Community
cancel
Showing results for 
Search instead for 
Did you mean: 
Highlighted
New Contributor III
94 Views

Step By Step OS + Kernel Installation On Xeon Phi

Jump to solution

Hi All,

I have Ninja Developer Platform which was running CentOS and file system got corrupted. Will have to do fresh installation, can Intel share step by step guide on how to do it? Do I need to follow the user guide shared here

The file size of CentOS 7.3 in Xeon Phi Software page is very small. So, do I need to install CentOS 7.3 and then patch these files shared here?

Hopefully, I won't loose any kernel functionality.

Thanks.

Chetan Arvind Patil
0 Kudos

Accepted Solutions
Highlighted
New Contributor I
94 Views

Hi Chetan,

I just installed a similar system from scratch - after 4 tries I think I got it working correctly.

- You need to download CentOS 7.3 without netinstall otherwise it upgrades to 7.4
Did you have any window manager or did you just have a CLI? I just installed a CLI.

- Install the Xeon Phi software in the link

- If you have a compiler license install it otherwise install the Intel compiler runtime via yum

- now you should be able to run the performance test micprun

- If you use gcc you need to install a newer version (remember to install the c++ too)
https://centos.pkgs.org/7/centos-sclo-rh-testing/devtoolset-7-gcc-7.1.1-2.1.el7.x86_64.rpm.html

 

View solution in original post

0 Kudos
13 Replies
Highlighted
New Contributor I
95 Views

Hi Chetan,

I just installed a similar system from scratch - after 4 tries I think I got it working correctly.

- You need to download CentOS 7.3 without netinstall otherwise it upgrades to 7.4
Did you have any window manager or did you just have a CLI? I just installed a CLI.

- Install the Xeon Phi software in the link

- If you have a compiler license install it otherwise install the Intel compiler runtime via yum

- now you should be able to run the performance test micprun

- If you use gcc you need to install a newer version (remember to install the c++ too)
https://centos.pkgs.org/7/centos-sclo-rh-testing/devtoolset-7-gcc-7.1.1-2.1.el7.x86_64.rpm.html

 

View solution in original post

0 Kudos
Highlighted
New Contributor III
94 Views

Hi Per,

Thank you for sharing details.

Can you please share the link to CentOS 7.3 without netinstall? CentOS website has link that will download 7 and I am not sure how it will get 7.3 version if I download that?

By window manage, if you mean GNOME or KDE, then yes it had the interface. 

Thanks.

Chetan Arvind Patil
0 Kudos
Highlighted
New Contributor I
94 Views
0 Kudos
Highlighted
New Contributor I
94 Views

Hi again,

I have just run some tests on OpenFOAM I compiled with gcc 7.1 and Intel MPI which does not work - I get some odd behavior where the calculations suddenly diverges and crashes. I don't know where the problem is but I will try different compile MPI combinations. The same calculations on an older Xeon works fine

Best regards,

Per

0 Kudos
Highlighted
New Contributor III
94 Views

Hi Per,

Do you have the same hardware system I linked above? Am trying to get help from them too.

I think important step is to configure all the performance libraries that will either come bundle with license of Intel® Parallel Studio XE or using these performance libraries.

Someone from Intel on this forum should be able to help as well, as system do crash and getting back to same system configuration is important for Xeon Phi architecture performance.

Thanks.

Chetan Arvind Patil
0 Kudos
Highlighted
New Contributor I
94 Views

Hi Chetan,

It looks the same - a Supermicro SYS-5038K-i-NF1 but I bought it through a Danish vendor without preinstalled software.

I don't have a Parallel Studio XE license so I tried the performance libraries. Do they need a special configuration to work? That could explain my crashes.

Right now I try to compile OpenFOAM with another MPI - I'll see if it works in the morning...

Best regards,

 

Per

0 Kudos
Highlighted
New Contributor III
94 Views

Hi Per,

I know following are required to be exported to make use of MKL etc:

source /opt/intel/bin/compilervars.sh intel64
source /opt/intel/mkl/bin/mklvars.sh intel64
source /opt/intel/impi/2017.1.132/intel64/bin/mpivars.sh intel64

I haven't got any response from vendor. Things will get clear out when they do. I think this shouldn't be as difficult as it looks.

Thanks.

Chetan Arvind Patil
0 Kudos
Highlighted
New Contributor III
94 Views

Hi Chetan,

I've got the same hardware as you and am running CentOS 7.3 on it; I have not seen much performance impact of installing the xppsl tools - the only *kernel* patch that is added is a module known as 'zonesort_module' : this module has some impact on performance of accessing memory (regular&MCDRAM, I believe) when running in SNC4 topology. 

Also, in the xppsl-1.5.3-centos7.3.tar file the source RPMs are included, so you can rebuild the RPMs for CentOS 7.4 as well - it's a pity that Intel did release a version for RHEL 7.4 but did not for CentOS 7.4 ...

 

0 Kudos
Highlighted
New Contributor I
94 Views

Hi Chetan,

gcc 5.3 and gcc 6.3 works with Intel MPI and give similar execution times on OpenFOAM so the configuration files works :-)

Did you get 1 year of support/upgrade on the Intel® Parallel Studio XE with the Ninja Developer Platform?

Best regards,

Per

0 Kudos
Highlighted
New Contributor III
94 Views

Hi Per,

There is support from vendor. For Parallel Studio, I think Intel provides too if produce is licensed and registered.

Thanks.

Chetan Arvind Patil
0 Kudos
Highlighted
New Contributor I
94 Views

Hi Chetan,

It looks like there is some improvements in XE 2018 for Xeon Phi, so I was wondering if the upgrade was included.

Best regards,

Per

0 Kudos
Highlighted
New Contributor III
94 Views

Hi Per,

I think the license is perpetual and upgrades are there. But, what improvments are done only Intel would know. I am using 2017 as it's more stable and was using that only earlier. I am more concerned with one click all library setup which PSXE provides. 

Also, for me MKL + OpenMP are helpful in exploring the thread distribution and affinity, so that way it's helpful in terms of optimization and speedup. Also, making use of cluster + memory mode can also substantially improve performance by reducing latency and increasing memoyr throughput.

I am not sure about your case, but you should look for Intel's distribution for OpenFOAM (this one is from OpenFOAM and not from Intel). 

Thanks.

Chetan Arvind Patil
0 Kudos
Highlighted
New Contributor I
94 Views

Hi Chetan,

Thanks, I will try to look at the Intel OpenFOAM

Best regards,

Per

0 Kudos