We’re proud to announce that Sherlock has reached another significant performance milestone. Building on past successes, Sherlock continues to evolve and expand, integrating new technologies and enhancing its capabilities to meet the growing computational needs of Stanford researchers.

Computing power doubles

Sherlock's theoretical computing power has nearly doubled in just a few months, with the addition of new compute nodes purchased by over 50 different Faculty groups. The cluster now features more than 2,000 compute nodes totalling over 70,000 CPU cores. And with more than 1,000 GPUs, Sherlock can now deliver over 11 Petaflops of computing power — equivalent to 11 thousand million million (10¹⁵) double-precision floating-point operations per second.

This expansion has been driven by strategic investments from various research groups and schools, including the Stanford Doerr School of Sustainability and the School of Humanities and Sciences.

Growth trajectory

Since its inception in 2014, Sherlock has grown 16x in terms of compute nodes, more than 35x in CPU cores, and over 100x in raw computing power! Four generations and as many Infiniband fabrics later, expansion continues stronger than ever:

Interconnectivity and Infrastructure

Sherlock's infrastructure has continued to grow, now occupying two datacenter rooms with over 50 racks of equipment and consuming more than 800kW of power. The system relies on over 70 Ethernet switches to connect all nodes and utilizes an extensive network of over 130 Infiniband switches for high-speed connections between storage and nodes. This setup requires more than 6,500 Infiniband cables, which collectively span over 20 miles—equivalent to the distance from campus to the San Francisco International Airport.

For even more facts and numbers, checkout the Sherlock Facts page!

Sherlock 2.0 retirement

As Sherlock expands, we will soon begin retiring the aging Sherlock 2.0 generation of compute nodes and its supporting infrastructure. Launched in 2017, Sherlock 2.0 has been a reliable computing platform for Stanford research for the last 8 years, twice its planned lifespan. However, maintaining large-scale computing operations on older hardware presents various challenges, including reduced performance, lower power efficiency, obsolescence issues, and declining reliability. Decommissioning this older equipment will enable us to concentrate on future developments and advance Sherlock into the next era of computing.

Thank you!

Finally, we’d like to extend our gratitude to the entire Sherlock user community for their ongoing support and enthusiasm. Your contributions and feedback are invaluable, and we look forward to continuing to support your groundbreaking research endeavors.

As usual, please don’t hesitate to reach out if you have any questions, and happy computing!