Struct GpuInstance 

pub struct GpuInstance(/* private fields */);

Implementations

impl GpuInstance

pub const unsafe fn from_raw(handle: nvmlGpuInstance_t) -> Self

pub const fn as_raw(&self) -> nvmlGpuInstance_t

pub const fn is_null(&self) -> bool

pub fn info(&self) -> Result<GpuInstanceInfo>

Returns GPU instance information.

For Ampere or newer fully supported devices. Supported on Linux only.

Errors

Returns an error if the handle or output arguments are rejected by NVML, if the current process does not have permission to query the instance, or if NVML has not been initialized.

pub fn parent_device(&self) -> Result<Device>

pub fn id(&self) -> Result<u32>

pub fn profile_id(&self) -> Result<u32>

pub fn compute_instance_profile_info( &self, profile: u32, engine_profile: ComputeInstanceEngineProfile, ) -> Result<ComputeInstanceProfileInfo>

Versioned wrapper that requests compute-instance profile information using the latest supported NVML output layout.

This wrapper sets the version field on the output structure before calling NVML.

For Ampere or newer fully supported devices. Supported on Linux only.

Errors

Returns an error if the GPU instance, profile, engine profile, or structure version is invalid, if the profile is not supported, if the current process does not have permission to perform the operation, or if NVML has not been initialized.

pub fn compute_instance_remaining_capacity( &self, profile_id: u32, ) -> Result<u32>

Returns compute instance profile capacity.

For Ampere or newer fully supported devices. Supported on Linux only. Requires privileged access.

Errors

Returns an error if the GPU instance or profile_id is invalid, if the profile is not supported, if the current process does not have permission to perform the operation, or if NVML has not been initialized.

pub fn compute_instance_possible_placements( &self, profile_id: u32, ) -> Result<Vec<ComputeInstancePlacement>>

Returns compute instance placements.

For Ampere or newer fully supported devices. Supported on Linux only. Requires privileged access.

A placement represents the location of a compute instance within a GPU instance. Returns all possible placements for the given profile. A created compute instance occupies compute slices described by its placement. Creating a compute instance fails if its placement overlaps already occupied compute slices.

Errors

Returns an error if the GPU instance or profile_id is invalid, if MIG mode is not enabled or the profile is not supported, if the current process does not have permission to perform the operation, or if NVML has not been initialized.

pub fn compute_instance_by_id(&self, id: u32) -> Result<ComputeInstance>

Returns compute instance for the given instance ID.

For Ampere or newer fully supported devices. Supported on Linux only. Requires privileged access.

Errors

Returns an error if the GPU instance or ID is invalid, if the compute instance is not found, if MIG mode is not enabled, if the current process lacks permission, or if NVML has not been initialized.

pub fn create_compute_instance( &self, profile_id: u32, ) -> Result<OwnedComputeInstance>

Creates a compute instance.

For Ampere or newer fully supported devices. Supported on Linux only. Requires privileged access.

If the parent device is unbound or reset, or if the parent GPU instance or compute instance is destroyed, the compute instance handle becomes invalid. The compute instance must be recreated to acquire a valid handle.

Errors

Returns an error if the requested compute instance cannot be created, if the GPU instance or profile_id is invalid, if the profile is not supported, if the current process lacks permission, or if NVML has not been initialized.

pub fn create_compute_instance_with_placement( &self, profile_id: u32, placement: ComputeInstancePlacement, ) -> Result<OwnedComputeInstance>

Creates a compute instance with the specified placement.

For Ampere or newer fully supported devices. Supported on Linux only. Requires privileged access.

If the parent device is unbound or reset, or if the parent GPU instance or compute instance is destroyed, the compute instance handle becomes invalid. The compute instance must be recreated to acquire a valid handle.

Errors

Returns an error if the requested compute instance cannot be created, if the GPU instance, profile_id, or placement is invalid, if the profile is not supported, if the current process lacks permission, or if NVML has not been initialized.

pub fn compute_instances(&self, profile_id: u32) -> Result<Vec<ComputeInstance>>

Returns compute instances for the given profile ID.

For Ampere or newer fully supported devices. Supported on Linux only. Requires privileged access.

