mirror of
https://github.com/NVIDIA/cuda-samples.git
synced 2024-10-06 14:29:16 +08:00
33 lines
1.3 KiB
Plaintext
33 lines
1.3 KiB
Plaintext
|
[simpleMultiCopy] - Starting...
|
||
|
|
> Using CUDA device [0]: NVIDIA H100 PCIe
|
||
|
|
[NVIDIA H100 PCIe] has 114 MP(s) x 128 (Cores/MP) = 14592 (Cores)
|
||
|
|
> Device name: NVIDIA H100 PCIe
|
||
|
|
> CUDA Capability 9.0 hardware with 114 multi-processors
|
||
|
|
> scale_factor = 1.00
|
||
|
|
> array_size = 4194304
|
||
|
|
|
||
|
|
|
||
|
|
Relevant properties of this CUDA device
|
||
|
|
(X) Can overlap one CPU<>GPU data transfer with GPU kernel execution (device property "deviceOverlap")
|
||
|
|
(X) Can overlap two CPU<>GPU data transfers with GPU kernel execution
|
||
|
|
(Compute Capability >= 2.0 AND (Tesla product OR Quadro 4000/5000/6000/K5000)
|
||
|
|
|
||
|
|
Measured timings (throughput):
|
||
|
|
Memcpy host to device : 0.610592 ms (27.476966 GB/s)
|
||
|
|
Memcpy device to host : 0.694048 ms (24.172991 GB/s)
|
||
|
|
Kernel : 0.033408 ms (5021.915549 GB/s)
|
||
|
|
|
||
|
|
Theoretical limits for speedup gained from overlapped data transfers:
|
||
|
|
No overlap at all (transfer-kernel-transfer): 1.338048 ms
|
||
|
|
Compute can overlap with one transfer: 1.304640 ms
|
||
|
|
Compute can overlap with both data transfers: 0.694048 ms
|
||
|
|
|
||
|
|
Average measured timings over 10 repetitions:
|
||
|
|
Avg. time when execution fully serialized : 1.325424 ms
|
||
|
|
Avg. time when overlapped using 4 streams : 1.203120 ms
|
||
|
|
Avg. speedup gained (serialized - overlapped) : 0.122304 ms
|
||
|
|
|
||
|
|
Measured throughput:
|
||
|
|
Fully serialized execution : 25.315998 GB/s
|
||
|
|
Overlapped using 4 streams : 27.889513 GB/s
|