Dualsphysics only consuming 54% GPU-Util Tesla K80
Hey folks -
I've been running some simulations on a Tesla K80 (driver 376.84) and noticed that the GPU utilisation is only running at ~50% (50% of half a card, so 25% of the whole card), using 114 out of 11447MB of GPU memory, and less than 5% RAM.
Running nvidia-smi -q -d PERFORMANCE I can see that the card is being throttled due to "Applications Clocks Setting" (Power-cap and/or overheating aren't an issue).
Looking at the clock the card is running at 562Mhz (out of 875).
Wondering whether -blocksize: will have an impact on this, but feel there is some inherent software control.
Running Windows Server 2016, 56GB Ram, 380GB SSD.
Has anyone seen this behaviour or know how to remove this limitation?
Kind regards,
John
Output:
==============NVSMI LOG==============
Timestamp : Wed Jun 14 12:12:24 2017
Driver Version : 376.84
Attached GPUs : 1
GPU 1B88:00:00.0
Performance State : P0
Clocks Throttle Reasons
Idle : Not Active
Applications Clocks Setting : Active
SW Power Cap : Not Active
HW Slowdown : Not Active
Sync Boost : Not Active
Unknown : Not Active
Wed Jun 14 12:19:38 2017
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 376.84 Driver Version: 376.84 |
|-------------------------------+----------------------+----------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla K80 TCC | 1B88:00:00.0 Off | 0 |
| N/A 51C P0 59W / 149W | 114MiB / 11447MiB | 54% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 3164 C ...12016\Executables\DualSPHysics4_win64.exe 114MiB |
+-----------------------------------------------------------------------------+
GPU 1B88:00:00.0
Clocks
Graphics : 627 MHz
SM : 627 MHz
Memory : 2505 MHz
Video : 540 MHz
Applications Clocks
Graphics : 562 MHz
Memory : 2505 MHz
Default Applications Clocks
Graphics : 562 MHz
Memory : 2505 MHz
Max Clocks
Graphics : 875 MHz
SM : 875 MHz
Memory : 2505 MHz
Video : 540 MHz
I've been running some simulations on a Tesla K80 (driver 376.84) and noticed that the GPU utilisation is only running at ~50% (50% of half a card, so 25% of the whole card), using 114 out of 11447MB of GPU memory, and less than 5% RAM.
Running nvidia-smi -q -d PERFORMANCE I can see that the card is being throttled due to "Applications Clocks Setting" (Power-cap and/or overheating aren't an issue).
Looking at the clock the card is running at 562Mhz (out of 875).
Wondering whether -blocksize: will have an impact on this, but feel there is some inherent software control.
Running Windows Server 2016, 56GB Ram, 380GB SSD.
Has anyone seen this behaviour or know how to remove this limitation?
Kind regards,
John
Output:
==============NVSMI LOG==============
Timestamp : Wed Jun 14 12:12:24 2017
Driver Version : 376.84
Attached GPUs : 1
GPU 1B88:00:00.0
Performance State : P0
Clocks Throttle Reasons
Idle : Not Active
Applications Clocks Setting : Active
SW Power Cap : Not Active
HW Slowdown : Not Active
Sync Boost : Not Active
Unknown : Not Active
Wed Jun 14 12:19:38 2017
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 376.84 Driver Version: 376.84 |
|-------------------------------+----------------------+----------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla K80 TCC | 1B88:00:00.0 Off | 0 |
| N/A 51C P0 59W / 149W | 114MiB / 11447MiB | 54% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 3164 C ...12016\Executables\DualSPHysics4_win64.exe 114MiB |
+-----------------------------------------------------------------------------+
GPU 1B88:00:00.0
Clocks
Graphics : 627 MHz
SM : 627 MHz
Memory : 2505 MHz
Video : 540 MHz
Applications Clocks
Graphics : 562 MHz
Memory : 2505 MHz
Default Applications Clocks
Graphics : 562 MHz
Memory : 2505 MHz
Max Clocks
Graphics : 875 MHz
SM : 875 MHz
Memory : 2505 MHz
Video : 540 MHz
Comments
To make it fair I went back to windows, recompiled there to use CUDA8.0 which worked, but still seeing performance throttling.