The Swiss University of Applied Sciences Graubünden challenged Pi’s computational world record, claiming to have calculated the mathematical constant at 62.8 trillion digits.
The university claimed yesterday that it broke the record, saying it broke the previous 50 trillion digit record held by Timothy Mullican of 12.8 trillion digits and did so 3.5 times faster .
Thankfully, the uni has also released details of the hardware used for their feat.
A pair of 32-core AMD Epyc 7542 processors powered the uni platform. AMD claims that the processor cores spend most of their time at 2.9GHz, can reach 3.4GHz, have 128MB of L3 cache, and happily run 64 threads each. A server with 1TB of RAM was also used, with Ubuntu Linux 20.04 installed on a pair of SSDs of unspecified size.
A JBOD housed 38 hard drives at 7200 rpm, each with a capacity of 16 TB.
Thirty-four of these disks were used to store values exchanged from RAM – a design chosen because memory is very expensive. Hard drives were preferred over SSDs because SSD performance degrades over time and university designers were concerned their compute-intensive problems would cause problems. In total, uni said 510TB of disk space has been used.
The uni’s platform was modest compared to the cloudy 96 vCPU effort Google used to calculate Pi to 31.4 trillion decimal places in 2019.
An application called y-Cruncher did the job and was configured to move data in parallel from the server to the 34 drives at around 8.5 GB / s.
Experienced readers will have noted that the JBOD housed 38 disks – the other four were used to store the value of Pi itself.
The last ten digits stored on these disks are 7817924264 and these are now the last known numbers of Pi, if the uni calculations are correct – and they have not yet been independently verified. If the world record is confirmed by Guinness, the full number will be released, the uni said. ®