Idphix
From Digitalis
(Difference between revisions)
(→Setup your access to the Xeon Phi) |
(→idphix-default 20131213) |
||
| Line 307: | Line 307: | ||
GDDR Frequency : 2500000 kHz | GDDR Frequency : 2500000 kHz | ||
GDDR Voltage : 1501000 uV | GDDR Voltage : 1501000 uV | ||
| + | |||
| + | === idphix-default 20140219 === | ||
| + | * Fixed mic-setup-my-user which was broken since the upgrade to MPSS 3.1 | ||
| + | * Add mount point for /grid5000 | ||
| + | * Change /opt/intel link to /grid5000/software/intel, which as of 2014-02-17 links to Intel Cluster Studio XE 2013.1 (icc 14) | ||
| + | * Add packages: | ||
| + | ** emacs-common-23.1-25.el6.x86_64 | ||
| + | ** emacs-nox-23.1-25.el6.x86_64 | ||
| + | ** autoconf-2.63-5.1.el6.noarch | ||
| + | ** automake-1.11.1-4.el6.noarch | ||
| + | ** automake16-1.6.3-18.el6.1.noarch | ||
| + | ** automake15-1.5-27.el6.1.noarch | ||
| + | ** automake14-1.4p6-19.2.el6.noarch | ||
| + | ** autoconf213-2.13-20.1.el6.noarch | ||
| + | ** autoconf-archive-2012.09.08-1.el6.noarch | ||
| + | ** tcl-8.5.7-6.el6.x86_64 | ||
| + | ** expect-5.44.1.15-5.el6_4.x86_64 | ||
| + | ** dejagnu-1.4.4-17.el6.noarch | ||
| + | ** tig-0.17-1.el6.x86_64 | ||
== References == | == References == | ||
Revision as of 14:45, 19 February 2014
This page describes the first steps to play with the Xeon Phi installed in the Digitalis platform in idphix.
Contents |
Overview
A Xeon Phi 5110P is installed in the idphix machine
The setup is the following:
We use an internal network on the host (idphix) for the Xeon Phi: 192.168.1.0/24.
This gives the following scheme:
=======================================
| ------------- ----------- |
|| host | | Xeon Phi | |
|| idphix | |idphix-mic0| |
Grid5000 network <---> eth0 mic0 <----> mic0 | |
172.16.21.6 192.168.1.254 192.168.1.1 | |
|| | | | |
| ------------- ----------- |
=======================================
How to experiment ?
For now (2013-09-13) the operating system is a Centos 6 (OS officially supported by Intel for the Xeon Phi).
Use the machine as follows:
Connect to digitalis/grid5000
workstation$ ssh digitalis.grenoble.g5k
In case of any issue (network issue, connection hanging...) please read more info here or ask on the Digitalis mailing list.
Reserve the machine/run a job
digitalis$ oarsub -p "machine='idphix'" -l walltime=2 -I
Setup your access to the Xeon Phi
(This might not be relevant if your are using some level of abstraction (COI, ...)
- WARNING
- there is currently a bug due to the last version of Intel MPSS, please report any issue to the mailing list.
Make sure you created your G5K ssh keys (keys without passephrase protection, dedicated to G5K and stored in your NFS home directory)
Run:
idphix$ sudo /usr/local/bin/mic-setup-my-user
If you get the error:
Error: Cannot get your SSH keys, please run 'chmod 755 ~/.ssh && chmod 644 ~/.ssh/{authorized_keys,id_*.pub}'
Please run:
idphix$ chmod 755 ~/.ssh && chmod 644 ~/.ssh/{authorized_keys,id_*.pub}
Output should be:
idphix$ sudo /usr/local/bin/mic-setup-my-user *** Configure user pneyron:users (10106:8000) on mic0 *** Copy user files to mic0 id_rsa.pub 100% 420 0.4KB/s 00:00 authorized_keys 100% 916 0.9KB/s 00:00 .profile 100% 62 0.1KB/s 00:00 *** Force creation of pneyron:users on mic0 and fix file ownership and permission Done
You shoud now be able to ssh to the coprocessor:
idphix$ ssh mic0 idphix-mic0$ uname -a Linux idphix-mic0.grenoble.grid5000.fr 2.6.38.8-g5f2543d #2 SMP Tue Apr 30 14:05:06 PDT 2013 k1om k1om k1om GNU/Linux
Intel compiler, compilation for the MIC
Intel compiler (icc & co) is installed in /opt/intel/applis/digitalis/intel.
It is configured to use Inria's license server (FlexLM).
This licence server provides 2 tokens which are shared among all Inria's users. Only 2 compilations can be performed at a time.
If icc complains about the license (error message talking about FlexLM), please report the issue to digitalis@lists.grid5000.fr
Rebooting the MIC
If needed, you can reboot the MIC from within a exclusive job (doing so in shared job session would risk annoying other users).
oarsub -t exclusive -p "machine='idphix'" ....
