|* Consistent with 48 C.F.R.12.212 and 48 C.F.R. 227.7202-1 through *|
|* 227.7202-4 (JUNE 1995), all U.S. Government End Users acquire the *|
|* source code with only those rights set forth herein. *|
|* *|
|* Any use of this source code in individual and commercial software must *|
|* include, in the user documentation and internal comments to the code, *|
|* the above Disclaimer and U.S. Government End Users Notice. *|
|* *|
|* *|
\***************************************************************************/
#include <stdio.h>
#include <nvml.h>
#include <stdlib.h>
int main(void)
{
nvmlReturn_t result;
unsigned int device_count, i;
// First initialize NVML library
result = nvmlInit();
if (NVML_SUCCESS != result)
{
printf("Failed to initialize NVML: %s\n", nvmlErrorString(result));
return 1;
}
result = nvmlDeviceGetCount(&device_count);
if (NVML_SUCCESS != result)
{
printf("Failed to query device count: %s\n", nvmlErrorString(result));
goto Error;
}
printf("Found %u device%s\n", device_count, device_count != 1 ? "s" : "");
printf("Listing devices:\n");
for (i = 0; i < device_count; i++)
{
nvmlDevice_t device;
char name[NVML_DEVICE_NAME_BUFFER_SIZE];
nvmlPciInfo_t pci;
// Query for device handle to perform operations on a device
// You can also query device handle by other features like:
// nvmlDeviceGetHandleBySerial
// nvmlDeviceGetHandleByPciBusId
result = nvmlDeviceGetHandleByIndex(i, &device);
if (NVML_SUCCESS != result)
{
printf("Failed to get handle for device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
result = nvmlDeviceGetName(device, name, NVML_DEVICE_NAME_BUFFER_SIZE);
if (NVML_SUCCESS != result)
{
printf("Failed to get name of device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
// pci.busId is very useful to know which device physically you're talking to
// Using PCI identifier you can also match nvmlDevice handle to CUDA device.
result = nvmlDeviceGetPciInfo(device, &pci);
if (NVML_SUCCESS != result)
{
printf("Failed to get pci info for device %u: %s\n", i, nvmlErrorString(result));
goto Error;
}
printf("%u. %s [%s]\n", i, name, pci.busId);
// This is an example to get the supported vGPUs type names
unsigned int vgpuCount = 0;
nvmlVgpuTypeId_t *vgpuTypeIds = NULL;
unsigned int j;
result = nvmlDeviceGetSupportedVgpus(device, &vgpuCount, NULL);
if (NVML_ERROR_INSUFFICIENT_SIZE != result)
goto Error;
if (vgpuCount != 0)
{
vgpuTypeIds = malloc(sizeof(nvmlVgpuTypeId_t) * vgpuCount);
if (!vgpuTypeIds)
{
printf("Memory allocation of %d bytes failed \n", (int)(sizeof(*vgpuTypeIds)*vgpuCount));
goto Error;
}
result = nvmlDeviceGetSupportedVgpus(device, &vgpuCount, vgpuTypeIds);
if (NVML_SUCCESS != result)
{
printf("Failed to get the supported vGPUs with status %d \n", (int)result);
goto Error;
}
printf(" Displaying vGPU type names: \n");
for (j = 0; j < vgpuCount; j++)
{
char vgpuTypeName[NVML_DEVICE_NAME_BUFFER_SIZE];
unsigned int bufferSize = NVML_DEVICE_NAME_BUFFER_SIZE;
if (NVML_SUCCESS == (result = nvmlVgpuTypeGetName(vgpuTypeIds[j], vgpuTypeName, &bufferSize)))
{
printf(" %s\n",vgpuTypeName);
}
else
{
printf("Failed to query the vGPU type name with status %d \n", (int)result);
}
}
}
if (vgpuTypeIds)
free(vgpuTypeIds);
}
result = nvmlShutdown();
if (NVML_SUCCESS != result)
printf("Failed to shutdown NVML: %s\n", nvmlErrorString(result));
printf("All done.\n");
return 0;
Error:
result = nvmlShutdown();
if (NVML_SUCCESS != result)
printf("Failed to shutdown NVML: %s\n", nvmlErrorString(result));
return 1;
}

13607
pkgs/cuda-13.0/targets/x86_64-linux/include/nvml.h Normal file
View File

File diff suppressed because it is too large Load Diff

BIN
pkgs/cuda-13.0/targets/x86_64-linux/lib/stubs/libnvidia-ml.a Normal file
View File

Binary file not shown.

BIN
pkgs/cuda-13.0/targets/x86_64-linux/lib/stubs/libnvidia-ml.so Executable file
View File

Binary file not shown.