Call it a New Year’s resolution: the EuroHPC Joint Undertaking—which serves as the concerted supercomputing play of the European Union—has issued a new call for a tender as it restarts the procurement process for the long-troubled MareNostrum5 supercomputer. The new call, which solicits a system at a value of some €151,410,000 (around $170.8 million), closes on January 31, 2022.

A couple of years ago, EuroHPC announced plans to procure and deploy eight systems: five petascale systems and three pre-exascale systems. Earlier this year, EuroHPC launched several of those petascale systems, earning a series of spots on the Top500 list. Meanwhile, the pre-exascale LUMI supercomputer at Finland’s CSC has begun deployment and the pre-exascale Leonardo system at Italy’s Cineca is more or less on schedule for deployment in the front half of 2022.

The outlier has been the third system: MareNostrum5, set to be hosted by the Barcelona Supercomputing Center (BSC). Despite assurances that the procurement for MareNostrum5 was set to complete around the end of 2020, the process was still characterized as “pending” by March 2021. Then, in June, the purported reasons for the delay were reported by Politico: broadly, the publication alleged, there was a deadlock between the Spanish stakeholders—who favored a bid from IBM/Lenovo that offered more bang for the buck—and the EU stakeholders, who instead favored a bid from European firm Atos that would reinforce EuroHPC’s priority of boosting homegrown HPC tech. Publicly, those involved in the process chalked up the delays to changes in system specifications. (To learn more about these events, read the prior feature coverage.)

Whatever the cause, the tender for MareNostrum5 was canceled. In June, BSC’s Sergio Girona said that they were “expecting that the relaunch of the procurement [was] in short [order],” pending action by EuroHPC’s governing board, and that they expected “some partitions” of the system to be ready for installation by the end of 2020.

Clearly, that has not come to pass—but now, at long last, there is once again forward movement on the status of MareNostrum5.

The new tender specifically calls for “a pre-exascale high performance computer which will be specifically tailored to strengthen European medical research through drug research, the development of vaccines, virus spread simulations as well as artificial intelligence and big data processing applications,” with other noted applications including climate research, engineering, material science and earth sciences. (This emphasis on biomedical applications lines up with the reported letter from EuroHPC Executive Director Anders Dam Jensen that indicated that MareNostrum5’s tender was being reformulated to better support pandemic-oriented use cases.)

EuroHPC expects the system to be capable of “at least” 205 sustained Linpack petaflops, relatively unchanged from prior expectations of at least 200 petaflops, driven by a heterogeneous machine with both accelerator-based and CPU-based partitions. The system will include a minimum of 204 petabytes of net high-performance storage capacity with 1.6TB/s read and 1.2TB/s write speeds.

EuroHPC is targeting installation of MareNostrum5 beginning in Q3 2022, with at least the compute and storage partitions operational by the end of the year; the “next-generation” technologies long-intended to contribute to the system, however, need only be operational in 2023. As with the first procurement, half the cost will be covered by EuroHPC, with the other half covered by the lead countries (Spain, Portugal and Turkey).

Taking a cue from the ultra-sustainable LUMI system, the tender also indicates that MareNostrum5 will be “energy efficient, fully powered with green energy, and will utilize heat reuse technology,” with a targeted PUE of less than 1.08.

EuroHPC also issued two other calls this month, representing the first systems in the “next generation of EuroHPC supercomputers.” The systems from that call—which ends on February 14, 2022—will include “several mid-range supercomputers” and “one high-end supercomputer” with exascale capabilities. “The JU will co-fund up to 50 percent of the total cost of the high-end supercomputer with a maximum financial contribution of €250 million,” the release reads, “and up to 35 percent of the total cost of the mid-range supercomputers with a total maximum contribution of €120 million.”