University of Maryland researcher Uzi Vishkin
BOFFINS at the University of Maryland have emerged from their labs with a desktop parallel computing system they say is 100 times faster than current PCs.
According to Network World, the computer uses a circuit board about the size of a license plate upon which the boffins have stuck 64 parallel processors.
As could be expected the electronics was not that difficult, but writing the software which could efficiently use the 64 processors was a bit tricky.
However Vishkin said the algorithms the project used made it possible to use 64 processors efficiently for general-purpose computing tasks for the first time.
How are they handling the heat?
Why is the BS meter always at 7.6?
Man I hate to be the cranky geek some times, but we’ve heard these claims before. Even if they did manage a breakthrough, the scheduling algorithms are going to be patented from here to next Tuesday which means no mainstream OS will adopt them. Moreover, I barely trust developers (myself included) to write multi-threaded apps for single and dual symmetric processing…let alone parallel.
I wonder how much they cooked the code to get those results. Even if this is accurate, the only systems this will be used in are research crunchers & simulation environment generators.
Geez, it’s not that hard from an engineering POV, to emulate a standard AMD / Intel CPU in a large PCI card with lots of real-estate.
Just like a RAID-5 (with six hard disks, one acts as parity, load spread across five disks. Read & Writes are faster)
Between the motherboard & the CPU, where the CPU normally sits, you wire 1-to-1 the pins to a cable to a daughterboard.
The daughterboard acts as one virtual CPU, yet behind it, you have a controller CPU allocating and routing CPU instructions to different CPU’s.
Thus Windows “thinks” there’s only 1 CPU, but in reality, on the daughterboard there are X CPU’s (how many depends on real-estate). Existing non multi-threaded programs just get executed faster.
The controller CPU, being an EEPROM, can be reprogrammed to emulate any number of CPU’s out there.
The 100 CPU’s don’t have to be very fast, they could be underclocked cheap P4’s 1.6 Ghz with little heat / power specs.
I saw a picture on another website, it was plugged into the motherboard as a PCI card (PCI-X?)
The most important thing I see here for this Tech is breathing new life into an old computer, especially for costly 5U servers that have full room inside. Just plug this daughterboard in the PCI-X slot and wire into the existing CPU’s.
Gee, I hope they hurry up and bring this to market. Word is soooo sluggish.
2–made me laugh. I was thinking why not call it the “truthiness meter” and get on Colbert?
But it is indeed a BS meter and probably brought into use when a floor value of 7.6 is initially spotted. Editor can correct us all showing an example of 8-9-10? but that would not be as much fun.
Just like that #5? I dont think thats how it works.
if these guys have indeed masked the complexity of parallel programming and truly have allowed normal apps like simple data entry, etc, to remain simple single-threaded apps, and left the heavy lifting like logistics and heavy data analysis to the parallel guys, then hooyah!
and yeah, Word kinda sucks the life out of any PC it’s on
I’ll believe it when it’s on my desktop (at an affordable price, naturally).
I bet there’s going to be a Mac user saying “My dual core iMac can outperform that machine.”
A $500 prize is being offered for the winning name submitted for this computer. Oh, boy! $500!
http://tinyurl.com/25c995
How about “iDon’tBelieveIt!” or “iThinkDvorakHatesHype”?
…so it turns out you can have nine women make a baby in just one month. Kewl.
64 parallel processors are not going to yeild 100 fold increase. Some fraction of 64 fold should be closer to whats expected.
#2 I want to know why the BS meter is still analog and not digital. Also when was it last calibrated and the certs to back it up.
#2,
Why is the BS meter always at 7.6?
Because 7.5 was taken.
Technically, cramming 64 processors into a box isn’t a big deal. I built a 64 processor desktop machine for my Masters project in ’93. (64 T800 Transputers at 25MHz).
I was able to get a near linear relationship of speed vs. # of processors only on very specific tasks, like finite element analysis, using highly optimized code. I graduated, and other grad students tried to follow on with writing parallelization tools etc. I don’t think anything ever came of it.
Anyway, although my insight is 14 years out-of-date, I smell something fishy with this. 75MHz processors? This development system is 10 years old! This is just an EE (CPE) prof and his grad student who have some very preliminary data. It will make a nice PhD thesis, but this is just likely an attempt by a couple of academic types to attract funding.
Cool — the return of OCCAM, anyone?