Category Archives: HPC

Archive, HPC

Distributed computing – teraflop/s statistics august 2007

What have I done…
…besides being busy again at work…

Meanwhile I’ve set up an almost automatic teraflop/s data-collector for distributed projects computing power.

The old dinosaur of distributed computing distributed.net has recently been available via Berkely Open Infrastructure for Network Computing (BOINC) in the yoyo@home project.
Yoyo@home wraps the real last distributed.net project the OGR-25 challenge to BOINC clients, since RSA has cancelled the pricemoney for the RC-72 challenge (has the RSA algorithm been fallen ?).
Worldcommunity-Grid, also counts into BOINC for some longer time, with their projects to find cure for: Aids, Malaria, Dengue, Muscular Dystrophy, Alzheimer, Parkinson, …

And finally a not yet integrated into BOINC project, but even bigger in calculation power: Folding@home (FAH)
FAH has once almost reached 1 petaflop/s as the first non-commercial distributed project in may this year, when GPU’s and the PS3 were added as folding clients.
But then came the hot/wet summer, earthquakes and power-outages, …
At least the summer is almost over, and as you can see in the graph FAH is getting faster again, ready to break the petaflop mark.

Using Mac OS X grapher tool, it produces some nice graphics like this one:
?
On the other side, this years supercomputing conference predicted the first commercial installation to reach more than 1 petaflop in october this year.

So, fire up your engines… let the race begin…
😉 Kobaan.

Archive, HPC

Folding at Home at PS3 once again – UPDATE

Sony Playstation 3
firmware update 1.6 is just out of the door, featuring then
Cure@PS3 folding
at home client, and just 20 hours later 3000 PS3 contributors
already reached the mark of 100 TFlop/s processing power.
Unbelievable, as this is the average equivalent of 120000 computers
folding under Windows OS.
Unbelievable further, that PS3 is not even released yet in Europe,
South Africa and Australia, which will happen in just a couple of
hours.
Probably the first people will do tomorrow is to upgrade their
firmware, which means they’ll all get the folding client installed
on their hardware, and if they’re curious enough might become
contributers as well. So might do the people which had to work
during the week, and just waited for the coming weekend.
The expectations are that the 1 PetaFlop/s mark, will be reached for the first time in
the world plus on a non-commercial and non-government project, by
public distributed computing, at about 24000 participants which
could be reached within the next 5 days.UPDATE:
Well we just missed the petaflop in the first approach on sunday
25th, instead after reaching 990 teraflop/s Stanford’s network
collapsed, sort of, although they blame the people playing PS3
instead of folding Winking

Stay tuned for the petaflop announcement which will definitely
happen within the next few week, I guess.

Archive, HPC

Cure@PS3 – Folding at Home on your PS3 finally available

Finally, finally, finally…
…just in time for the european release of Sony’s Playstation 3
on friday, march 23rd, Sony presents the final “now public” client
software for contributing 100 GFlop/s of CELL
computing power to stanford university’s Folding@Home
project.PS3-shot-00006
shot-00007
Let’s all help to fight deceases and boost scientific calculations
for the good of all mankind.
(A note for the green enthusiast: YES, it will consume 180 watts,
but it’s 20 times faster than the average PC, now think
again.)
Kobaan.

Archive, HPC

SUPERCOMPUTING 2006 Tutorial on GPGPU on november 12th

Right in time to this years
supercomputing conference 2006 GPGPU will release lots of tutorials on how to use
graphics cards GPU’s for computation in supercomputer scale.
Most interesting will be the direct use of NVIDIA’s CUDA respective
ATI’s CTM, both a C-like programming language, which should make it
easy for he average joe to use the mighty processing power of GPU’s
for the homeuser.You may also checkout and/or join Stanfords latest development in
F@H
protein folding, as they now support multiple GPU’s (Crossfire
or SLI) in their latest client.
saver2

See further details on supercomputing for everyone on http://gpgpu.org or for the more business
type here: http://sc06.supercomputing.org.

