# tensorplay.cuda API

Source: https://www.tensorplay.cn/docs/api/tensorplay.cuda.html

## Functions 58

[#](#api-tensorplay.cuda.caching_allocator_alloc)

### caching_allocator_alloc

function[Full reference ↗](/docs/generated/tensorplay.cuda.caching_allocator_alloc.html)

```python
tensorplay.cuda.caching_allocator_alloc(size, device: Any = None, stream=None)
```

Perform a memory allocation using the CUDA memory allocator.

Not exposed by this TensorPlay build.

[#](#api-tensorplay.cuda.caching_allocator_delete)

### caching_allocator_delete

function[Full reference ↗](/docs/generated/tensorplay.cuda.caching_allocator_delete.html)

```python
tensorplay.cuda.caching_allocator_delete(mem_ptr)
```

Delete memory allocated using the CUDA memory allocator.

[#](#api-tensorplay.cuda.can_device_access_peer)

### can_device_access_peer

function[Full reference ↗](/docs/generated/tensorplay.cuda.can_device_access_peer.html)

```python
tensorplay.cuda.can_device_access_peer(device: Any, peer_device: Any) → bool
```

Check if peer access between two devices is possible.

[#](#api-tensorplay.cuda.change_current_allocator)

### change_current_allocator

function[Full reference ↗](/docs/generated/tensorplay.cuda.change_current_allocator.html)

```python
tensorplay.cuda.change_current_allocator(alloc)
```

Change the currently used memory allocator (not supported by this build).

[#](#api-tensorplay.cuda.check_error)

### check_error

function[Full reference ↗](/docs/generated/tensorplay.cuda.check_error.html)

```python
tensorplay.cuda.check_error(res: int) → None
```

Raise an error if the result of a CUDA runtime API call is not success.

[#](#api-tensorplay.cuda.classproperty)

### classproperty

function[Full reference ↗](/docs/generated/tensorplay.cuda.classproperty.html)

```python
tensorplay.cuda.classproperty(func: Callable) → _ClassPropertyDescriptor
```

[#](#api-tensorplay.cuda.clock_rate)

### clock_rate

function[Full reference ↗](/docs/generated/tensorplay.cuda.clock_rate.html)

```python
tensorplay.cuda.clock_rate(device: Any = None) → int
```

Return the clock speed of the GPU SM in MHz (megahertz) over the past sample period as given by nvidia-smi.

[#](#api-tensorplay.cuda.cudart)

### cudart

function[Full reference ↗](/docs/generated/tensorplay.cuda.cudart.html)

```python
tensorplay.cuda.cudart()
```

Retrieves the CUDA runtime API module.

This function initializes the CUDA runtime environment if it is not already
initialized and returns the CUDA runtime API module (_cudart).

Returns:

The CUDA runtime API module, or None when no
ctypes runtime binding is exposed by this build.

Return type:

module or None

[#](#api-tensorplay.cuda.current_blas_handle)

### current_blas_handle

function[Full reference ↗](/docs/generated/tensorplay.cuda.current_blas_handle.html)

```python
tensorplay.cuda.current_blas_handle()
```

Return cublasHandle_t pointer to current cuBLAS handle

[#](#api-tensorplay.cuda.current_device)

### current_device

function[Full reference ↗](/docs/generated/tensorplay.cuda.current_device.html)

```python
tensorplay.cuda.current_device() → int
```

Return the index of a currently selected device.

[#](#api-tensorplay.cuda.current_solver_handle)

### current_solver_handle

function[Full reference ↗](/docs/generated/tensorplay.cuda.current_solver_handle.html)

```python
tensorplay.cuda.current_solver_handle()
```

Return cusolverDnHandle_t pointer to current cuSOLVER handle

[#](#api-tensorplay.cuda.current_stream)

### current_stream

function[Full reference ↗](/docs/generated/tensorplay.cuda.current_stream.html)

```python
tensorplay.cuda.current_stream(device: Any = None) → Stream
```

Return the currently selected Stream for a given device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
the currently selected Stream for the current device, given
by [current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is None
(default).

