LHC@home

My colleague has cleaned database and I think that is the end of http errors etc etc.
I have submitted new work and I am always getting results anyway. There is still a whole
bag of worms around sse2 sse3 ssse3 pni and whatever, not helped by Intel's ifort
refusal to run optimised code on non-Intel hardware.

Igor has much improved version distribution and some people are getting "PNI"
versions. The important thing is that SSE2 upwards is much faster than the generic
version. Don't want to waste resources. All versions are completely numerically
portable (I hope so) but when panic is over I shall be looking at all rejected results
as I believe they are due to hardware failures (over-clocked?).

If all goes well I shall try and issue an update to whatever happened to lhc@home
this weekend.

In the meantime someone has changed the WWW pages, or whatever and I don't even know if you
can read this. All my bookmarks failed and usual start page NOT available.

Eric (from his new super MAC notebook pro, bought at great personal expense,
but have never had the time to set up. I am going to try and install BOINC now.)

---

I think I am back to normal now. Happily crunching 4 tasks at the moment. It's just been a heck of a week for everybody!

I never had any doubt that the problems would get resolved, and I don't worry about loss of credit because it's all for science anyway. And such great science it is. I'm with this project all the way.

Still not getting new Rasks.

I suspended the other Projects I am running and did an Update (twice). Received no Tasks in spite of the fact that Server Status shows 5,000+ available.

Here's the log showing where I suspended the other Projects and then "pumped" LHC ...

7/11/2012 11:27:39 | Einstein@Home | project suspended by user
7/11/2012 11:27:39 | Test4Theory@Home | project suspended by user
7/11/2012 11:27:39 | SETI@home | project suspended by user
7/11/2012 11:27:41 | LHC@home 1.0 | update requested by user
7/11/2012 11:27:43 | LHC@home 1.0 | Sending scheduler request: Requested by user.
7/11/2012 11:27:43 | LHC@home 1.0 | Requesting new tasks for CPU
7/11/2012 11:27:47 | LHC@home 1.0 | Scheduler request completed: got 0 new tasks
7/11/2012 11:27:47 | LHC@home 1.0 | No tasks sent
7/11/2012 11:28:04 | LHC@home 1.0 | update requested by user
7/11/2012 11:28:07 | LHC@home 1.0 | Sending scheduler request: Requested by user.
7/11/2012 11:28:07 | LHC@home 1.0 | Requesting new tasks for CPU
7/11/2012 11:28:09 | LHC@home 1.0 | Scheduler request completed: got 0 new tasks
7/11/2012 11:28:09 | LHC@home 1.0 | No tasks sent

Same thing here, many tasks available but boinc says none available:

wo 11 jul 20:42:18 2012 | LHC@home 1.0 | update requested by user
wo 11 jul 20:42:21 2012 | LHC@home 1.0 | Sending scheduler request: Requested by user.
wo 11 jul 20:42:21 2012 | LHC@home 1.0 | Requesting new tasks for CPU
wo 11 jul 20:42:27 2012 | LHC@home 1.0 | Scheduler request completed: got 0 new tasks
wo 11 jul 20:42:27 2012 | LHC@home 1.0 | No tasks sent

Running on one of the brand new shiny MacBook Pro's (Retina) as well, i see there are no SSE2/SSE3 applications available for Mac OSX?

Last time around this was the disk was full so, I'm assuming the same again?

NO; actually. We now have the correct and planned amount
of disk space. There are scheduling issues out with my
comprehension. The MAC executable uses SSE always. (There are
other issues with it though.....). Eric.

---

Sorry Eric, thanks for all the effort that your putting into this!

SNAFU on our side. Sending jobs with bad disk quota.
Fixed, but you will probably see some jobs fail rather
quickly. At least not wasting so much CPU. Eric.

---

2 Eric Mcintosh

What "PNI" version means? Generec (non-optimized, standart x87 FPU)?
If so - then what is the difference from the normal version (without specifying a modification) and "PNI" version?

I get a wild mix of versions - simple, "PNI" and "SSE2".
http://lhcathomeclassic.cern.ch/sixtrack/results.php?userid=193270SixTrack v444.01
SixTrack v444.01 (pni)
SixTrack v444.01 (sse2)

In general all except correct for my CPU - SSE3 :)
The problem with the work of SSE3 code on AMD CPU has not been solved yet?
(My main CPU AMD Phenom II X6)
If so sheduler must send SSE2 - it work and validate OK. Why it mix versions?

P.S.
Speed comparation of versions (actualy not speed but Cr/hour ratio, but i read before that credits in project are awarded now in proportion to the volume of real work done, not the CPU time * CPU benchmark)

standart 22.7 Cr/h (hour = CPU time, not runtime)
PNI - 34.08 (average from last 9 tasks, 126k sec total)
SSE2 - 37.71 (average from last 6 tasks, 42k sec total)

PNI stands for Prescott New Instructions. It's just another name of SSE3 instructions.

I am getting both sse2 tasks and pni tasks on my Opteron 1210. I am unable to say which run faster because the tasks may have different runtimes. All execute with no error and that is the important thing.
Tullio

Assuming they work correctly (and I have tested all versions
for numerical compatibility) SSE2/SSE3/PNI versions ate typically
at least two times faster than the generic version which will run
on any PC. Eric.

---

Assuming they work correctly (and I have tested all versions
for numerical compatibility) SSE2/SSE3/PNI versions ate typically
at least two times faster than the generic version which will run
on any PC. Eric.

Now that we have the sse2/sse3/pni versions working quite impressively, any news about the development of an CUDA App?

Thanks. Filipe.

---

I am still getting a few PNI and SSE2 at times. Not knowing the meaning of PNI (=SSE3) I aborted those alongside the SSE2. What about having an option to select apps in project preferences?

PS: Congrats on your likely sensational finding at LHC! \o/ Never hoped we would see this so soon.

---

I can't see this happening, the effort to demonstate mathmatical equilvence would exceed the benifit of using CUDA.

I can't see this happening, the effort to demonstate mathmatical equilvence would exceed the benifit of using CUDA.

There was a post from Eric talking about a GPU app (Before the cleaning of the news forum. I can't find it know).

---

Dear All; bit overwhelmed by our progress. Trying to keep up and
keep the work flowing. Must take some time to analyse the results
(very) soon. Even more important I have to finish the "user" doc so
my colleagues can take advantage of this powerful tool.

Plan to produce new executables with the new elements for the LHC, sse4
support too. Need to fix the MAC executable as well. GPUs
are on the back burner at the moment; first SixTrack is
Fortran and also I would need to track many more particles in each job.
That might imply some serious re-design....Still I have used openMP and
perhaps I could use openCL...all a bit vague. Personally I MUST take the
time to publish my work on numeric portability.

We are also planning a major improvement on the (CERN) user end to facilitate
work submission and result retrieval. Eric.
P.S. Using GPU's shouldl NOT be a problem for "mathematical equivalence".
The parallelism is over particles which are independent during a turn. So we
perform exactly the same computations, synchronise at each turn and so on.
This would be fun and exciting but I can't do everything ! :-)

---

Thanks for the news Eric!


Filipe

---