Then, you will be able to run
sudo /usr/local/bin/mic-reboot
(no password should be requested).
Rebooting the MIC takes some time. You can look at the output of dmesg or miccrtl -s meanwhile.
Data on the mic: no persistence
Please mind the fact that data pushed to the mic are not persistent. Your home directory on idphix-mic0 could be emptied for various reasons.
System changelogs
idphix-default 20130926
- Centos 6
- Linux 2.6.32-358.el6.x86_64
- MPSS packages
intel-mic-flash-2.1.386-2.el6.x86_64 intel-mic-kmod-2.1.6720-13.2.6.32.358.el6.x86_64 intel-mic-sysmgmt-2.1.6720-13.el6.x86_64 intel-mic-gpl-2.1.6720-13.el6.x86_64 intel-mic-mpm-2.1.6720-13.el6.x86_64 intel-mic-2.1.6720-13.el6.x86_64 intel-mic-micmgmt-2.1.6720-13.2.6.32.358.el6.x86_64 intel-mic-perf-data-2.1.6720-13.el6.x86_64 intel-mic-gdb-2.1.6720-13.el6.x86_64 intel-mic-cdt-2.1.6720-13.el6.x86_64 intel-mic-perf-2.1.6720-13.el6.x86_64
- Intel Cluster Studio XE 2013
- MIC info
[root@idphix ~]# /opt/intel/mic/bin/micinfo MicInfo Utility Log Created Mon Dec 9 20:13:37 2013 System Info HOST OS : Linux OS Version : 2.6.32-358.el6.x86_64 Driver Version : 6720-13 MPSS Version : 2.1.6720-13 Host Physical Memory : 66077 MB Device No: 0, Device Name: mic0 Version Flash Version : 2.1.02.0386 SMC Firmware Version : 1.14.4616 SMC Boot Loader Version : 1.8.4326 uOS Version : 2.6.38.8-g5f2543d Device Serial Number : ADKC32100375 Board Vendor ID : 0x8086 Device ID : 0x2250 Subsystem ID : 0x2500 Coprocessor Stepping ID : 3 PCIe Width : x16 PCIe Speed : 5 GT/s PCIe Max payload size : 256 bytes PCIe Max read req size : 512 bytes Coprocessor Model : 0x01 Coprocessor Model Ext : 0x00 Coprocessor Type : 0x00 Coprocessor Family : 0x0b Coprocessor Family Ext : 0x00 Coprocessor Stepping : B1 Board SKU : INVALID SKU ECC Mode : Enabled SMC HW Revision : Product 225W Passive CS Cores Total No of Active Cores : 60 Voltage : 985000 uV Frequency : 1052631 kHz Thermal Fan Speed Control : N/A Fan RPM : N/A Fan PWM : N/A Die Temp : 35 C GDDR GDDR Vendor : Elpida GDDR Version : 0x1 GDDR Density : 2048 Mb GDDR Size : 7936 MB GDDR Technology : GDDR5 GDDR Speed : 5.000000 GT/s GDDR Frequency : 2500000 kHz GDDR Voltage : 1501000 uV
idphix-default 20131209
- Updated Centos packages, except linux kernel
- Moved Intel Cluster Studio XE 2013 to NFS /applis/digitalis/intel
- No change to the MIC software (MPSS)
idphix-default 20131213
- Activated Hyperthreading
[root@idphix ~]# lstopo --no-caches
Machine (63GB) + Socket L#0
Core L#0
PU L#0 (P#0)
PU L#1 (P#8)
Core L#1
PU L#2 (P#1)
PU L#3 (P#9)
Core L#2
PU L#4 (P#2)
PU L#5 (P#10)
Core L#3
PU L#6 (P#3)
PU L#7 (P#11)
Core L#4
PU L#8 (P#4)
PU L#9 (P#12)
Core L#5
PU L#10 (P#5)
PU L#11 (P#13)
Core L#6
PU L#12 (P#6)
PU L#13 (P#14)
Core L#7
PU L#14 (P#7)
PU L#15 (P#15)
- Updated MPSS 2.1 -> 3.1.1
