LHC@home

will be nice i think ;-)

Common Criteria (ISO 15408)

Nice. How much does such a 'puter cost?
I wish one for xmas... :-p

if the project have more WU s to crunch
MB run it 2-3 days and we are Team No -1- for long time ;-)))
will be nice and a feeling like a day at the beach with 2 nice girls like PAM hohoho ;-)))
feel free to visit www.guidowaldenmeier.de

Why dont you run CPDN on it? I am sure folks there would be happy with some units actually finished. And we would now how long would it take for Cray to do one CPDN unit. Go for it! Do at least one run! ;-)

The problem with something like the cray xd1, is that is uses Opteron processors. It just has alot of them. It will therefore not run a single
instance of sixtrack or CPDN much faster than a server with a normal Opteron. It might be slightly faster because of better memory and disk systems.
The advantage is that it has a lot of processors, so it can crunch many WU at the same time. So it will still take 500 hours to return a unit, but after those 500 hours, the cray will return maybe 500 WU and therefore have an average WU time of 1 hour.


Christian Søttrup
LHC@home Developer

> The problem with something like the cray xd1, is that is uses Opteron
> processors. It just has alot of them. It will therefore not run a single
> instance of sixtrack or CPDN much faster than a server with a normal Opteron.
> It might be slightly faster because of better memory and disk systems.
> The advantage is that it has a lot of processors, so it can crunch many WU at
> the same time. So it will still take 500 hours to return a unit, but after
> those 500 hours, the cray will return maybe 500 WU and therefore have an
> average WU time of 1 hour.
>
>
> Christian Søttrup
> LHC@home Developer
>

This is what i Would expect from BOINC (that is running one WU on one prcessor). That's wy it made me curious, when MB has written that Cray can crunch WU in 2.5secs. Maybe it has some system of 'merging' CPUs so they look like one for an application? I just can't believe that Opterons could be so efficient. I know these are state of the art CPUs, but I wouldn't expect such a difference.

No, I don't think they can either. If they actually finish a WU in 2,5 sec.
I am pretty sure it is because it crashes or calculates wrong. Since we have not released a version of sixtrack compiled directly for the opteron it will have to run it in 32 bit compatabillity mode, which should not be particularly fast. We had some nice tricks in sixtrack to do vectorisations of some math on alpha processors, but we had to take it out because it gave different results on Pentium and AMD.

Cray is also known for making very fast processors, but you have to know what you are doing when you are coding for them. In one of Cray's machine you had to write a comparison as:
if (1.0 * a == b) then ..
If you did not write the "1.0 *" it would sometimes give the wrong answer. they did this because then they could save an AND gate(basically a transistor)on the critical path, making the processor faster.

I do not know if cray's still need tricks like this, but if they do i guess that could cause sixtrack to finish very early.

Christian Søttrup
LHC@home Developer

> I do not know if cray's still need tricks like this, but if they do i guess
> that could cause sixtrack to finish very early.

I was my understanding that most of the Crays only provided an approximation for division operations, and not a very accurate one at that ...
<p>
For BOINC Documentation: Click Me!

That seem sensible to me. I read that Cray's OS is not Windows (that didn't surprise me) nor linux (that was not a big surprise either) but something called Super Unix. A surprise to me is that sixtrack could run on it without any alterations in code.
MB, did you get any output files from your Cray experiment?

> so, does guido talk bulls*it and in addition owns an XD1 he wants to donate ?
>
>

Cray's are dirt cheap these days.
http://cgi.ebay.com/ws/eBayISAPI.dll?ViewItem&category=1484&item=5724870440&rd=1

And your personal attack on Guido is uncalled for, especially in this thread which was on topic of crunching.

Ever consider that maybe he wasn't trying to be annoying (in your view) on purpose?
Maybe it's a language thing. Maybe he's enthusiastic about crunching for BOINC. Maybe that person is 'just like that' and there's no changing their personality.

I know that when I visited Japan, none of the Kanji made sense, so to survive, (to eat, to use the subway, find the washroom, to go shopping, etc.), I had to rely on pictures.
So, when I see someone else using pictures to communicate, I don't have as much a problem with that.

> So, when I see someone else using pictures to communicate, I don't have as
> much a problem with that.

The main problem is his leeching. Most of the pics he posts live not on his webspace but somebody elses. Everytime the thread is opened that site takes hits so reducing its bandwidth and in some cases actuall costing them real money. Its the likes of Guido who cause major problems without any thought for others.

Guido if you post pics make sure you own them and they are on your webspace!

Angs

thread closed by GOV