Re: AMD dual core vs Intel core 2 quad

2008-02-07 Thread Daniel Tryba
On Wed, Feb 06, 2008 at 04:44:41PM +, Jo Shields wrote:
 To be specific: when I run molecular dynamics (MD) on Debian amd64
 with my shared memory Thunder with 2 AMD dual core, parallelization
 support OpenMPI (everything compiled with Intel ifort/icc), the MD
 program sees 4 cpus and top -i indicates four cpu at work. The
 gain with respect to running MD in serial mode is about threefold,
 i.e. the parallelization is not bad at all. What can I expect - from
 this viewpoint - with four Intel Core 2 quad?

 - but the achilles heel of the Intel platform is memory contention,
 and for heavily memory-bound codes, scalability will be poor.
 Conversely, for CPU-bound codes, performance will be SIGNIFICANTLY
 better. Roughly speaking, using a 4-way GROMACs job as an example, the
 same job could complete about 25% faster, clock for clock, on Intel.

A bit offtopic, but maybe Francesco would benefit from a PS3 if (big IF)
you can fit your computations in 256MB and can do a decent job of
programming the SPUs. Or maybe even faster (and more memory) a GPU.

-- 

   Daniel Tryba


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



Re: DG965ss

2008-02-07 Thread Lennart Sorensen
On Thu, Feb 07, 2008 at 05:07:22PM +0100, Dr. Helmut G. Enders wrote:
 I still have one strange error.
 When I do a apt-get install large deb file,
 e.g. linux-image-2.6..., this file
 is corrupted (apt-get said gzip: invalid
 compressed data). If I fetch this deb-file
 with wget into /var/cache/apt/archives
 and do an apt-get install package
 it worked.  I tried within my net some
 filetransfers without a problem. So the
 interface seems to be ok.
 
 And apt-get from my other maschine with
 the same net/firewall to the same mirror
 worked smoothly.

Do you use a http proxy?  I have had my proxy get out of sync and end up
caching a bad Packages file, so every time it downloaded it the proxy
gave me the wrong one and I got messages that the checksum on the
downloaded package was wrong when in fact the package was right but
Packages.gz was corrupt.  Forcing the proxy to refresh the file solved
the problem (as did turning of the use of the proxy temporarily).

--
Len Sorensen


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



Re: DG965ss

2008-02-07 Thread Justin Piszcz



On Thu, 7 Feb 2008, Dr. Helmut G. Enders wrote:


I tried an amd64 installation on an Intel DG965SS board with 8GByte.
The system is very very slow (installation more than 1 day.) and
I got corrupt files from the net (e.g. apt-get update).


Justin Piszcz:


Use the following boot option:
append=mem=8832M


Yes it works, thanks.


I wouldn't touch the BIOS
Justin.


You are right, I first changed back in the bios.
Tt worked with the mem problem, but I got a lot
of new troubles. So I ruefully come back to your
advice and updated the bios to the new
version again and with your option it worked
nearly perfect.  Even my grub problem with the
raid controller (3ware) disappeared.



I still have one strange error.
When I do a apt-get install large deb file,
e.g. linux-image-2.6..., this file
is corrupted (apt-get said gzip: invalid
compressed data). If I fetch this deb-file
with wget into /var/cache/apt/archives
and do an apt-get install package
it worked.  I tried within my net some
filetransfers without a problem. So the
interface seems to be ok.
Strange, do you use an http proxy by any chance?  Also you could try a 
different mirror just incase there was some temporal weirdness.  I assume 
you ran the latest version of memtest86 and let it cook for some time?



And apt-get from my other maschine with
the same net/firewall to the same mirror
worked smoothly.

Helmut



--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]




--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



DG965SS

2008-02-07 Thread Dr. Helmut G. Enders


 Justin Piszcz:
...  I assume you ran the latest version of memtest86 and
  let it cook for some time?

Grub said  28: does not fit into memory?  and I have *no*
floppy :-(.

I'll try to burn the image onto a cd.

Helmut


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



Re: DG965SS

2008-02-07 Thread Justin Piszcz



On Thu, 7 Feb 2008, Dr. Helmut G. Enders wrote:




Justin Piszcz:

...  I assume you ran the latest version of memtest86 and
 let it cook for some time?

Grub said  28: does not fit into memory?  and I have *no*
floppy :-(.

I'll try to burn the image onto a cd.

Helmut


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



I am curious of the results :)

If its good ram, probably http proxy or other misc error, else you got bad 
ram :)


Justin.


