This entry was posted
on Thursday, January 31st, 2008 at 4:24 pm and is filed under geek.
You can follow any responses to this entry through the RSS 2.0 feed.
You can leave a response, or trackback from your own site.
I’m not sure that proves that you’re human. It does prove that you’re a geek. Of course, just going to a website to download random numbers already pretty much established that.
I have a question though, how much of a need do you really have for truly random numbers? If it’s a critical need, it might be handy to get your own machine to do it. VIA has a hardware random number generator that uses thermal noise to generate random numbers in the newer versions of their processors. This lets you generate up to like 20 million truly random numbers per second. These processors are available on mini-ITX motherboards just like the one that dactyl uses, so you could build a full machine in the $300 range. (Or, for around $200, you could try to use grant money to just upgrade dactyl.)
Actually, looking at wikipedia, it looks like the processor in dactyl might actually have some less fancy hardware RNGs onboard to make real random numbers. I’ll look into whether that’s actually true or not.
These types of process might be useful for systems where one is actually seeking a vague notion of randomness, such as rpg’s and visual/sonic related graphics rendering. For physics applications which use O(10^9) rng’s then it is actually safer to use a well understood pseudo-rng algorithm which as mathematically proven distributions and correlations. After all, most physics applications do not really on the actual “randomness” of each number, but the distribution that those number create when looked at together. This basically casts doubt on any physically induced rng that come from a “complex dynamical system” such as on the via chip. The complexity of the system sounds by it’s description to be actually lacking compared to say the mersenne primes algorithm (mersenne twister). http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html
We actually don’t really need that “true” random number, but the old pseduo-rng in Matlab produces quite significant patterns. However I think the new method Matlab implemented after 7.3 - the Twister method as pointed out by Joel - provides good enough random numbers for most purposes. I tried several rng services from the web and this quantum one seems to be the best. They are selling the unit too so I think in some way the free service they provide is a good ad. Nevertheless the distribution of the random numbers matters more. Some of these generators may be very “random”, but they often don’t produce a nice distribution. Actually I found out that a relatively popular matlab rng code which uses different sources such as random.org doesn’t really provide good uniform distribution at all, somehow it favors small numbers very close to 0.
One very good thing about the pesudo-rng is the simulation will be “repeatable” if using the same seeds.
February 1st, 2008 at 4:13 pm Using
I’m not sure that proves that you’re human. It does prove that you’re a geek. Of course, just going to a website to download random numbers already pretty much established that.
I have a question though, how much of a need do you really have for truly random numbers? If it’s a critical need, it might be handy to get your own machine to do it. VIA has a hardware random number generator that uses thermal noise to generate random numbers in the newer versions of their processors. This lets you generate up to like 20 million truly random numbers per second. These processors are available on mini-ITX motherboards just like the one that dactyl uses, so you could build a full machine in the $300 range. (Or, for around $200, you could try to use grant money to just upgrade dactyl.)
Actually, looking at wikipedia, it looks like the processor in dactyl might actually have some less fancy hardware RNGs onboard to make real random numbers. I’ll look into whether that’s actually true or not.
February 1st, 2008 at 5:31 pm Using
These types of process might be useful for systems where one is actually seeking a vague notion of randomness, such as rpg’s and visual/sonic related graphics rendering. For physics applications which use O(10^9) rng’s then it is actually safer to use a well understood pseudo-rng algorithm which as mathematically proven distributions and correlations. After all, most physics applications do not really on the actual “randomness” of each number, but the distribution that those number create when looked at together. This basically casts doubt on any physically induced rng that come from a “complex dynamical system” such as on the via chip. The complexity of the system sounds by it’s description to be actually lacking compared to say the mersenne primes algorithm (mersenne twister). http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/emt.html
February 1st, 2008 at 8:44 pm Using
We actually don’t really need that “true” random number, but the old pseduo-rng in Matlab produces quite significant patterns. However I think the new method Matlab implemented after 7.3 - the Twister method as pointed out by Joel - provides good enough random numbers for most purposes. I tried several rng services from the web and this quantum one seems to be the best. They are selling the unit too so I think in some way the free service they provide is a good ad. Nevertheless the distribution of the random numbers matters more. Some of these generators may be very “random”, but they often don’t produce a nice distribution. Actually I found out that a relatively popular matlab rng code which uses different sources such as random.org doesn’t really provide good uniform distribution at all, somehow it favors small numbers very close to 0.
One very good thing about the pesudo-rng is the simulation will be “repeatable” if using the same seeds.