[#](#api-tensorplay.cuda.default_stream)

### default_stream

function[Full reference ↗](/docs/generated/tensorplay.cuda.default_stream.html)

```python
tensorplay.cuda.default_stream(device: Any = None) → Stream
```

Return the default Stream for a given device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
the default Stream for the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is None
(default).

[#](#api-tensorplay.cuda.device_count)

### device_count

function[Full reference ↗](/docs/generated/tensorplay.cuda.device_count.html)

```python
tensorplay.cuda.device_count() → int
```

Return the number of GPUs available.

> **Note**
>
> This API will NOT poison fork if NVML discovery succeeds.

[#](#api-tensorplay.cuda.device_memory_used)

### device_memory_used

function[Full reference ↗](/docs/generated/tensorplay.cuda.device_memory_used.html)

```python
tensorplay.cuda.device_memory_used(device: Any = None) → int
```

Return used global (device) memory in bytes as given by nvidia-smi.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
statistic for the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is
None (default).

[#](#api-tensorplay.cuda.empty_cache)

### empty_cache

function[Full reference ↗](/docs/generated/tensorplay.cuda.empty_cache.html)

```python
tensorplay.cuda.empty_cache() → None
```

Release all unoccupied cached memory currently held by the caching
allocator so that those can be used in other GPU application and visible in
nvidia-smi.

> **Note**
>
> empty_cache() doesn’t increase the amount of GPU memory available for TensorPlay. However, it may help reduce fragmentation of GPU memory in certain cases.

[#](#api-tensorplay.cuda.get_allocator_backend)

### get_allocator_backend

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_allocator_backend.html)

```python
tensorplay.cuda.get_allocator_backend() → str
```

Returns the active allocator backend as a string. Always native here.

[#](#api-tensorplay.cuda.get_arch_list)

### get_arch_list

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_arch_list.html)

```python
tensorplay.cuda.get_arch_list() → list[str]
```

Return list CUDA architectures this library was compiled for.

[#](#api-tensorplay.cuda.get_device_capability)

### get_device_capability

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_device_capability.html)

```python
tensorplay.cuda.get_device_capability(device: Any = None) → tuple[int, int]
```

Get the cuda capability of a device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int) or [str](https://docs.python.org/3/builtins/stdtypes.html#str), optional) – device for which to return the
device capability. This function is a no-op if this argument is
a negative integer. It uses the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is None
(default).

Returns:

the major and minor cuda capability of the device

Return type:

[tuple](https://docs.python.org/3/builtins/stdtypes.html#tuple)([int](https://docs.python.org/3/builtins/functions.html#int), [int](https://docs.python.org/3/builtins/functions.html#int))

[#](#api-tensorplay.cuda.get_device_name)

### get_device_name

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_device_name.html)

```python
tensorplay.cuda.get_device_name(device: Any = None) → str
```

Get the name of a device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int) or [str](https://docs.python.org/3/builtins/stdtypes.html#str), optional) – device for which to return the
name. This function is a no-op if this argument is a negative
integer. It uses the current device, given by [current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device),
if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is None (default).

Returns:

the name of the device

Return type:

[str](https://docs.python.org/3/builtins/stdtypes.html#str)

[#](#api-tensorplay.cuda.get_device_properties)

### get_device_properties

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_device_properties.html)

```python
tensorplay.cuda.get_device_properties(device: Any = None) → _CudaDeviceProperties
```

Get the properties of a device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int) or [str](https://docs.python.org/3/builtins/stdtypes.html#str), optional) – device for which to return the
properties of the device.  It uses the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is None
(default).