--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



Re: AMD dual core vs Intel core 2 quad

2008-02-07 Thread Lennart Sorensen
On Wed, Feb 06, 2008 at 11:59:14PM -0800, Francesco Pietra wrote:
 Together with other suggestions that were kindly provided, that above makes an
 alluring prospect for me. Thanks. My point of view is that when a 8xxx series
 is offered, they already have a 8 series ready for offer. I rely much on
 the quality of the software, too, in particular the algorithm for
 parallelization and the model, i.e., for MD the force field. Bad ones fight
 against any advancement in hardware. I would be much indebted for a
 clarification of details, at the reader convenience.
 
 8xx series (without their box/fan) are found for little. Should the four
 8xx be the same clock? (as they are found as single, or at most as pairs).
 Also, is the box/fan specific of the 8xx series or can I use the box/fan for
 the 2xx I have? (and only look for two additional boxes/fans).

I believe that they have to be the same speed, and in general it is
recommended that they be the same revision (stepping) as well.  If they
have the same code on the box, then they should be good to work
together.

 In the European union - where I am - a four-way mainboard for 8xx and 184
 pin ECC is not easy to find. Worldwide, I came across brand new Supermicro
 H8QCE-B Quad Opteron 800 Dual Server motherboard and Supermicro CK804 Quad
 Opteron 800 Dual Server motherboard at an affordable price (though as much as
 $200 are charged for shipping).

The problem is that the old DDR based systems are obsolete and the DDR2
based ones are the ones currently made.  No one wants to stock obsolete
equipment making it hard to find.

 As to the mainboard, which specifications should I check in order that
 there is no too narrow bottleneck for the 8xx cpus and ECC 400MHz? Optionally,
 it might help to have a connection to make a cluster with an 8-way server,
 should I decide one day to transfer my private server to my institution.

The standard interconnects on the opteron ought to be plenty to handle
the memory bandwidth.

 Can different memory slots (8 2GB ECC and 8 1GB ECC, all 400MHz and same
 voltage) be assembled into the 16 slots of the mainboard?

Yes.  Preferably you should balance the ram between the CPUs if
possible, but most important is to make the dual channel work so each
bank has to be matched.  If you have 24GB, you could put 6GB per CPU
with 3GB per bank on each CPU.

of course ram is probably (not sure about registered ecc ram) cheap
enough that it really is a rather insignificant part of the cost of a
full system.  It may make more sense to go for the new style 8xxx
instead with new ram.  DDR2 has more bandwidth than your current ram, so
if your application uses a lot of ram that could make a big difference.

After all if you want 8 cores, that could be done with either 2 opteron
2300 series, or 4 8000 or 8xx series, or 2 8300 series.  4 8300 series
would of course give you 16 cores on a 4 socket board.

Looking at prices I see a quad core opteron 2350 is around $400, while
an 8350 is around $1200.  A dual core opteron 870 costs about $600 (so
half the cores of a 2350 at the same clock speed for 50% more money).
So to get 8 cores you could get a dual socket board with two 2350's for
$800 for CPUs, while for an 870 you would need a 4 socket board and
$2400 worth of CPUs.  How much ram could you buy for the difference?  A
two socket board is likely to be a lot less than a four socket board as
well, and being the new 2xxx series with DDR2 it is likely to be much
easier to find.  DDR2 ram also costs less than DDR ram these days.  You
can get a dual socket system that holds up to 32GB DDR2 ram with 2xxx
series CPUs for about $400, like for example this one (no idea how good
it is):
http://www.directcanada.com/products/?sku=10530BD5632vpn=KFN4-DREmanufacture=ASUS

Basicly it can be VERY expensive to try to reuse old parts.

--
Len Sorensen


-- 
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]



DG965ss

2008-02-07 Thread Dr. Helmut G. Enders

 I tried an amd64 installation on an Intel DG965SS board with 8GByte.
 The system is very very slow (installation more than 1 day.) and
 I got corrupt files from the net (e.g. apt-get update).

Justin Piszcz:

Use the following boot option:
append=mem=8832M

Yes it works, thanks.

I wouldn't touch the BIOS
Justin.

You are right, I first changed back in the bios.
Tt worked with the mem problem, but I got a lot
of new troubles. So I ruefully come back to your
advice and updated the bios to the new
version again and with your option it worked
nearly perfect.  Even my grub problem with the
raid controller (3ware) disappeared.



I still have one strange error.
When I do a apt-get install large deb file,
e.g. linux-image-2.6..., this file
is corrupted (apt-get said gzip: invalid
compressed data). If I fetch this deb-file
with wget into /var/cache/apt/archives
and do an apt-get install package
it worked.  I tried within my net some
filetransfers without a problem. So the
interface seems to be ok.

And apt-get from my other maschine with
the same net/firewall to the same mirror
worked smoothly.

Helmut



--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]