IBM's "Blue Gene" supercomputer will lose a core every few days - to cosmic rays?!?

[CTho9305]

According to Wired 9.07 (July 2001) in their Blue Gene article, on page 156, last full paragraph:


<< The computer is intended to be highly fault-tolerant. When processors break down, as they will - cosmic rays should knock out a core every few days - the machine is meant to isolate them and route the data around the blockage. >>



If cosmic rays kill CPUs, then wouldn't there be a lot more DOA cpus in the retail market? Even if its b/c these are .13u chips, then we will have lots of DOA P4's soon, and geforce3s (.15u I think)

 

[TunaBoo]

Thier CPU's are probably a lot more sensitive. Your CPU prolly gets hit an average of once per 6 months with a cosmic ray. Cosmic rays are the very reason thy have ECC ram (A ray can flip a bit or 2 in the ram).

 

[MrHelpful]

Yeah, but Blue Genes have many many processors... More than hundreds of thousands, I believe. Chances of losing processors increase greatly.

 

[McCarthy]

When they say &quot;break down&quot; do they mean it actually damaged the CPU and rendered it useless or caused a fault/lockup in that CPU which would work again if reset?

Even if it has hundreds, thousands, of CPUs...well, we around here have that many and most of the dead ones can be explained pretty easily.

--Mc

 

[CTho9305]

It has 25,000 processors. no more, no less (well, less when some die of course)

 

[Doomguy]

What exactly is a cosmic ray?

 

[SCSIRAID]

Alpha particle hits from cosmic rays generate errors but do not damage the processors. Memory and register arrays are the most vulnerable but logic can take hits too.

 

[zephyrprime]

Cosmic rays are naked atomic nuclei originating from outer space. They are are very high energy owing to their high velocity. They are the most energetic rays known to man - exceeding the energy of particles generated by earthly particle accelerators. Their origins are not really known.

 

[Soccerman]

the un-forgiving universe strikes again!

you think that's bad? the atmosphere filters out most of the cosmic rays (or is it the magnetic field of the earth? yeah that sounds right), but when you're up in space, your radiation dosage increases by alot. CPUs up there are specially hardened against the radiation. I wouldn't be surprised if that wasn't what was causing some of the problems on the ISS/Alpha..

 

[zephyrprime]

http://www.sciam.com/0197issue/0197swordybox1.html

 

[TunaBoo]

<<

the un-forgiving universe strikes again!

you think that's bad? the atmosphere filters out most of the cosmic rays (or is it the magnetic field of the earth? yeah that sounds right), but when you're up in space, your radiation dosage increases by alot. CPUs up there are specially hardened against the radiation. I wouldn't be surprised if that wasn't what was causing some of the problems on the ISS/Alpha.. >>



Actually, like I year ago I read on /. that all spaceships proc's are radiation hardened 486 66's that cost about 10k EACH. They are supposed to be more resistant to rays and stuff I guess.

 

[whitelight]

interesting.. never knew about cosmic rays.

 

[McCarthy]

Guess I'm still not sure, though it sounds like they're dead, done, broke. In which case I have to ask -

Any of you personally, or know someone firsthand (not friend of a friend's brother in law's goat's mother) who has had a perfectly good CPU, running within temp specs, not hit by lightening, just up and die?

I realize the chances are small, but that said every few days with 25,000. If the same rate happens with our CPUs you'd think it'd be known?

Sounds more interesting than &quot;Doh, I cracked the core putting on my heatsink&quot; anyway

--Mc

 

[Soccerman]

well I know Stephan had a Celeron 300@450 that wouldn't overclock anymore.. I don't know if Cosmic rays had anything to do with it.