Returns:

the properties of the device

Return type:

_CudaDeviceProperties

[#](#api-tensorplay.cuda.get_gencode_flags)

### get_gencode_flags

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_gencode_flags.html)

```python
tensorplay.cuda.get_gencode_flags() → str
```

Return NVCC gencode flags this library was compiled with.

[#](#api-tensorplay.cuda.get_per_process_memory_fraction)

### get_per_process_memory_fraction

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_per_process_memory_fraction.html)

```python
tensorplay.cuda.get_per_process_memory_fraction(device: Any = None) → float
```

Get memory fraction for a process. Always returns 1.0 in this build.

[#](#api-tensorplay.cuda.get_stream_from_external)

### get_stream_from_external

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_stream_from_external.html)

```python
tensorplay.cuda.get_stream_from_external(data_ptr: int, device: Any = None) → Stream
```

Return a Stream from an externally allocated CUDA stream.

[#](#api-tensorplay.cuda.get_sync_debug_mode)

### get_sync_debug_mode

function[Full reference ↗](/docs/generated/tensorplay.cuda.get_sync_debug_mode.html)

```python
tensorplay.cuda.get_sync_debug_mode() → int
```

Return current value of debug mode for cuda synchronizing operations.

Always returns 0 in this build (mode is not enforced).

[#](#api-tensorplay.cuda.host_memory_stats_as_nested_dict)

### host_memory_stats_as_nested_dict

function[Full reference ↗](/docs/generated/tensorplay.cuda.host_memory_stats_as_nested_dict.html)

```python
tensorplay.cuda.host_memory_stats_as_nested_dict() → dict[str, Any]
```

Return the result of [host_memory_stats()](/docs/generated/tensorplay.cuda.host_memory_stats.html#tensorplay.cuda.host_memory_stats) as a nested dictionary.

[#](#api-tensorplay.cuda.host_memory_stats)

### host_memory_stats

function[Full reference ↗](/docs/generated/tensorplay.cuda.host_memory_stats.html)

```python
tensorplay.cuda.host_memory_stats() → dict[str, Any]
```

Return a dictionary of pinned (host) allocator statistics.

[#](#api-tensorplay.cuda.init)

### init

function[Full reference ↗](/docs/generated/tensorplay.cuda.init.html)

```python
tensorplay.cuda.init()
```

Initialize TensorPlay’s CUDA state.

You may need to call this explicitly if you are interacting with
TensorPlay via its C API, as Python bindings for CUDA functionality
will not be available until this initialization takes place.
Ordinary users should not need this, as all of TensorPlay’s CUDA methods
automatically initialize CUDA state on-demand.

Does nothing if the CUDA state is already initialized.

[#](#api-tensorplay.cuda.ipc_collect)

### ipc_collect

function[Full reference ↗](/docs/generated/tensorplay.cuda.ipc_collect.html)

```python
tensorplay.cuda.ipc_collect()
```

Force collects GPU memory after it has been released by CUDA IPC.

[#](#api-tensorplay.cuda.is_available)

### is_available

function[Full reference ↗](/docs/generated/tensorplay.cuda.is_available.html)

```python
tensorplay.cuda.is_available() → bool
```

Return a bool indicating if CUDA is currently available.

> **Note**
>
> This function will NOT poison fork if the environment variable TENSORPLAY_NVML_BASED_CUDA_CHECK=1 is set.

[#](#api-tensorplay.cuda.is_bf16_supported)

### is_bf16_supported

function[Full reference ↗](/docs/generated/tensorplay.cuda.is_bf16_supported.html)

```python
tensorplay.cuda.is_bf16_supported(including_emulation: bool = True)
```

Return a bool indicating if the current CUDA device supports dtype bfloat16.

[#](#api-tensorplay.cuda.is_gds_available)

### is_gds_available

function[Full reference ↗](/docs/generated/tensorplay.cuda.is_gds_available.html)

```python
tensorplay.cuda.is_gds_available() → bool
```

Return whether GDS is available. Always False in this build.

[#](#api-tensorplay.cuda.is_initialized)

### is_initialized

function[Full reference ↗](/docs/generated/tensorplay.cuda.is_initialized.html)

```python
tensorplay.cuda.is_initialized()
```

Return whether TensorPlay’s CUDA state has been initialized.

[#](#api-tensorplay.cuda.is_tf32_supported)

### is_tf32_supported

function[Full reference ↗](/docs/generated/tensorplay.cuda.is_tf32_supported.html)

```python
tensorplay.cuda.is_tf32_supported() → bool
```

Return a bool indicating if the current CUDA device supports dtype tf32.

[#](#api-tensorplay.cuda.list_gpu_processes)

### list_gpu_processes

function[Full reference ↗](/docs/generated/tensorplay.cuda.list_gpu_processes.html)

```python
tensorplay.cuda.list_gpu_processes(device: Any = None) → str
```

Return a human-readable printout of the running processes and their GPU memory use for a given device.

[#](#api-tensorplay.cuda.max_memory_allocated)

### max_memory_allocated

function[Full reference ↗](/docs/generated/tensorplay.cuda.max_memory_allocated.html)

```python
tensorplay.cuda.max_memory_allocated(device: Any = None) → int
```

Return the maximum GPU memory occupied by tensors in bytes for a given device.

By default, this returns the peak allocated memory since the beginning of
this program. [reset_peak_memory_stats()](/docs/generated/tensorplay.cuda.reset_peak_memory_stats.html#tensorplay.cuda.reset_peak_memory_stats) can be used to
reset the starting point in tracking this metric.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
statistic for the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is
None (default).

[#](#api-tensorplay.cuda.max_memory_reserved)

### max_memory_reserved

function[Full reference ↗](/docs/generated/tensorplay.cuda.max_memory_reserved.html)

```python
tensorplay.cuda.max_memory_reserved(device: Any = None) → int
```

Return the maximum GPU memory managed by the caching allocator in bytes for a given device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
statistic for the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is
None (default).

[#](#api-tensorplay.cuda.mem_get_info)

### mem_get_info

function[Full reference ↗](/docs/generated/tensorplay.cuda.mem_get_info.html)

```python
tensorplay.cuda.mem_get_info(device: Any = None) → tuple[int, int]
```

Return the global free and total GPU memory for a given device using cudaMemGetInfo.

[#](#api-tensorplay.cuda.memory_allocated)

### memory_allocated

function[Full reference ↗](/docs/generated/tensorplay.cuda.memory_allocated.html)

```python
tensorplay.cuda.memory_allocated(device: Any = None) → int
```

Return the current GPU memory occupied by tensors in bytes for a given device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
statistic for the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is
None (default).

[#](#api-tensorplay.cuda.memory_reserved)

### memory_reserved

function[Full reference ↗](/docs/generated/tensorplay.cuda.memory_reserved.html)

```python
tensorplay.cuda.memory_reserved(device: Any = None) → int
```

Return the current GPU memory managed by the caching allocator in bytes for a given device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
statistic for the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is
None (default).

[#](#api-tensorplay.cuda.memory_snapshot)

### memory_snapshot

function[Full reference ↗](/docs/generated/tensorplay.cuda.memory_snapshot.html)

```python
tensorplay.cuda.memory_snapshot(mempool_id=None, include_traces=True)
```

Return a snapshot of the CUDA memory allocator state across all devices.

Interpreting the output of this function requires familiarity with the
memory allocator internals. Not exposed by this TensorPlay build.

[#](#api-tensorplay.cuda.memory_stats_as_nested_dict)

### memory_stats_as_nested_dict

function[Full reference ↗](/docs/generated/tensorplay.cuda.memory_stats_as_nested_dict.html)

```python
tensorplay.cuda.memory_stats_as_nested_dict(device: Any = None) → dict[str, Any]
```

Return the result of [memory_stats()](/docs/generated/tensorplay.cuda.memory_stats.html#tensorplay.cuda.memory_stats) as a nested dictionary.

