←back to thread

359 points sdsykes | 1 comments | | HN request time: 0.293s | source
Show context
p5a0u9l ◴[] No.41887886[source]
Are there statistics on the scale of compute available to GIMPS for this search? Is there any evidence that by crowdsourcing the clients, we are searching faster than, eg, a dedicated cluster financed by a government or a corporation? What is the impact of GIMPS as a distributed problem solving tool? Like, if there was a practical application, how much money would it take to exceed GIMPS throughput, that curious people provide for free?

I’d like it to be astronomical, but given the niche of this, and the low cost of cloud compute, the answer is predicable depressing, like, “$50k/year in AWS costs would equal current GIMPS search throughput”

replies(2): >>41888774 #>>41889000 #
1. areyousure ◴[] No.41889000[source]
https://www.mersenne.org/primenet/ suggests 127 PFlop/s average over the last month, which would put it in the top 10 on https://top500.org/lists/top500/2024/06/

I used a random estimate online for computing cost which had 5.6e17 Flops per dollar on A100s gives about a dollar every 4.4 seconds or ~$7 million per year.

Sadly, I do not vouch for the correctness of any part of this, though I did try.