Skip to content

Latest commit

 

History

History
148 lines (120 loc) · 7.56 KB

Install_guide.md

File metadata and controls

148 lines (120 loc) · 7.56 KB

Intel(R) XPU Manager Installation Guide

Requirements

  • Intel GPU driver (GPU Driver Installation Guides). XPU-SMI is included in the GPU driver repository. If you want to use XPU Manager, please uninstall xpu-smi and install XPU Manager.
  • Intel(R) Graphics Compute Runtime for oneAPI Level Zero (intel-level-zero-gpu and level-zero in package repositories)
  • Intel(R) Graphics System Controller Firmware Update Library (intel-gsc in package repositories)
  • Intel(R) Media Driver (intel-media-va-driver-non-free or intel-media in package repositories)
  • Intel(R) Media SDK Utilities (libmfx-tools or intel-mediasdk-utils in package repositories)
  • Intel(R) oneVPL GPU Runtime (libmfxgen1 in package repositories)
  • Intel(R) Metrics Library for MDAPI (intel-metrics-library or libigdml1 in package repositories)
  • Intel(R) Metrics Discovery Application Programming Interface (intel-metrics-discovery or libmd1 in package repositories)

intel-metrics-library (libigdml1) and intel-metrics-discovery (libmd1) are optional. You may use the parameter like "--force-all" to ignore them when installing Intel(R) XPU Manager.

DEB install on Ubuntu

After adding the repository and installing the required kernel/run-time packages on GPU Driver Installation Guides, you may download the latest installer package from the Github release page and run apt command below to install XPU Manager and the required dependencies.

sudo apt install ./xpumanager.xxxxxxxx.xxxxxx.xxxxxxxx.deb

DEB uninstall

sudo dpkg -r xpumanager

RPM install on RHEL and CentOS Stream

After importing the repository and required kernel/run-time packages on GPU Driver Installation Guides, you may download the latest installer package from the Github release page and run dnf command below to install XPU Manager and the required dependencies.

sudo dnf install xpumanager.xxxxxxxx.xxxxxx.xxxxxxxx.rpm

RPM install on SLES

After importing the repository and required kernel/run-time packages on GPU Driver Installation Guides, you may download the latest installer package from the Github release page and run zypper command below to install XPU Manager and the required dependencies.

sudo zypper install xpumanager.xxxxxxxx.xxxxxx.xxxxxxxx.rpm

Start to use XPU Manager

By default, XPU Manager is installed the folder, /usr/bin, /usr/lib and /usr/lib64. The command line tool is /usr/bin/xpumcli. Please refer to "CLI_user_guide.md" for how to use the command line tool.

RPM relocation install

rpm -i --prefix=/usr/local xpumanager.xxxxxxxx.xxxxxx.xxxxxxxx.rpm

You need set the environmental variable LD_LIBRARY_PATH if you change the installation folder.

RPM uninstall

sudo rpm -e xpumanager

How to enable or disable some daemon monitor metrics

By default, XPU Manager has provided as many GPU metrics as possible without changing the system settings. You may follow the steps below to collect more metrics or disable some metrics.

  1. edit file "/lib/systemd/system/xpum.service" or "/etc/systemd/system/xpum.service" in some system. add "-m metric-indexes" to ExecStart. Use "/usr/bin/xpumd -h" to get detailed info.
    Sample: ExecStart=/usr/bin/xpumd -p /var/xpum_daemon.pid -d /usr/lib/xpum/dump -m 0,4-38
  2. Run command "sudo systemctl daemon-reload"
  3. Run command "sudo systemctl restart xpum"

Metric types:

  1. GPU Utilization (%), GPU active time of the elapsed time, per tile
  2. GPU EU Array Active (%), the normalized sum of all cycles on all EUs that were spent actively executing instructions, per tile (Disabled by default)
  3. GPU EU Array Stall (%), the normalized sum of all cycles on all EUs during which the EUs were stalled. Per tile. At least one thread is loaded, but the EU is stalled, per tile. (Disabled by default)
  4. GPU EU Array Idle (%), the normalized sum of all cycles on all cores when no threads were scheduled on a core. per tile. (Disabled by default)
  5. GPU Power (W), per tile
  6. GPU Energy Consumed (J), per tile
  7. GPU Frequency (MHz), per tile
  8. GPU Core Temperature (Celsius Degree), per tile
  9. GPU Memory Used (MiB)
  10. GPU Memory Utilization (%), per tile
  11. GPU Memory Bandwidth Utilization (%), per tile
  12. GPU Memory Read (kB), per tile
  13. GPU Memory Write (kB), per tile
  14. GPU Memory Read Throughput(kB/s), per tile
  15. GPU Memory Write Throughput(kB/s), per tile
  16. GPU Compute Engine Group Utilization (%), per tile
  17. GPU Media Engine Group Utilization (%), per tile
  18. GPU Copy Engine Group Utilization (%), per tile
  19. GPU Render Engine Group Utilization (%), per tile
  20. GPU 3D Engine Group Utilization (%), per tile
  21. Reset Counter, per GPU
  22. Programming Errors, per tile
  23. Driver Errors, per tile
  24. Cache Errors Correctable, per tile
  25. Cache Errors Uncorrectable, per tile
  26. Display Errors Correctable, per tile (Not supported so far)
  27. Display Errors Uncorrectable, per tile (Not supported so far)
  28. Memory Errors Correctable, per tile
  29. Memory Errors Uncorrectable, per tile
  30. GPU Requested Frequency, per tile
  31. GPU Memory Temperature, per tile
  32. GPU Frequency Throttle Ratio, per tile (Not supported so far)
  33. GPU PCIe Read Throughput (kB/s), per GPU (Disabled by default)
  34. GPU PCIe Write Throughput (kB/s), per GPU (Disabled by default)
  35. GPU PCIe Read (bytes), per GPU (Disabled by default)
  36. GPU PCIe Write (bytes), per GPU (Disabled by default)
  37. GPU Engine Utilization, per GPU engine
  38. Fabric Throughput (kB/s), per tile
  39. Throttle reason, per tile

Change the system settings to enable some GPU advanced metrics

  • GPU PCIe Read/Write Throughput: if these metrics are enabled, XPU Manager automatically loads MSR module by command 'modprobe msr', but XPU Manager will not automatically unload the MSR module. If you want to unload it, please run the command 'modprobe -r msr'.

Disable some metrics to reduce the CPU usage of XPU Manager daemon

  • We have tried our best to reduce the CPU usage of XPU Manager daemon. If you have many GPUs (10+) and still have concern with the CPU usage of XPU Manager daemon, you may disable some the RAS related metrics below to reduce the CPU usage further.
      1. Reset Counter, per GPU
      1. Programming Errors, per tile
      1. Driver Errors, per tile
      1. Cache Errors Correctable, per tile
      1. Cache Errors Uncorrectable, per tile
      1. Display Errors Correctable, per tile (Not supported so far)
      1. Display Errors Uncorrectable, per tile (Not supported so far)
      1. Memory Errors Correctable, per tile
      1. Memory Errors Uncorrectable, per tile

GPU memory ECC on/off

XPU Manager provides the GPU memory ECC on/off feature based on IGSC. GPU memory ECC on/off starts to work since IGSC 0.8.4. If you want to use this feature, please make sure that you install IGSC 0.8.4 or newer version.

How to install the latest version of libcurl on CentOS 7

CentOS 7 still has the old version of libcurl. If you need update the AMC firmware through Redfish host interface, please follow the steps below to build and install libcurl.

yum update -y

yum install wget gcc openssl-devel make -y

wget https://curl.se/download/curl-7.56.1.tar.gz

tar xzf curl-7.56.1.tar.gz

cd curl-7.56.1

./configure --with-openssl --prefix=/usr

make 

sudo make install

curl --version

XPU Manager Windows CLI tool

The Windows GPU driver version should be 31.0.101.3902 or newer.