The native allocator reports the fragmentation-aware matrix directly
(segments, free-block histogram, pending cross-stream blocks, graph
pools, capture state); it is exposed under "allocator" alongside the

[#](#api-tensorplay.cuda.memory_stats)

### memory_stats

function[Full reference ↗](/docs/generated/tensorplay.cuda.memory_stats.html)

```python
tensorplay.cuda.memory_stats(device: Any = None) → dict[str, Any]
```

Return a dictionary of CUDA memory allocator statistics for a given device.

The return value of this function is a dictionary of statistics, each of
the full key layout; keys that are not tracked by this TensorPlay build
are always reported as zero.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
statistics for the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is
None (default).

[#](#api-tensorplay.cuda.memory_summary)

### memory_summary

function[Full reference ↗](/docs/generated/tensorplay.cuda.memory_summary.html)

```python
tensorplay.cuda.memory_summary(device: Any = None, abbreviated: bool = False) → str
```

Return a human-readable printout of the current memory allocator statistics for a given device.

This can be useful to display periodically during training, or when
handling out-of-memory exceptions.

Parameters:

- device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns printout for the current device, given by [current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is None (default).

- abbreviated ([bool](https://docs.python.org/3/builtins/functions.html#bool), optional) – whether to return an abbreviated summary (default: False).

[#](#api-tensorplay.cuda.memory_usage)

### memory_usage

function[Full reference ↗](/docs/generated/tensorplay.cuda.memory_usage.html)

```python
tensorplay.cuda.memory_usage(device: Any = None) → int
```

Return the percent of time over the past sample period during which global (device)
memory was being read or written as given by nvidia-smi.

Warning: Each sample period may be between 1 second and 1/6 second,
depending on the product being queried.

[#](#api-tensorplay.cuda.power_draw)

### power_draw

function[Full reference ↗](/docs/generated/tensorplay.cuda.power_draw.html)

```python
tensorplay.cuda.power_draw(device: Any = None) → int
```

Return the average power draw of the GPU sensor in mW (MilliWatts)
over the past sample period as given by nvidia-smi for Fermi or newer fully supported devices.

[#](#api-tensorplay.cuda.reset_accumulated_host_memory_stats)

### reset_accumulated_host_memory_stats

function[Full reference ↗](/docs/generated/tensorplay.cuda.reset_accumulated_host_memory_stats.html)

```python
tensorplay.cuda.reset_accumulated_host_memory_stats() → None
```

Reset the “accumulated” (historical) stats tracked by the host memory allocator.

[#](#api-tensorplay.cuda.reset_accumulated_memory_stats)

### reset_accumulated_memory_stats

function[Full reference ↗](/docs/generated/tensorplay.cuda.reset_accumulated_memory_stats.html)

```python
tensorplay.cuda.reset_accumulated_memory_stats(device: Any = None) → None
```

Reset the “accumulated” (historical) stats tracked by the CUDA memory allocator.

This TensorPlay build does not track historical totals; the call is a no-op.

[#](#api-tensorplay.cuda.reset_peak_host_memory_stats)

### reset_peak_host_memory_stats

function[Full reference ↗](/docs/generated/tensorplay.cuda.reset_peak_host_memory_stats.html)

```python
tensorplay.cuda.reset_peak_host_memory_stats() → None
```

Reset the “peak” stats tracked by the host memory allocator.

[#](#api-tensorplay.cuda.reset_peak_memory_stats)

### reset_peak_memory_stats

function[Full reference ↗](/docs/generated/tensorplay.cuda.reset_peak_memory_stats.html)

```python
tensorplay.cuda.reset_peak_memory_stats(device: Any = None) → None
```

Reset the “peak” stats tracked by the CUDA memory allocator.