[root@idphix ~]# rpm -qa | grep -e intel -e mpss mpss-miccheck-3.1.1-r1.glibc2.12.2.x86_64 mpss-sysmgmt-python-3.1.1-1.glibc2.12.2.x86_64 mpss-sciftutorials-3.1.1-1.glibc2.12.2.x86_64 mpss-eclipse-cdt-mpm-3.1.1-1.glibc2.12.2.x86_64 mpss-myo-doc-3.1.1-1.glibc2.12.2.x86_64 mpss-coi-doc-3.1.1-1.glibc2.12.2.x86_64 mpss-flash-3.1.1-1.glibc2.12.2.x86_64 mpss-myo-3.1.1-1.glibc2.12.2.x86_64 mpss-metadata-3.1.1-1.glibc2.12.2.x86_64 mpss-myo-dev-3.1.1-1.glibc2.12.2.x86_64 intel-composerxe-compat-k1om-3.1.1-1.x86_64 mpss-micmgmt-doc-3.1.1-1.glibc2.12.2.x86_64 mpss-sysmgmt-micsmc-gui-3.1.1-1.glibc2.12.2.x86_64 mpss-miccheck-bin-3.1.1-r1.glibc2.12.2.x86_64 mpss-sysmgmt-micras-3.1.1-1.glibc2.12.2.x86_64 mpss-modules-2.6.32-358.el6.x86_64-3.1.1-1.el6.x86_64 mpss-modules-dev-2.6.32-358.el6.x86_64-3.1.1-1.el6.x86_64 mpss-sciftutorials-doc-3.1.1-1.glibc2.12.2.x86_64 mpss-license-3.1.1-1.glibc2.12.2.x86_64 mpss-rasmm-kernel-3.1.1-1.glibc2.12.2.x86_64 mpss-micmgmt-python-3.1.1-1.glibc2.12.2.x86_64 mpss-coi-3.1.1-1.glibc2.12.2.x86_64 mpss-modules-headers-3.1.1-1.glibc2.12.2.x86_64 mpss-sdk-k1om-3.1.1-1.x86_64 mpss-metadata-dev-3.1.1-1.glibc2.12.2.x86_64 mpss-daemon-dev-3.1.1-1.glibc2.12.2.x86_64 mpss-coi-dev-3.1.1-1.glibc2.12.2.x86_64 mpss-micmgmt-3.1.1-1.glibc2.12.2.x86_64 mpss-mpm-doc-3.1.1-1.glibc2.12.2.x86_64 mpss-mpm-3.1.1-1.glibc2.12.2.x86_64 mpss-boot-files-3.1.1-1.glibc2.12.2.x86_64 mpss-daemon-3.1.1-1.glibc2.12.2.x86_64 mpss-modules-headers-dev-3.1.1-1.glibc2.12.2.x86_64
- Updated the MIC SMC and Flash (part of the MPSS upgrade)
[root@idphix ~]# micinfo MicInfo Utility Log Created Fri Dec 13 19:00:30 2013 System Info HOST OS : Linux OS Version : 2.6.32-358.el6.x86_64 Driver Version : 3.1.1-1 MPSS Version : 3.1.1 Host Physical Memory : 66076 MB Device No: 0, Device Name: mic0 Version Flash Version : 2.1.02.0390 SMC Firmware Version : 1.16.5078 SMC Boot Loader Version : 1.8.4326 uOS Version : 2.6.38.8+mpss3.1.1 Device Serial Number : ADKC32100375 Board Vendor ID : 0x8086 Device ID : 0x2250 Subsystem ID : 0x2500 Coprocessor Stepping ID : 3 PCIe Width : x16 PCIe Speed : 5 GT/s PCIe Max payload size : 256 bytes PCIe Max read req size : 512 bytes Coprocessor Model : 0x01 Coprocessor Model Ext : 0x00 Coprocessor Type : 0x00 Coprocessor Family : 0x0b Coprocessor Family Ext : 0x00 Coprocessor Stepping : B1 Board SKU : B1PRQ-5110P/5120D ECC Mode : Enabled SMC HW Revision : Product 225W Passive CS Cores Total No of Active Cores : 60 Voltage : 1021000 uV Frequency : 1052631 kHz Thermal Fan Speed Control : N/A Fan RPM : N/A Fan PWM : N/A Die Temp : 36 C GDDR GDDR Vendor : Elpida GDDR Version : 0x1 GDDR Density : 2048 Mb GDDR Size : 7936 MB GDDR Technology : GDDR5 GDDR Speed : 5.000000 GT/s GDDR Frequency : 2500000 kHz GDDR Voltage : 1501000 uV
idphix-default 20140219
- Fixed mic-setup-my-user which was broken since the upgrade to MPSS 3.1
- Add mount point for /grid5000
- Change /opt/intel link to /grid5000/software/intel, which as of 2014-02-17 links to Intel Cluster Studio XE 2013.1 (icc 14)
- Add packages:
- emacs-common-23.1-25.el6.x86_64
- emacs-nox-23.1-25.el6.x86_64
- autoconf-2.63-5.1.el6.noarch
- automake-1.11.1-4.el6.noarch
- automake16-1.6.3-18.el6.1.noarch
- automake15-1.5-27.el6.1.noarch
- automake14-1.4p6-19.2.el6.noarch
- autoconf213-2.13-20.1.el6.noarch
- autoconf-archive-2012.09.08-1.el6.noarch
- tcl-8.5.7-6.el6.x86_64
- expect-5.44.1.15-5.el6_4.x86_64
- dejagnu-1.4.4-17.el6.noarch
- tig-0.17-1.el6.x86_64
References
- Main Intel resource page
- Documentation to install and configure the Xeon Phi
- Coding docs
- Intel webinars
- Directive-based programming extensions
- Intrisics on Xeon Phi
- Memory transfer
- Getting the highest bandwidth possible out of your Phi application:
