SmokingSkillz Posted January 26, 2004 Share Posted January 26, 2004 I bumped into an old mate in a pub, and he works in the field of micro-gravity geophysics. He was telling me about a super computer at his uni (keele), made using 64 standard motherboards and athlons with tons of bizare cables. He claimed that the result of this configuration gives a processing power of each cpu to the power of 64! I said surely it it will just run a program in at most a 64th of the time one processor will. But he seemed sure of his claim. Anyway has anyone a clue as to how the units would be wired together? He said the code that distributes the load is only a few lines long and the parallel nature of the system is transparent to the user, I guess it is hacked into the linux kernel at a very low level. But how on earth this has been done with standard motherboards is a mystery to me! Link to comment Share on other sites More sharing options...
JaKiri Posted January 26, 2004 Share Posted January 26, 2004 Raised to the power of 64? He's talking bollocks. If it was made with 2GHz processors, according to him it would have an overall processor speed equivilent of 1.8x10^778Ghz. Which is more possible states than exist in the universe. ps. It will be, in practice, either a little slower or a little faster than 64x the original. From the description, slower. Link to comment Share on other sites More sharing options...
YT2095 Posted January 26, 2004 Share Posted January 26, 2004 the typical layout for parralel proccesing is a Pyramid design, there`s one interface, and that then delegates the task load downwards in the array, one may deal with floating point operations another with raw division and so on, that`s then passed back upwards to the "top of the pyramid" (actualy its a port from an integrator but thats how YOU see it when using it). it won`t be to the 64`th POWER at all, in fact it`ll likely be roughly 60 times the speed allowing for integration multiplexing and pass on time, sure his puter will be WELL FAST! but not quite as he claims, maybe he confused the terms (it was in a PUB after all:) Link to comment Share on other sites More sharing options...
LuTze Posted January 26, 2004 Share Posted January 26, 2004 I doubt it's "64 standard motherboards and a load of cables". It's more likely to be a 64 node cluster, and the speed at which it will do anything depends very much on what it's being used for. It'd be pants at Quake 3. Link to comment Share on other sites More sharing options...
YT2095 Posted January 26, 2004 Share Posted January 26, 2004 using standard mobos isn`t all that good either, polling time and IRQ`s for port access will slow them down quite a bit, REAL PP uses dedicated boards and thereby elliminates this problem in the main, but I`m sure his puter is quite fast, with different architecture and ROM s/ware and few DMA addaptions it would be "up to speed" as it could be Link to comment Share on other sites More sharing options...
SmokingSkillz Posted January 26, 2004 Author Share Posted January 26, 2004 Yeah I was pretty sure about the overall power of the system. He implied that they work of each other almost like a P2P network, but I had also assumed one would be concerned with distributing instructions and retrieving the results. Also for what he described I would have thought the mboards would need some sort of hardware syncronisation and interfacing, as serial or parallel interfaces would be too slow and would require a software layer. It sounded good though! Link to comment Share on other sites More sharing options...
YT2095 Posted January 26, 2004 Share Posted January 26, 2004 p2p`s fine, but running it under s/ware is dead slow (the 2`nd gen PP`s were ran that way though), hadwiring kicks micro seconds ass when it come to fan out and pass on though ports are great where a user interface or slow device (printer, scanner, mouse etc...) are involved, task delegation needs to be there, as does a collator (data integrator). as the saying goes "if you can cut a sub with chip DO IT, the customers paying for it anyway". as in, hardware`s much faster than s/ware even though it costs more Link to comment Share on other sites More sharing options...
SmokingSkillz Posted January 26, 2004 Author Share Posted January 26, 2004 Aye, have you heard of altnet? I can't recall the owner of the network, but it's packaged with freeware like gain and all that crap. It's processing time is sold for things like weather calculations and genome processing. Sounds pretty cool technology, and I guess it could be hacked! I might have a poke around with the exe, I wonder how the listening process differentiates between bonafide altnet instructions and hacker joe's sam file cracking. Link to comment Share on other sites More sharing options...
YT2095 Posted January 26, 2004 Share Posted January 26, 2004 to be honest with ya, the last PP I had any experience with was the N Cube and one or 2 others that I can`t rem the names of as it was so long ago (very late 80`s early to mid 90`s). much of todays modern stuff this or that GFX card or whatever, I know little to nothing about. yes I can and have built a fully working compter from the ground up and writen its firmware, todays stuff just leaves me cold man Link to comment Share on other sites More sharing options...
Radical Edward Posted January 26, 2004 Share Posted January 26, 2004 probably multiplied by 64. a power thing would be impressive indeed. aah, I should buld myself a computer again. I built one using wires and dip switches and I programmed it in binary. twas cool. failing that, learn about parallel processing and build a monster. anyone seen Serial Experiments Lain? The computer she has towards the end of the series using the HTTP8 protocol (I think), That is the sort of computer I want. Link to comment Share on other sites More sharing options...
SmokingSkillz Posted January 26, 2004 Author Share Posted January 26, 2004 well thats the thing he's on about: http://www.esci.keele.ac.uk/geophysics/kage/ and the bizzare cables would be 100MB/s network cables. Your right it'd be crap for Quake3! Link to comment Share on other sites More sharing options...
Guest Syntax Posted January 27, 2004 Share Posted January 27, 2004 Just to clear this up alittle bit, the speed of the processor in NO way has anything to do with how many processors you have, I don't care if you have 10000000000 processors, the computer itself would only become faster due to the amount of processes it can do, not at how fast it does it. It's similar to, if a person was jogging with his clothes on, and a bookbag over his back, and he was running next to a person who was running with a tree trunk on his back, both running their fastest, but at the same speed. Does this clear it up alittle bit? Link to comment Share on other sites More sharing options...
JaKiri Posted January 27, 2004 Share Posted January 27, 2004 Syntax said in post # :Just to clear this up alittle bit, the speed of the processor in NO way has anything to do with how many processors you have, I don't care if you have 10000000000 processors, the computer itself would only become faster due to the amount of processes it can do, not at how fast it does it. It's similar to, if a person was jogging with his clothes on, and a bookbag over his back, and he was running next to a person who was running with a tree trunk on his back, both running their fastest, but at the same speed. Does this clear it up alittle bit? I'm the only one who mentioned resultant processor speed in terms of GHz, and I KNOW; 'overall processor speed equivilent' is what I actually said. Link to comment Share on other sites More sharing options...
YT2095 Posted January 27, 2004 Share Posted January 27, 2004 Syntax, If I give you 1,000 bricks to stack up in a neat pile, one at a time and it takes you 10 hours to do, if there were 2 of you working on the same pile, it would take about 5 and a half hours, just over half the time, and almost twice as fast, we would allow the extra half hour for mistakes about you both going to pick up the same brick or getting in each others way etc... the CPU equiv would be IRQ delays polling delays and integration. that`s probably a bit closer an example than the joggers Link to comment Share on other sites More sharing options...
JaKiri Posted January 27, 2004 Share Posted January 27, 2004 I think he's talking in terms of clockspeed, in which case he's perfectly correct. Link to comment Share on other sites More sharing options...
YT2095 Posted January 27, 2004 Share Posted January 27, 2004 Syntax said in post # :Just to clear this up alittle bit, the speed of the processor in NO way has anything to do with how many processors you have, I don't care if you have 10000000000 processors, the computer itself would only become faster due to the amount of processes it can do, not at how fast it does it. well that in itself sounds contradictory and self defeating. maybe it`s just the way i`m reading it? Link to comment Share on other sites More sharing options...
Sayonara Posted January 29, 2004 Share Posted January 29, 2004 "Parallel processing is done in parallel" shocker Link to comment Share on other sites More sharing options...
YT2095 Posted January 29, 2004 Share Posted January 29, 2004 ya don`t say! Honest? Link to comment Share on other sites More sharing options...
Recommended Posts
Create an account or sign in to comment
You need to be a member in order to leave a comment
Create an account
Sign up for a new account in our community. It's easy!
Register a new accountSign in
Already have an account? Sign in here.
Sign In Now