Peak stats correspond to the “peak” key in each individual stat dict.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device. Returns
statistic for the current device, given by
[current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device), if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is
None (default).

[#](#api-tensorplay.cuda.set_device)

### set_device

function[Full reference ↗](/docs/generated/tensorplay.cuda.set_device.html)

```python
tensorplay.cuda.set_device(device: Any) → None
```

Set the current device.

Usage of this function is discouraged in favor of [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device). In most
cases it’s better to use CUDA_VISIBLE_DEVICES environmental variable.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int)) – selected device. This function is a no-op
if this argument is negative.

[#](#api-tensorplay.cuda.set_per_process_memory_fraction)

### set_per_process_memory_fraction

function[Full reference ↗](/docs/generated/tensorplay.cuda.set_per_process_memory_fraction.html)

```python
tensorplay.cuda.set_per_process_memory_fraction(fraction, device: Any = None) → None
```

Set memory fraction for a process.

Not enforced by this TensorPlay build; the signature is retained for API
compatibility.

Parameters:

- fraction ([float](https://docs.python.org/3/builtins/functions.html#float)) – Range: 0~1. Allowed memory equals total_memory * fraction.

- device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – selected device.

[#](#api-tensorplay.cuda.set_stream)

### set_stream

function[Full reference ↗](/docs/generated/tensorplay.cuda.set_stream.html)

```python
tensorplay.cuda.set_stream(stream: Stream)
```

Set the current stream. This is a wrapper API to set the stream.

Usage of this function is discouraged in favor of the stream
context manager.

Parameters:

stream ([Stream](/docs/generated/tensorplay.cuda.streams.Stream.html#tensorplay.cuda.streams.Stream)) – selected stream. This function is a no-op
if this argument is None.

[#](#api-tensorplay.cuda.set_sync_debug_mode)

### set_sync_debug_mode

function[Full reference ↗](/docs/generated/tensorplay.cuda.set_sync_debug_mode.html)

```python
tensorplay.cuda.set_sync_debug_mode(debug_mode: int | str) → None
```

Set the debug mode for cuda synchronizing operations.

Not enforced by this TensorPlay build; the signature is retained for API
compatibility.

Parameters:

debug_mode ([str](https://docs.python.org/3/builtins/stdtypes.html#str) or [int](https://docs.python.org/3/builtins/functions.html#int)) – if “default” or 0, don’t error or warn on synchronizing operations,
if “warn” or 1, warn on synchronizing operations, if “error” or 2, error out synchronizing operations.

[#](#api-tensorplay.cuda.stream)

### stream

function[Full reference ↗](/docs/generated/tensorplay.cuda.stream.html)

```python
tensorplay.cuda.stream(stream_: Stream | None) → StreamContext
```

Wrap around the Context-manager StreamContext that selects a given stream.

Parameters:

stream ([Stream](/docs/generated/tensorplay.cuda.streams.Stream.html#tensorplay.cuda.streams.Stream)) – selected stream. This manager is a no-op if it’s
None.

> **Note**
>
> Streams are per-device.

[#](#api-tensorplay.cuda.synchronize)

### synchronize

function[Full reference ↗](/docs/generated/tensorplay.cuda.synchronize.html)

```python
tensorplay.cuda.synchronize(device: Any = None) → None
```

Wait for all kernels in all streams on a CUDA device to complete.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int), optional) – device for which to synchronize.
It uses the current device, given by [current_device()](/docs/generated/tensorplay.cuda.current_device.html#tensorplay.cuda.current_device),
if [device](/docs/generated/tensorplay.cuda.device.html#tensorplay.cuda.device) is None (default).

[#](#api-tensorplay.cuda.temperature)

### temperature

function[Full reference ↗](/docs/generated/tensorplay.cuda.temperature.html)

```python
tensorplay.cuda.temperature(device: Any = None) → int
```

Return the average temperature of the GPU sensor in Degrees C (Centigrades).

The average temperature is computed based on past sample period as given by nvidia-smi.

[#](#api-tensorplay.cuda.utilization)

### utilization

function[Full reference ↗](/docs/generated/tensorplay.cuda.utilization.html)

```python
tensorplay.cuda.utilization(device: Any = None) → int
```

Return the percent of time over the past sample period during which one or
more kernels was executing on the GPU as given by nvidia-smi.

Warning: Each sample period may be between 1 second and 1/6 second,
depending on the product being queried.

## Classes 5

[#](#api-tensorplay.cuda.CUDAPluggableAllocator)

### CUDAPluggableAllocator

class[Full reference ↗](/docs/generated/tensorplay.cuda.CUDAPluggableAllocator.html)

```python
class tensorplay.cuda.CUDAPluggableAllocator(so_file: str, alloc_fn_name: str = 'my_alloc', free_fn_name: str = 'my_free')
```

CUDA memory allocator plugin (not supported by this TensorPlay build).

[#](#api-tensorplay.cuda.device_of)

### device_of

class[Full reference ↗](/docs/generated/tensorplay.cuda.device_of.html)

```python
class tensorplay.cuda.device_of(obj)
```

Context-manager that changes the current device to that of given object.

You can use both tensors and storages as arguments. If a given object is
not allocated on a GPU, this is a no-op.

Parameters:

obj ([Tensor](/docs/generated/tensorplay.Tensor.html#tensorplay.Tensor) or Storage) – object allocated on the selected device.

[#](#api-tensorplay.cuda.device)

### device

class[Full reference ↗](/docs/generated/tensorplay.cuda.device.html)

```python
class tensorplay.cuda.device(device: Any)
```

Context-manager that changes the selected device.

Parameters:

device ([tensorplay.Device](/docs/generated/tensorplay.Device.html#tensorplay.Device) or [int](https://docs.python.org/3/builtins/functions.html#int)) – device index to select. It’s a no-op if
this argument is a negative integer or None.

[#](#api-tensorplay.cuda.MemPool)

### MemPool

class[Full reference ↗](/docs/generated/tensorplay.cuda.MemPool.html)

```python
class tensorplay.cuda.MemPool(*args, **kwargs)
```

MemPool context (not supported by this TensorPlay build).

[#](#api-tensorplay.cuda.StreamContext)

### StreamContext

class[Full reference ↗](/docs/generated/tensorplay.cuda.StreamContext.html)

```python
class tensorplay.cuda.StreamContext(stream: Stream | None)
```

Context-manager that selects a given stream.

All CUDA kernels queued within its context will be enqueued on a selected
stream.

Parameters:

Stream ([Stream](/docs/generated/tensorplay.cuda.streams.Stream.html#tensorplay.cuda.streams.Stream)) – selected stream. This manager is a no-op if it’s
None.

> **Note**
>
> Streams are per-device.

## Exceptions 4

[#](#api-tensorplay.cuda.AcceleratorError)

### AcceleratorError

exception[Full reference ↗](/docs/generated/tensorplay.cuda.AcceleratorError.html)

```python
exception tensorplay.cuda.AcceleratorError
```

[#](#api-tensorplay.cuda.CudaError)

### CudaError

exception[Full reference ↗](/docs/generated/tensorplay.cuda.CudaError.html)

```python
exception tensorplay.cuda.CudaError(code: int)
```

[#](#api-tensorplay.cuda.DeferredCudaCallError)

### DeferredCudaCallError

exception[Full reference ↗](/docs/generated/tensorplay.cuda.DeferredCudaCallError.html)

```python
exception tensorplay.cuda.DeferredCudaCallError
```

[#](#api-tensorplay.cuda.OutOfMemoryError)

### OutOfMemoryError

exception[Full reference ↗](/docs/generated/tensorplay.cuda.OutOfMemoryError.html)

```python
exception tensorplay.cuda.OutOfMemoryError
```