PDA

View Full Version : Sandia Red Storm upgrades for the future


Yousuf Khan
July 30th 04, 01:56 AM
They are talking about replacing the slower existing Opterons with faster
dual-core Opterons to take it upto 100 Tflops!

http://news.com.com/Sandia+supercomputer+to+get+dual-core+Opterons/2100-7337_3-5289251.html

Yousuf Khan

--
Humans: contact me at ykhan at rogers dot com
Spambots: just reply to this email address ;-)

Klaus Fehrle
July 31st 04, 12:41 AM

"Yousuf Khan" > schrieb im Newsbeitrag
t.cable.rogers.com...
> They are talking about replacing the slower existing Opterons with faster
> dual-core Opterons to take it upto 100 Tflops!

>
> http://news?.com......

Hmm. http://www.theinquirer.net/?article=16819
>
KF

Robert Myers
July 31st 04, 02:28 PM
Klaus Fehrle wrote:
>
> "Yousuf Khan" > schrieb im Newsbeitrag
> t.cable.rogers.com...
>
>>They are talking about replacing the slower existing Opterons with faster
>>dual-core Opterons to take it upto 100 Tflops!
>
>
>>http://news?.com......
>
>
> Hmm. http://www.theinquirer.net/?article=16819
>

The Sandia/Cray PR team certainly seems to be working overtime

"Sandia supercomputer to be world’s fastest, yet smaller and less
expensive than any competitor"

http://www.sandia.gov/news-center/news-releases/2004/comp-soft-math/redstormrising.html

(note the title of the html page; compare to the title of Intel/HP's
book about Itanium). The Register seems more likely to have got it right

http://www.theregister.co.uk/2004/07/30/cray_red_drizzle/

"Cray pours Red Drizzle over anxious investors"

Cray is laying off:

http://www.theregister.co.uk/2004/07/28/cray_q2_flop/

"Cray's Q2 revenue gigaflops"

On the surface, the story at Cray looks like déjà-vu all over again

"Meanwhile, for all its vaunted muscle, the Cray X1 posted negligible
sales. The number-crunching monster is looking every inch a dinosaur,
burdened by a hefty price, limited market niche and pricing pressure
from nimbler competitors."

"The company is heavily reliant on the U.S. government, which accounted
for 80% of last year's sales. It blamed the quarterly shortfall on a
drop in a defense appropriation and a delayed a major contract. Worse,
X1's successor has run into production snags, forcing Cray to cut its
sales outlook."

Not everyone thinks all the money pouring out of the free money machine
is being wisely spent:

http://www.computerworld.com/hardwaretopics/hardware/story/0,10801,94607,00.html

and all the (prospective) good! news! about Red! Storm! isn't
necessarily the good news Sandia and Cray would like to make it seem to
be. In the end, Red Storm is just another commodity cluster (albeit
with a fancy router chip), especially with fermilab reporting 4.5
microsecond MPI latency with infiniband and pci-xpress:

http://lqcd.fnal.gov/benchmarks/newib/

RM

Klaus Fehrle
August 1st 04, 12:50 PM
--

"Robert Myers" > schrieb im Newsbeitrag
news:[email protected]_s03...
> Klaus Fehrle wrote:
> >
> > "Yousuf Khan" > schrieb im Newsbeitrag
> > t.cable.rogers.com...
> >
> >>They are talking about replacing the slower existing Opterons with
faster
> >>dual-core Opterons to take it upto 100 Tflops!
> >
> >
> >>http://news?.com......
> >
> >
> > Hmm. http://www.theinquirer.net/?article=16819
> >
>
> The Sandia/Cray PR team certainly seems to be working overtime
>
> "Sandia supercomputer to be world’s fastest, yet smaller and less
> expensive than any competitor"
>
>
http://www.sandia.gov/news-center/news-releases/2004/comp-soft-math/redstormrising.html
>
> (note the title of the html page; compare to the title of Intel/HP's
> book about Itanium). The Register seems more likely to have got it right
>
> http://www.theregister.co.uk/2004/07/30/cray_red_drizzle/
>
> "Cray pours Red Drizzle over anxious investors"
>
> Cray is laying off:
>
> http://www.theregister.co.uk/2004/07/28/cray_q2_flop/
>
> "Cray's Q2 revenue gigaflops"
>
> On the surface, the story at Cray looks like déjà-vu all over again
>
> "Meanwhile, for all its vaunted muscle, the Cray X1 posted negligible
> sales. The number-crunching monster is looking every inch a dinosaur,
> burdened by a hefty price, limited market niche and pricing pressure
> from nimbler competitors."
>
> "The company is heavily reliant on the U.S. government, which accounted
> for 80% of last year's sales. It blamed the quarterly shortfall on a
> drop in a defense appropriation and a delayed a major contract. Worse,
> X1's successor has run into production snags, forcing Cray to cut its
> sales outlook."
>
> Not everyone thinks all the money pouring out of the free money machine
> is being wisely spent:
>
>
http://www.computerworld.com/hardwaretopics/hardware/story/0,10801,94607,00.html
>
> and all the (prospective) good! news! about Red! Storm! isn't
> necessarily the good news Sandia and Cray would like to make it seem to
> be. In the end, Red Storm is just another commodity cluster (albeit
> with a fancy router chip), especially with fermilab reporting 4.5
> microsecond MPI latency with infiniband and pci-xpress:

I basically agree. What I took home from Heidelberg in a nutshell is
Supercomping-Architecture really comes down to economical approaches
nowadays - making economies of scale accessible, to put it more precise,
this implying to widen the focus of designs beyond just deploying technology
for scientific use. From this point of view, the golden standard question in
the Hot Chair sessions "What commercial use for your concept do you see?" is
the crucial point of distinction. Red-Storm architecture looks charming for
its use of very balanced CPUs in terms of FP/integer capabilities in
comparison with other approaches at first glance. However, it's beyond my
scope to overlook how the combination of Linux and Catamount Os' works out
for running commercial stuff on it and how the mesh will deal with such
application, so I would appreciate any comments in this respect.

KF