←back to thread

One year, 41M digits: How Luke Durant found the largest known prime number

(www.washingtonpost.com)
60 points ortusdux | 3 comments | 06 Nov 24 20:56 UTC | HN request time: 0s | source
Show context
evilsaloon ◴[07 Nov 24 04:13 UTC] No.42073337[source]▶
> Durant, a graduate of the California Institute of Technology, found the new prime number using only publicly available unused cloud storage space.

> I was able to find this number that’s astonishingly large … but I was able to do it just by using big tech’s leftovers.

Did he do this by taking advantage of spot pricing? It isn't actually mentioned how he uses those leftovers in the article.

replies(1): >>42073846 #
1. NavinF ◴[07 Nov 24 05:59 UTC] No.42073846[source]▶
Yikes! If so, he could have saved a lotta money by purchasing 4090s instead of spot pricing. Last time I did the math, break even between buying and renting was only 3 months. I wonder if the search took a lot longer than he anticipated
replies(1): >>42074808 #
2. nwiswell ◴[07 Nov 24 08:37 UTC] No.42074808[source]▶
Once physical hardware is involved, the problem gets a lot more complicated.

If you're talking about $2M worth of 4090s you have to concern yourself not only with putting them in hosts, but also with adequate space, power distribution, cooling, and networking. You have to figure out how to gracefully handle hardware failures. You have to install operating systems, or at least come up with some mechanism to deploy automatically. And you just straight up have to supply labor to get 1000+ of these things plugged in.

All that before you can get even get to the "fun part". If you're a multi-millionaire you might be willing to just pay the spot price premium.

replies(1): >>42075234 #
3. NavinF ◴[07 Nov 24 09:53 UTC] No.42075234[source]▶
Naw the dude had a year. It's not that hard to pay 1 contractor $30/hr to set up PXE boot, assemble machines and GPUs, air handling units, GPUs, and electrical sub panels. I've done a similar "jack of all trades" job for a weed company in Oakland way back before I became a software engineer. 1000 GPUs only takes 2 weeks to deploy.