Graywolf mused: > > I just did a little research (very little actually). The current fastest > supercomputer is made by IBM and just beat out NEC at 36.1 Tera-FLOPS (trillion > floating point opreations per second). Next year they will be delivering one > with almost 10 times that power. Even a little 1 cabinet Cray XD1 does 5 G-FLOPS > (billion flops) and will only set you back $100K or so. > > Tell me again how secure your cipher is from brut force decryption (grin). Next > year's IBM should be able to do that 2048 bit jobby in one year all by its lonesome.
360 TeraFlops? Well, assuming FLOPS & IPS are roughly comparable, that machine will do roughly 360 million MIPS-years in a year of operation. So, by the figures quoted in the document you cited, it should be able to complete the 300,000,000,000,000,000,000 MIPS-years needed to break 2048-bit RSA encryption in, more or less, one million million years. Sounds pretty secure to me. You want to check those calculations again? > BTW the IBM uses AMD processors and most likely runs Linux. I find it amazing > that current state of the art supercomputers are just like our desktops only > they have a few more proccessors (16000 or so). That's been the fastest way to make computers for some time. For some classes of problems (which includes factorisation, weather forecasting, aerodynamics, and simulating nuclear explosions) this kind of approach works really well, becaise it's easy to break the problem into small pieces with only limited communication between the individual parts. For other problems the biggest problem is moving data between the different processors. This is why SGI (and IBM) manage to sell 4000-processor machines which cost a great deal more than 4000 times the price of a single processor; the extra cost is in the massive crossbar data switches, multi-port data cacheing systems, etc.
