LHC@home

Hello,
Do you have an estimated date for the GPU support for the LHC project? the gain in computation time will be phenomenal.

Thx

Read the first post in this thread.

First, thankyou for all the feed back,and offers of help.
I shall be reviewing all this with Igor tomorrow I hope
(Massimo is absent for a week).

My current absolute priority is to check the physics results
from the new executables produced by Intel ifort. I had to
introduce rounding of the input data due to a strange 1 ULP
difference between ifort on Linux an Windows. I am
hoping, indeed praying, that the results will be valid.

If so, I can pass directly to producig new executables which
will allow for systems with/without SSE2. Should be possible.
I have my Mac but will produce a MAC version only after the
above issues are resolved. (I might also have problems to access
the ifort compiler for MAC as although CERN has an openlab
project with Intel I have to pay for a licence to debug the compiler!)

Then we can worry about credits and graphics, and many thanks to
Eric Myers for his detailed and constructive comments and to another
volunteer who offered to help implement them.

Well; I had hoped to be compiler independent by now, but .....
Only then can I think about GPUs. I had hoped by
May,then before the end of year, but it now looks lik next year, and
then I have to write it up and publish. Priority to getting
a high quality, well supported service (with graphics :-).

I will report soonest on the result of the physics checks. Eric

---

Any progress on the physics checks? the .09 version did run a lot faster on my machines that have sse2.

---

Well I have been on vacation since 1st November but
still trying to do what I can. (The sunshine makes
it difficult to read the screen on my netbook! :-)
In addition the simple putty service often hangs or
breaks the connection to CERN which is very frustrating.
I therefore am using batch jobs instead of just running
scripts to submit work and get results back. Further
the CERN services are very busy and evrything is rather
slow.

I am trying to run a series of studies lhc300/400/500 which we
have already run on LSF batch at CERN. The goal is to dtermine
if the physics results are valid using the new SixTrack, the
new ifort compiler and the new DBL(SNGL()) of input data.
These jobs are "only" 10,000 turns and many cases are unstable.
Hence the shorter execution times. I cannot submit work fast
enough either. I am supposed to produce the procedures for
physics verification but I am stymied since I have no X-windows
and cannot check my gnuplot scripts. That will have to wait until
I return to CERN on 15th November.

I will also try and agree with Igor an rcp type system for
submitting jobs and retrieving results which will be much more
efficient and avoid using intermediate AFS buffers. Also if I
can get agreement I would set up a dedicated machine (or two)
for the physicists to use. AFS would be used for the scripts but
everything else would be on a say 1 Terabyte local disk system.

I have produced a new executable which uses SSE2 if available
and I am reviewing with Intel. It would be great if somebody has
an old Pentium 3 machine (NO SSE2) Linux so we could check it
works on "any" PC. I cannot use Windows at CERN remotely so that
too will have to await the end of my vacation.

Maybe I shall manage to progress on compiler independence. My personal
goal of 0ULP on any IEEE machine with any Fortran compiler at any
optimisation might get closer. I think this is vital because I can
then be sure we have no compiler bugs affecting results and I can
send convincing e-mails to you when results are treated as invalid.

I haven' forgotten the MACs. I'll try and produce that executable as well
at CERN.

A lot depends on finding an acceptable solution to the formatted I/O
problems. I am more and more thinking I shall have to use C++ strtod()
decimal to binary conversion as I can't wait for 2003 standard
compliant compilers. Still maybe the DBLE/SNGL() workround will be
satisfactory even if inelegant.

I haven't forgotten graphics; just can't do everything right now.

Thanks for all your support and patience. Please understand that there
may be a pause in the work submission while all this is sorted out.
There is no point in submitting million turn jobs and LHC upgrade studies
until we are confident in the results. Eric.

---

Workingon it. I think I have it OK now
Versin 4311 but I am reviewing with Intel
next week- Eric.

---

Thanks Eric,

Don't worry about the graphics....the LHC is more important.
We are always ready for a new batch of tasks here.

(btw it is 4am here so I'm having no trouble with the sun shining through the window here )

---

Volunteer Mad Scientist For Life

Thanks for the update Eric, even though a lot of it went over my head :) Hope your having a good holiday!

Maybe I shall manage to progress on compiler independence. My personal
goal of 0ULP on any IEEE machine with any Fortran compiler at any
optimisation might get closer. I think this is vital because I can
then be sure we have no compiler bugs affecting results and I can
send convincing e-mails to you when results are treated as invalid.

Achieving 0ULP across all Fortran compilers is a worthy goal from an academic perspective and you've done very well in that regard. However, from a practical perspective it makes more sense to run exactly the same version of Sixtrack on all platforms by running it in a virtual machine. Test4Theory uses that approach and it works very well. Another benefit of using a virtual machine is that you wouldn't have to make a version for OS X and people running BSD could crunch Sixtrack too.

The oldest CPU I have is a pentium 4.

Maybe it would be easier to find someone with an Athlon XP/MP or Duron/Sempron as AMD didn't implament SSE2 until a few years after Intel.

Eric,

I have two hosts running SixTrack without sse2. However, they're both running Win2k not Linux. This one is a Pentium 3, and this one a Pentium 4.

Thanks for the updates, they're much appreciated, even if I don't understand a lot of it.

John.

It would be great if somebody has
an old Pentium 3 machine (NO SSE2) Linux so we could check it
works on "any" PC.

I have an old motherboard with Socket A CPU (I don't remember if under the fan there's an Athlon or a Duron) that support only SSE (no SSE2). If it can be useful I can find some pieces to build a PC on the fly and install a Linux onto that.

---

I have a bunch of AthlonXP/Sempron boxes already running Linux (SLC4)... and a Pentium II, if 10 000 turns is a good enough test!

A lot of time passed since last update... any news on developing the new client version? The 530.10 is really slow :-(


Thanks

---