Archive, HPC

SUPERCOMPUTING 2006 Tutorial on GPGPU on 12th november

Right in time to this years
supercomputing conference 2006 GPGPU will release lots of tutorials on how to use
graphics cards GPU’s for computation in supercomputer scale.
Most interesting will be the direct use of NVIDIA’s CUDA respective
ATI’s CTM, both a C-like programming language, which should make it
easy for he average joe to use the mighty processing power of GPU’s
for the homeuser.

You may also checkout and/or join Stanfords latest development in
F@H
protein folding, as they now support multiple GPU’s (Crossfire
or SLI) in their latest client.
saver2

See further details on supercomputing for everyone on http://gpgpu.org or for the more business
type here: http://sc06.supercomputing.org.

Archive, HPC

Open House Day – Science CampusGarching 15.10.2006

Inform yourself about the latest
progress on different science projects at the Science Campus Garching on sunday, october 15th 2006.
At this event you have the chance to talk to scientist about their
work, get things explained and make your own suggestions or simply
ask them questions you wanna know.
Science Campus Garching consists of several Max Planck Institutes
(MPI), the European
Southern Observatory (ESO), Ludwig Maximillians University
Munich (LMU), Leibnitz Rechenzentrum (LRZ) and more.
Further the subway construction to Science Campus Garching is
finished and will open the day before.
LRZ will present one of Germany’s top supercomputer SGI Altix 4700 with 27 TFlops (doubled next
year).
garching_kl
You can take a look at the latest fusion reactor experiments, reach
for the stars at ESO‘s
remote controlled Very Large Telescope(VLT) in Chile, or or or…

Archive, HPC

Open House Day – Science Campus Garching 15.10.2006

Inform yourself about the latest
progress on different science projects at the Science Campus Garching on sunday, october 15th 2006.
At this event you have the chance to talk to scientist about their
work, get things explained and make your own suggestions or simply
ask them questions you wanna know.
Science Campus Garching consists of several Max Planck Institutes
(MPI), the European
Southern Observatory (ESO), Ludwig Maximillians University
Munich (LMU), Leibnitz Rechenzentrum (LRZ) and more.
Further the subway construction to Science Campus Garching is
finished and will open the day before.
LRZ will present one of Germany’s top supercomputer SGI Altix 4700 with 27 TFlops (doubled next
year).
garching_kl
You can take a look at the latest fusion reactor experiments, reach
for the stars at ESO‘s
remote controlled Very Large Telescope(VLT) in Chile, or or or…
Archive, HPC

GPU Power outperforming CPU Power

According to the latest
F@H statistics, graphics cards processors are pretty much
outperforming traditional CPU’s.
The current factor is about 75 GFlops on GPU against an average of
1GFlops on a CPU. (this varies because of very old processors and
modern processors with about 14 GFlops).

Anyway at the moment we see 370 GPUs doing the same amount of work
as 29000 CPUs. This is really great for the F@H project, although the GPU
cores at the moment are only capable of doing basic calculations
with the gromacs application. Other applications have to be heavily
rewritten, and/or are not able to run every calculation on GPUs and
therefore have to use the main CPU for those parts of the
calculation.

Archive, HPC

GPU Power outperforming CPU Power

According to the latest
F@H statistics, graphics cards processors are pretty much
outperforming traditional CPU’s.
The current factor is about 75 GFlops on GPU against an average of
1GFlops on a CPU. (this varies because of very old processors and
modern processors with about 14 GFlops).

Anyway at the moment we see 370 GPUs doing the same amount of work
as 29000 CPUs. This is really great for the F@H project, although the GPU
cores at the moment are only capable of doing basic calculations
with the gromacs application. Other applications have to be heavily
rewritten, and/or are not able to run every calculation on GPUs and
therefore have to use the main CPU for those parts of the
calculation.

Archive, HPC

Science Sports – Calculate PI

And another game for the next birthday
party:dart-pi
Let your guests calculate Pi by playing some rounds of
“darts”.

Do this by targeting the blue shaped square on the above figure.
And then simply count the darts you are throwing and that will be
inside the square, and additionally the darts that miss the board
and hit inside the marked blue area.

And after a few rounds, let’s say 1.000-100.000 Winking should be enough
to see some good results.
Now use this formula:

dart-pi-formula

A small hint for the frustrated, invite people that are able to aim
the upper left square, but are not so much pros that they will
always hit the board only.