The High-Performance Computing Center Stuttgart (HLRS) has decided on the components of its next supercomputer, Herder. It will again feature processors and GPU accelerators from AMD; this time, however, in separate form.
According to current plans, Herder will receive 1488 accelerators from AMD’s upcoming 400 series, the Instinct MI430X, paired with 372 Epyc processors of the upcoming Venice type. It is the first processor with Zen 6 architecture and 2-nanometer technology from chip contract manufacturer TSMC. The [previous system, Hunter](https://www.heise.de/news/HRLS-Hunter-Erst…
The High-Performance Computing Center Stuttgart (HLRS) has decided on the components of its next supercomputer, Herder. It will again feature processors and GPU accelerators from AMD; this time, however, in separate form.
According to current plans, Herder will receive 1488 accelerators from AMD’s upcoming 400 series, the Instinct MI430X, paired with 372 Epyc processors of the upcoming Venice type. It is the first processor with Zen 6 architecture and 2-nanometer technology from chip contract manufacturer TSMC. The previous system, Hunter, still uses Instinct MI300A accelerators, which combine CPU and GPU on a single carrier.
Each Instinct MI430X comes with 432 GByte of High Bandwidth Memory (HBM4), which transmits 19.6 TByte/s. This results in nearly 643 TByte of HBM4 system-wide, in addition to the DDR5 RAM for the Epyc CPUs. The amount of DDR5 RAM is not yet known.
Two CPUs and eight GPUs each are housed on a motherboard from HPE’s new Cray GX5000 platform. Herder will feature liquid cooling, with the waste heat used to heat the Vaihingen campus of the University of Stuttgart.
Construction in 2027
Ultimately, Herder represents a smaller sibling of the US exascale system Discovery. However, HLRS will not have to wait quite as long: HPE will begin construction in the second half of 2027. The system is expected to be operational by the end of 2027.
The center predicts that Herder’s peak performance will increase by at least a factor of 7 compared to the current Hunter system. This would correspond to at least 340 Petaflops in double-precision FP64 (64-bit floating-point values). For AI applications, the computing power is significantly higher because simpler data formats like FP8 and INT8 (8-bit integers) are sufficient.
HLRS sees computer science, engineering, physics, chemistry, climate science, and biomedicine as application areas.
The costs amount to approximately 100 million Euros. The Federal Ministry of Research, Technology and Space, as well as the Ministry of Science, Research and the Arts of the state of Baden-Württemberg, will cover the costs.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externer Preisvergleich (heise Preisvergleich) geladen.
Preisvergleiche immer laden
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (heise Preisvergleich) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
(mma)
Don’t miss any news – follow us on Facebook, LinkedIn or Mastodon.
This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.