Pi has been computed to a record 62.8 trillion decimal places with two EPYC CPUs
A team from the Swiss Fachhochschule Graubünden has calculated pi to a record number of decimal places. With two AMD EPYC 7542 CPUs, the students and researchers reached 62.8 trillion decimal places after 108 days.
The Swiss University used a system with two AMD EPYC 7542 CPUs, each with 32 cores and 64 threads. Calculating pi to far behind the decimal point requires a lot of memory; the system used was equipped with 1 TB of ram and 510 TB of hdds, which were largely used as swap memory. The configuration consisted of 38 HDDs in JBOD with a capacity of 16TB each and a speed of 7200rpm. Of these, 34 were used as swap memory and 4 for pi storage.
The team used y cruncher from developer Alexander Yee for calculating pi. That program can handle properties of modern processors and has also been used for previous world record attempts. The Swiss used Ubuntu 20.04 and made several adjustments to speed up the calculation of pi. For example, security functions and power saving functions were disabled.
The aim of the researchers and students at the Swiss university of applied sciences was to show that they can calculate the irrational number pi to a record number of decimal places with a limited budget, relatively simple hardware and a small team. After 108 days and 9 hours, the team came to 62.8 trillion decimal places. That’s 12.8 trillion more than the previous record.
Images: Fachhochschule Graubünden
In their own words the calculation of the Swiss university of applied sciences was almost twice as fast as Google’s in 2019 and about 3.5 times faster than the last recorded world record from 2020. The researchers are awaiting confirmation from the Guinness Book of Records. When it comes in, they publish the full number. For now, they only give the last ten digits of the result: 7817924264.
In the Guinness Book of Records the most accurate calculation of pi to date is in the name of the American Timothy Mullican. In early 2020, he came to 50 trillion decimal places after eight months. He used for that a server with four Intel Xeon E7-4880v2 CPUs, with 15 cores and 30 threads each. Mullican broke Google’s record, which was set in 2019, with his attempt by a Google Cloud employee. Emma Haruka Iwao calculated on Google pi data center hardware to 31.4 trillion decimal places. An instance with 96 Intel Skylake cores was used for this.
Fachhochschule Graubünden vs. Google vs. Mullican