"I bet it still won't run Crysis lol"

actually i think ur right, since crysis can only multithread on like 4 cpu's --... not 45,000.... or something

Roadrunner uses Opterons clustered with IBM Cell processors . You have Opteron AM2 Santa Rosa cpus matched with the PowerXCell 8i (64 bit version of the PS3 cpu). In apps like radiation transfer the Opeteron /Cell is 6 times faster than the Opteron alone. So it really depends on what you do which one will be faster. It is much like Red Storm is 30% faster than Blue Gene L in Fast Fourier transforms but 75% slower in Global Linear Algebra apps. Jaguar uses a mix of about 50 tflops of XT3's with Denmark 185's in the original configuration in 2006(ranked 3rd in HPC's Fast Fourier benchmark for 2006 and 2007). http://www.hpcchallenge.org/ Late 2007 they added about 200 tflops of XT4's with Opteron 1222's or 1224's. Now they have Opteron Barcelona 9600's in the XT5 frame. Both the Shanghai and the next generation will fit in the XT5 socket. 2009 should see the Cray/AMD Cascade and the IBM P7 make their appearances at about 4+ Pflops or 3 times faster than Jaguar and Roadrunner. http://www.nitrd.gov/pubs/2006supplement/hec_rd.pdf

Wikipedia: Roadrunner is designed for a peak performance of 1.7 petaflops. If reverse engineer math with this statement.

If 45000 Opetrons in jaguar give 1.64petaflop then 6480 Opterons (in roadrunner will be contribuiting) 0.24 petaflops of roadrunner's peak 1.7 petaflop. also meaning each PowerXCell operates at 7.2 times the performance of each Opteron. though Opteron is documented to operate at PowerXCell connector only.

http://www.hpcwire.com/topic/processors/IBM_Roadrunner_Takes_the_Gold_in_the_Petaflop_Race.html
http://www.nzherald.co.nz/technology/news/article.cfm?c_id=5&objectid=10515465

If this how then
-----------------------------
6948 .44 x 1.7 ........ x=26845
6480 .44 x 1.7 ........ x=25036

jaguar is using 18155 (45000-26845) more opterons than math requires to reach 1.7petaflops (1.64 infact). this also means 2 times drop in opteron scaling.

6948 .44 12960 x
x=0.82 (is roadruner 0.82+0.44=1.26pf peak capable, wikipedia false?)

6948 .44 x 1.26 ........ x=19896 (opteron only config)

so intead of 6480+12960=19440 cocktail-cpus stunt, IBM should have just used 19896 opetrons only to reach 1.26pf i.e just 456 more opterons.

19896 1.26 45000 x ........ x=2.85pf (jaguar pf !! either jaguar is lieing or with 45000 opterons we have 4 times drop in scaling or they have infact a 2+ pf machine)

any ideas !?!

Finally a computer with that can run Vista.

We will see tomorrow but for the first time ever IBM's Blue Gene L may not make the DARPA HPC top 3 in any benchmark. Ranger at TACC passed BGL a couple of months ago. Roadrunner doubled Blue Gene L. 

Sylvie these benchmarks are a whole lot more reliable than what you cited in your story.

The Future News Headline:

"Computer scientists at Cray are stumped on why their newest supercomputer no longer seems to be working. Instead of the giving the correct answer to an array of problems presented to it, the machine simply shows a large '42' on its display."

"In unrelated news, the building hosting the supercomputer has been plagued with an outbreak of infestation from white mice."

Just kidding. Well done!

nVidia Quadro FX-5800 memory bandwidth 102GB a second vs Cray XT 286GB a second they should have stick in couple of nvidia gpu-s or is it gpgpu-s?

If there was just a PCI-e slot....

I bet it still won't run Crysis lol

Can they clip 45,000 AMD "Shanghai"s into this system? Have AMD already sold out a few months of Shanghai production?!