Thanks to visit codestin.com
Credit goes to github.com

Skip to content

Commit fc8be1c

Browse files
leofangleofang
authored
Merge pull request NVIDIA#474 from samaid/main
Device properties example
2 parents 5d86ebb + 8812b8d commit fc8be1c

File tree

1 file changed

+234
-0
lines changed

1 file changed

+234
-0
lines changed

‎cuda_core/examples/show_device_properties.py‎

Lines changed: 234 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,234 @@
1+
# Copyright (c) 2025, NVIDIA CORPORATION & AFFILIATES. ALL RIGHTS RESERVED.
2+
#
3+
# SPDX-License-Identifier: LicenseRef-NVIDIA-SOFTWARE-LICENSE
4+
5+
import sys
6+
7+
from cuda.core.experimental import Device, system
8+
9+
10+
# Convert boolean to YES or NO string
11+
def _yes_no(value: bool) -> str:
12+
return "YES" if value else "NO"
13+
14+
15+
# Convert value in bytes to MB or GB string
16+
BYTES_TO_MBYTES = 1 / (1024 * 1024)
17+
BYTES_TO_GBYTES = BYTES_TO_MBYTES / 1024
18+
19+
20+
def _bytes_to_mbytes(value):
21+
return f"{value * BYTES_TO_MBYTES:.2f}MB"
22+
23+
24+
def _bytes_to_gbytes(value):
25+
return f"{value * BYTES_TO_GBYTES:.2f}GB"
26+
27+
28+
# Convert value in KHz to GHz string
29+
KHZ_TO_GHZ = 1e-6
30+
31+
32+
def _khz_to_ghz(value):
33+
return f"{value * KHZ_TO_GHZ:.2f}GHz"
34+
35+
36+
# Print device properties to stdout
37+
def print_device_properties(properties):
38+
print("Properties:\n------------")
39+
print(f"- Can map host memory into the CUDA address space: {_yes_no(properties.can_map_host_memory)}")
40+
print(
41+
"- Can access host registered memory at the same virtual address as the CPU: "
42+
+ f"{_yes_no(properties.can_use_host_pointer_for_registered_mem)}"
43+
)
44+
print(f"- Clock rate: {_khz_to_ghz(properties.clock_rate)}")
45+
print(f"- Peak memory clock frequency: {_khz_to_ghz(properties.memory_clock_rate)}")
46+
print(
47+
"- Performance ratio (single precision)/(double precision): "
48+
+ f"{properties.single_to_double_precision_perf_ratio}"
49+
)
50+
print(
51+
f"- Compute capability: major={properties.compute_capability_major}, "
52+
+ f"minor={properties.compute_capability_minor}"
53+
)
54+
print(f"- Compute mode: {properties.compute_mode} (0 - default, 2 - prohibited, 3 - exclusive process)")
55+
print(f"- Support for Compute Preemption: {_yes_no(properties.compute_preemption_supported)}")
56+
print(
57+
"- Support for concurrent kernels execution within the same context: "
58+
+ f"{_yes_no(properties.concurrent_kernels)}"
59+
)
60+
print(
61+
"- Support for coherent access to managed memory concurrently with CPU: "
62+
+ f"{_yes_no(properties.concurrent_managed_access)}"
63+
)
64+
print(
65+
"- Support for deferred mapping in CUDA arrays and CUDA mipmapped arrays: "
66+
+ f"{_yes_no(properties.deferred_mapping_cuda_array_supported)}"
67+
)
68+
print(
69+
"- Support for direct access of managed memory on device without migration: "
70+
+ f"{_yes_no(properties.direct_managed_mem_access_from_host)}"
71+
)
72+
print(f"- ECC enabled: {_yes_no(properties.ecc_enabled)}")
73+
print(f"- Support for generic compression: {_yes_no(properties.generic_compression_supported)}")
74+
print(f"- Support for caching globals in L1 cache: {_yes_no(properties.global_l1_cache_supported)}")
75+
print(f"- Support for caching locals in L1 cache: {_yes_no(properties.local_l1_cache_supported)}")
76+
print(f"- Global memory bus widths: {properties.global_memory_bus_width} bits")
77+
print(f"- Support for GPUDirect RDMA: {_yes_no(properties.gpu_direct_rdma_supported)}")
78+
print(f"- GPUDirect RDMA flush-writes options bitmask: 0b{properties.gpu_direct_rdma_flush_writes_options:032b}")
79+
print(
80+
f"- GPUDirect RDMA writes ordering: {properties.gpu_direct_rdma_writes_ordering} "
81+
+ "(0 - none, 100 - this device can consume remote writes, "
82+
+ "200 - any CUDA device can consume remote writes to this device)"
83+
)
84+
print(
85+
"- Can concurrently copy memory between host and device while executing kernel: "
86+
+ f"{_yes_no(properties.gpu_overlap)}"
87+
)
88+
print(
89+
"- Support for exporting memory to a posix file descriptor: "
90+
+ f"{_yes_no(properties.handle_type_posix_file_descriptor_supported)}"
91+
)
92+
print(
93+
"- Support for exporting memory to a Win32 NT handle: "
94+
+ f"{_yes_no(properties.handle_type_win32_handle_supported)}"
95+
)
96+
print(
97+
"- Support for exporting memory to a Win32 KMT handle: "
98+
+ f"{_yes_no(properties.handle_type_win32_kmt_handle_supported)}"
