Computación con clústeres

Few issues with mic and mpssd


As I described in another post, mpss-3.2.1 is running on kernel 3.13.10 perfectly (Fedora 20).

 I can run programs on the processor and have no problems except:

1. mpssd daemon  is taking 100% of one host cpu all the time.

2. If mpssd is not started the coprocessor fan is always on. When mpssd starts it goes off when there is no activity. Since people may not want to use the card all the time the default fan speed should be off or low it seems like.

GFlops on MIC

Hello Intel,

I wrote "a kind of" meta compiler to generate SIMD code on multi platform (x86, power, PHI, etc ...). I will present my work in the ISC 2014 in june. I am preparing the Super Computing conference, where I would like present to result on the Phi platform I have an issue.

I am trying to calculate the GFLOP/s of my application, a first approach will be to count the number of operations and divide by the elapsed time, as usually done for dgemm benchmark. Unfortunately I have thousands of lines ...

performance issue of operating on overlap area between two arrays

Hi, My application program needs a lot of operators (copy/accumulate...) on two arrays' overlap. In most cases, the overlap area is not very large. A typical size is 100x100*2.  I have made a simple benchmark for these functions in the attached file.

Unfortunately, I found the performance on MIC is much poor than Xeon CPU. For all three cases, one OpenMP thread is the best. 

Offload: 0.026 seconds

Native: 0.018 seconds

CPU: 0.0009 seconds 

Any suggestion on improving performance on MIC. Can we achieve better performance on MIC than that on CPU?

SSH logins to the Phi coprocessor fail if user's shell is not bash or sh

Hello all,

I'm just pointing this out because it is not documented somewhere (I think). Because the host's /etc/passwd lines are copied "as is" to the coprocessor, there is a chance that ssh logins will fail due to a non-existent shell. Specifically, the coprocessor OS image only contains sh, bash and ash. If you are using anything else logins will fail silently. This is a bug. Should I report it somewhere properly?


mic0: lost

Hi !

I begin with Xeon Phi and MPI. After a lot of troubles, yesterday I managed to run MPI programs on symmetric mode (both host and co-processor) as well as directly on mic0 after ssh.

But today I have troubles again.

MPSS starts correctly and 3 minutes later the status of mic0 is lost and I can not reset it.

dgetrf performance drop above 10000


I'm doing performance testing using mkl composer_xe_2013_sp1.2.144 with the icc version 14.0.2 20140120 compiler and randomly generatoed matrices.

when the matrix size gets above 10000, I notice a performance drop of around 100GFlops ( size 9217 peforms wih 211GFlops but 10241 only has 109)

I tried using the align option to try aligning the data to 64 byte boundaries and this did offer a general speed up and reduced the drop to around 50GFlops but the drop is still there.

Run Intel MPI without mpirun/mpiexec


I am wondering does Intel MPI support a MPI run without mpirun/mpiexec in the command line?

I know that in MPI-2 standard, it supports the “dynamic process” feature, i.e., dynamically generate/spawn processes from existing MPI process.

What I am trying to do here is 1) Firstly, launch a singleton MPI process without mpirun/mpiexec in the command line; 2) Secondly, use MPI_Comm_spawn to spawn a set of process on the different host machines.

Buyers Remorse

So I can't set up the Xeon Phi in Linux like I need it without it deleting my entire hard drive and I refuse to spend another $3000 just to be able to compile on windows. Now after spending days on the phone with support and getting no help at all I am left with a $2000 piece of crap I can't even use.


Buyer beware!

Suscribirse a Computación con clústeres