The $1,000,000 PC set a WORLD RECORD
0 up · 0 down · 0 ratings
Description
Thanks to KIOXIA for sponsoring this video! Learn more about their fantastic SSDs at: americas.kioxia.com americas.kioxia.com When Linus was a young lad, Linus loved using Super PI, a benchmark that calculates PI (who would have thought). Back in the day, he could only calculate 32 million digits of PI, but that was with an old Athlon CPU. Find out how many we can calculate now with the help of KIOXIA (it's a lot). Discuss on the forum: linustechtips.com
Check out Y-Cruncher: lmg.gg
Check out Weka: lmg.gg
Promos
Check out NVIDIA's ConnectX-7 Network Adapters: lmg.gg Buy Micron's 128GB DDR5-5600MHz CL46 ECC RDIMM Memory: geni.us ► GET OUR MERCH: lttstore.com ► GET EXCLUSIVE CONTENT ON FLOATPLANE: lmg.gg ► GET A VPN: piavpn.com ► SPONSORS, AFFILIATES, AND PARTNERS: lmg.gg Purchases made through some store links may provide some compensation to Linus Media Group. CHAPTERS --------------------------------------------------- 0:00 Intro 1:47 How did we get here? 2:28 The hardware 5:07 Looking at the compute node 6:52 Some WEKA trickery was needed 8:35 We gotta tune it 9:42 How fast does she go 10:40 Survey says, very fast 11:58 KIOXIA SSDs ROCK 12:45 y-cruncher go time 14:30 Back to August 1, 2024 16:43 Let's look at some logs 17:30 How long will the record be safe for? 17:52 Linus with the real talk 18:04 It's all about the journey 18:55 Outro
The video documents Linus Tech Tips’ ambitious project to break a Guinness World Record for calculating pi, using a highly customized compute cluster assembled from nine servers, including a million-dollar PC and additional storage nodes. The hosts explain the constraints of pi digit calculations, noting that storage and memory bandwidth become the real bottlenecks as digits scale into the trillions. With sponsorship from Kioxia, they assemble a massive Gen 4 NVMe storage pool totaling over two petabytes of raw capacity across nine servers, enabling the required data throughput for the computation. The build includes high-end CPUs, multiple NVIDIA ConnectX network cards, and a Weta-based file system integration, all orchestrated to run YC Cruncher across a shared storage pool. After tuning software and hardware alignment, the team demonstrates sustained high throughput and reports on the final result, the 300 trillion digit pi calculation, achieved after a long, iterative process and a few outages along the way. They reflect on the journey, the teamwork, and the implications of such an extreme storage and compute setup, emphasizing the experience and collaborative effort over the destination alone.
Topics · technology · science · data_center · computing · ai_and_machine_learning · hardware
Questions answered
- What factors most limit the pi calculation at such scales?
- The main limiting factor is storage and memory bandwidth rather than raw compute power, so the project emphasizes very large, fast NVMe storage pools and high RAM capacity to feed the calculation efficiently.
- How many servers were used and why so many?
- Nine servers were used to expand storage capacity beyond what a single machine could hold and to optimize stripe width and performance, enabling a unified file system to feed YC Cruncher across the cluster.
- What is YC Cruncher and why is it used here?
- YC Cruncher is a software tool designed to compute digits of pi; it benefits from high I/O throughput and memory, which is why the team tuned storage, RAM, and NUMA configurations to maximize performance.
- Was the exact final digit important beyond breaking the record?
- The primary goal was to set a world record and demonstrate the capability of the setup; the exact final digit served as a milestone, while the project also highlighted the journey, coordination, and technological innovations involved.