99+
)
100+
print(
101+
"- Link between device and host supports native atomic operations: "
102+
+ f"{_yes_no(properties.host_native_atomic_supported)}"
103+
)
104+
print(f"- Device is integrated with memory subsystem: {_yes_no(properties.integrated)}")
105+
print(f"- Kernel execution timeout: {_yes_no(properties.kernel_exec_timeout)}")
106+
print(f"- L2 cache size: {_bytes_to_mbytes(properties.l2_cache_size)}")
107+
print(f"- Max L2 persisting lines capacity: {_bytes_to_mbytes(properties.max_persisting_l2_cache_size)}")
108+
print(f"- Support for managed memory allocation: {_yes_no(properties.managed_memory)}")
109+
print(f"- Max access policy window size: {_bytes_to_mbytes(properties.max_access_policy_window_size)}")
110+
print(f"- Max x-dimension of a block: {properties.max_block_dim_x}")
111+
print(f"- Max y-dimension of a block: {properties.max_block_dim_y}")
112+
print(f"- Max z-dimension of a block: {properties.max_block_dim_z}")
113+
print(f"- Max blocks in a multiprocessor: {properties.max_blocks_per_multiprocessor}")
114+
print(f"- Max x-dimension of a grid: {properties.max_grid_dim_x}")
115+
print(f"- Max y-dimension of a grid: {properties.max_grid_dim_y}")
116+
print(f"- Max z-dimension of a grid: {properties.max_grid_dim_z}")
117+
print(f"- Max pitch allowed by the memory copy functions: {_bytes_to_gbytes(properties.max_pitch)}")
118+
print(f"- Max number of 32-bit registers per block: {properties.max_registers_per_block}")
119+
print(f"- Max number of 32-bit registers in a multiprocessor: {properties.max_registers_per_multiprocessor}")
120+
print(f"- Max shared memory per block: {properties.max_shared_memory_per_block}B")
121+
print(f"- Max optin shared memory per block: {properties.max_shared_memory_per_block_optin}B")
122+
print(f"- Max shared memory available to a multiprocessor: {properties.max_shared_memory_per_multiprocessor}B")
123+
print(f"- Max threads per block: {properties.max_threads_per_block}")
124+
print(f"- Max threads per multiprocessor: {properties.max_threads_per_multiprocessor}")
125+
print(f"- Warp size: {properties.warp_size}")
126+
print(f"- Max 1D surface width: {properties.maximum_surface1d_width}")
127+
print(f"- Max layers in 1D layered surface: {properties.maximum_surface1d_layered_layers}")
128+
print(f"- Max 1D layered surface width: {properties.maximum_surface1d_layered_width}")
129+
print(f"- Max 2D surface width: {properties.maximum_surface2d_width}")
130+
print(f"- Max 2D surface height: {properties.maximum_surface2d_height}")
131+
print(f"- Max layers in 2D layered surface: {properties.maximum_surface2d_layered_layers}")
132+
print(f"- Max 2D layered surface width: {properties.maximum_surface2d_layered_width}")
133+
print(f"- Max 2D layered surface height: {properties.maximum_surface2d_layered_height}")
134+
print(f"- Max 3D surface width: {properties.maximum_surface3d_width}")
135+
print(f"- Max 3D surface height: {properties.maximum_surface3d_height}")
136+
print(f"- Max 3D surface depth: {properties.maximum_surface3d_depth}")
137+
print(f"- Max cubemap surface width: {properties.maximum_surfacecubemap_width}")
138+
print(f"- Max layers in a cubemap layered surface: {properties.maximum_surfacecubemap_layered_layers}")
139+
print(f"- Max cubemap layered surface width: {properties.maximum_surfacecubemap_layered_width}")
140+
print(f"- Max 1D texture width: {properties.maximum_texture1d_width}")
141+
print(f"- Max width for a 1D texture bound to linear memory: {properties.maximum_texture1d_linear_width}")
142+
print(f"- Max layers in 1D layered texture: {properties.maximum_texture1d_layered_layers}")
143+
print(f"- Max 1D layered texture width: {properties.maximum_texture1d_layered_width}")
144+
print(f"- Max mipmapped 1D texture width: {properties.maximum_texture1d_mipmapped_width}")
145+
print(f"- Max 2D texture width: {properties.maximum_texture2d_width}")
146+
print(f"- Max 2D texture height: {properties.maximum_texture2d_height}")
147+
print(f"- Max width for a 2D texture bound to linear memory: {properties.maximum_texture2d_linear_width}")
148+
print(f"- Max height for a 2D texture bound to linear memory: {properties.maximum_texture2d_linear_height}")
149+
print(
150+
"- Max pitch for a 2D texture bound to linear memory: "
151+
+ f"{_bytes_to_mbytes(properties.maximum_texture2d_linear_pitch)}"
152+
)
153+
print(f"- Max layers in 2D layered texture: {properties.maximum_texture2d_layered_layers}")
154+
print(f"- Max 2D layered texture width: {properties.maximum_texture2d_layered_width}")
155+
print(f"- Max 2D layered texture height: {properties.maximum_texture2d_layered_height}")
156+
print(f"- Max mipmapped 2D texture width: {properties.maximum_texture2d_mipmapped_width}")
157+
print(f"- Max mipmapped 2D texture height: {properties.maximum_texture2d_mipmapped_height}")
158+
print(f"- Max 3D texture width: {properties.maximum_texture3d_width}")
159+
print(f"- Max 3D texture height: {properties.maximum_texture3d_height}")
160+
print(f"- Max 3D texture depth: {properties.maximum_texture3d_depth}")
161+
print(f"- Alternate max 3D texture width: {properties.maximum_texture3d_width_alternate}")
162+
print(f"- Alternate max 3D texture height: {properties.maximum_texture3d_height_alternate}")
163+
print(f"- Alternate max 3D texture depth: {properties.maximum_texture3d_depth_alternate}")
164+
print(f"- Max cubemap texture width or height: {properties.maximum_texturecubemap_width}")
165+
print(f"- Max layers in a cubemap layered texture: {properties.maximum_texturecubemap_layered_layers}")
166+
print(f"- Max cubemap layered texture width or height: {properties.maximum_texturecubemap_layered_width}")
167+
print(f"- Texture base address alignment requirement: {properties.texture_alignment}B")
168+
print(
169+
"- Pitch alignment requirement for 2D texture references bound to pitched memory: "
170+
+ f"{properties.texture_pitch_alignment}B"
171+
)
172+
print(f"- Support for memory pools: {_yes_no(properties.memory_pools_supported)}")
173+
print(
174+
"- Bitmask of handle types supported with memory pool-based IPC: "
175+
+ f"0b{properties.mempool_supported_handle_types:032b}"
176+
)
177+
print(f"- Multi-GPU board: {_yes_no(properties.multi_gpu_board)}")
178+
print(f"- Multi-GPU board group ID: {properties.multi_gpu_board_group_id}")
179+
print(f"- Support for switch multicast and reduction operations: {_yes_no(properties.multicast_supported)}")
180+
print(f"- Number of multiprocessors: {properties.multiprocessor_count}")
181+
print(f"- NUMA configuration: {properties.numa_config}")
182+
print(f"- NUMA node ID of GPU memory: {properties.numa_id}")
183+
print(f"- Support for coherently accessing pageable memory: {_yes_no(properties.pageable_memory_access)}")
184+
print(
185+
"- Access pageable memory via host's page tables: "
186+
+ f"{_yes_no(properties.pageable_memory_access_uses_host_page_tables)}"
187+
)
188+
print(f"- PCI bus ID: {properties.pci_bus_id}")
189+
print(f"- PCI device (slot) ID: {properties.pci_device_id}")
190+
print(f"- PCI domain ID: {properties.pci_domain_id}")
191+
print(
192+
"- Support for registering memory that must be mapped to GPU as read-only: "
193+
+ f"{_yes_no(properties.read_only_host_register_supported)}"
194+
)
195+
print(
196+
"- Amount of shared memory per block reserved by CUDA driver: "
197+
+ f"{properties.reserved_shared_memory_per_block}B"
198+
)
199+
print(
200+
"- Support for sparse CUDA arrays and sparse CUDA mipmapped arrays: "
201+
+ f"{_yes_no(properties.sparse_cuda_array_supported)}"
202+
)
203+
print(f"- Using TCC driver: {_yes_no(properties.tcc_driver)}")
204+
print(f"- Constant memory available: {properties.total_constant_memory}B")
205+
print(f"- Support for unified address space with host: {_yes_no(properties.unified_addressing)}")
206+
print(f"- Support for virtual memory management: {_yes_no(properties.virtual_memory_management_supported)}")
207+
208+
209+
# Print info about all CUDA devices in the system
210+
def show_device_properties():
211+
ndev = system.num_devices
212+
print(f"Number of GPUs: {ndev}")
213+
214+
for device_id in range(ndev):
215+
device = Device(device_id)
216+
print(f"DEVICE {device.name} (id={device_id})")
217+
218+
device.set_current()
219+
# Extend example to show device context information after #189 is resolved.
220+
# ctx = device.context
221+
222+
cc = device.compute_capability
223+
prop = device.properties
224+
225+
print(f"Device compute capability: major={cc[0]}, minor={cc[1]}")
226+
print(f"Architecture: sm_{cc[0]}{cc[1]}")
227+
print(f"PCI bus id={device.pci_bus_id}")
228+
print_device_properties(prop)
229+
print("*****************************************************\n\n")
230+
231+
232+
if __name__ == "__main__":
233+
assert len(sys.argv) == 1, "no command-line arguments expected"
234+
show_device_properties()

0 commit comments

Comments
 (0)