SETI@Home -- Running On A PCI Card

levendis writes: "This has got to be the strangest piece of hardware I have ever seen. It's a PCI card with 6 embedded processors and a flash rom containing Linux and the SETI@home client. The manufacturer claims it can process 6 SETI work units in 16 hours, completely independent of the host CPU."

This is a truly intriguing piece of equipment. I especially liked this bit from the FAQ: "SETI accelerator® uses military surplus components. The chip used on this board was used for target vector calculations in the terrain following radar (TFR) component of the PR-964 Cruise Missile (NATO Codename SAMOWAR)." I wonder whether they could release similar cards adaptable to the emerging pay-for-cycles outfits like ProcessTree. If yes, maybe the card could pay for itself after a while.

Comments

This is such a hoax

[toastyman]

• I can't find any mention of a MP-105-D anywhere.
  
• I'm guessing that any "terrain following radar" processor would either be a standard microprocessor (i386, arm, 68k, etc), meaning there's no reason to use this over a standard proc. OR it's an ASIC, meaning it's impossible to make it do anything but process terrain.
  
• The picture of the six processor board is obviously faked. The processor closest to the camera is overlapping a crystal. The skew of the chips repeated isn't perspective corrected properly, either. Also, staggering the upper chips from the lower chips has to make routing impossible on the board.
  
• There are obviously no jumpers or sockets on either board, when the faq mentions them.
  
• If I were going to do something this silly, I'd use a ColdFire or StrongArm processor. In any case, I don't think there are ANY high performance microprocessors that have imbedded ram/flash/etc. The fastest thing out there like that would seem to be a 68HC11 which maxes out at around 4Mhz. The ColdFire *does* have 4k of SRAM internal to it, but definately not 2MB+ of flash, internal ram, cache, etc. Jamming that much stuff on one die is expensive, and you end up with low yields.
  
• The cost to layout the board and get 1000 or so boards in would be very very prohibitive for a product like this.
  

Bleh. :)

Does this affect results?

[Matt2000]

I was just wondering, since the Seti@Home dudes are always talking about how they need to keep the source code under control to keep any semblance of scientific integrity, something which I think makes sense, does the porting of the block cruncher to other systems affect this goal at all?

For example, if there's a rounding error in one of the floating point libraries for whatever OS you're compiling for, let's just take the common example of a 6 processor embedded PCI card, wouldn't that invalidate the results from that computer as much as the Seti@Home source being modified?


Hotnutz.com - Funny

This will really help...

[Fourier]

Now we can all analyze redundant data packets six times faster! :-)

Sometimes I wish there were a distributed computing project out there that I actually cared about. Maybe we should start a project to scan the space of all 2MB Linux ELF executables until we find one that corresponds to a fast, stable, standards-compliant web browser. The funny thing is, that might actually take less time than Mozilla...

Can you say "Hoax"?

[dan the person]

Looks like one to me. The pictures of the boards for instance, the 6 processor version the chips look like they are vut and pasted on top of all the other chips on the circuit board. The FAQ recommends you by the single processor version and then fill the slots up with the required number of processors to your preference. There are no slots! It runs the Linux Client? Since when was linux ported to this embedded processor of the US Militaries, and where is the source? And when was the Seti client ported to this platform. Looks like a hoax to me. At best it's a GPL violation as there is no source for the linux kernel modifications from the port.