Errors

Returns an error if the GPU instance or profile_id is invalid, if the profile is not supported, if the current process lacks permission, or if NVML has not been initialized.

pub fn creatable_vgpus(&self) -> Result<Vec<VgpuTypeId>>

Query the currently creatable vGPU types on a specific GPU instance.

Returns the vGPU types that can currently be created for this GPU instance. This wrapper performs the NVML size query internally and returns the results as a Vec.

The creatable vGPU types may differ over time, as there may be restrictions on what type of vGPUs can concurrently run on the device.

Errors

Returns an error if NVML rejects the versioned request, if the intermediate size query reports a larger buffer requirement than expected, if this GPU instance or query arguments are invalid, if the host or GPU does not support vGPU creation, if NVML has not been initialized, or if NVML reports an unexpected failure.

pub fn active_vgpus(&self) -> Result<Vec<VgpuInstance>>

Returns the active vGPU instances within a GPU instance.

This wrapper performs the NVML size query internally and returns the active vGPU instances as a Vec.

Errors

Returns an error if NVML rejects the versioned request, if the intermediate size query reports a larger buffer requirement than expected, if this GPU instance or query arguments are invalid, if the host or GPU does not support vGPU queries, if NVML has not been initialized, or if NVML reports an unexpected failure.

pub fn vgpu_heterogeneous_mode(&self) -> Result<EnableState>

Returns the vGPU heterogeneous mode for the GPU instance.

When in heterogeneous mode, a vGPU can concurrently host timesliced vGPUs with differing framebuffer sizes.

On success, returns the current vGPU heterogeneous mode as EnableState::Enabled or EnableState::Disabled.

For Blackwell &tm GB20x; or newer fully supported devices.

Errors

Returns an error if NVML rejects the versioned request, if this GPU instance or query arguments are invalid, if the host, GPU, or MIG mode does not support the query, if NVML has not been initialized, or if NVML reports an unexpected failure.

pub fn vgpu_type_creatable_placements( &self, vgpu_type_id: VgpuTypeId, ) -> Result<Vec<VgpuPlacementId>>

Query the creatable vGPU placement ID of the vGPU type within a GPU instance.

For Blackwell &tm GB20x; or newer fully supported devices.

The returned placement IDs correspond to the given vgpu_type_id. This wrapper performs the NVML size query internally and returns the placement IDs as a Vec. The creatable vGPU placement IDs may differ over time, as there may be restrictions on what type of vGPU the vGPU instance is running.

Errors

Returns an error if NVML rejects the versioned request, if the intermediate size query reports a larger buffer requirement than expected, if this GPU instance or query arguments are invalid, if the host or GPU does not support the query or vGPU heterogeneous mode is disabled, if NVML has not been initialized, or if NVML reports an unexpected failure.

pub fn vgpu_scheduler_state( &self, engine: VgpuSchedulerEngine, ) -> Result<VgpuSchedulerState>

Returns the vGPU scheduler state for the given GPU instance. The returned scheduler-state details are not relevant when the scheduler policy is best effort.

For Blackwell &tm GB20x; or newer fully supported devices.

Errors

Returns an error if this GPU instance or query arguments are invalid, if the host or GPU does not support vGPU scheduler queries, if NVML has not been initialized, or if NVML reports an unexpected failure.

pub fn vgpu_scheduler_log( &self, engine: VgpuSchedulerEngine, ) -> Result<VgpuSchedulerLog>

Returns the vGPU scheduler logs for this GPU instance. The number of returned elements never exceeds NVML_SCHEDULER_SW_MAX_LOG_ENTRIES.

To get the entire logs, call this method at least 5 times a second.

For Blackwell &tm GB20x; or newer fully supported devices.

Errors

Returns an error if this GPU instance or query arguments are invalid, if the host or GPU does not support vGPU scheduler queries, if NVML has not been initialized, or if NVML reports an unexpected failure.

Trait Implementations

impl Clone for GpuInstance

fn clone(&self) -> GpuInstance

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for GpuInstance

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Hash for GpuInstance

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl PartialEq for GpuInstance

fn eq(&self, other: &GpuInstance) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl Eq for GpuInstance

impl StructuralPartialEq for GpuInstance