Skip to content

Significant changes to Puhti & Mahti authentication coming in April! Read about the SSH key and multi-factor authentication requirements.

Technical details about Roihu

Note

This page contains preliminary information about CSC's next national supercomputer Roihu, which is projected to be in researchers' use early 2026. Please note that the details may evolve over time.

Compute

Roihu will have a total of 486 CPU nodes and 132 GPU nodes. The high-performance LINPACK (HPL) performance is estimated to be 10.5 PFlop/s for the CPU nodes and 23.4 PFlop/s for the GPU nodes, resulting in an aggregate HPL performance of 33.9 PFlop/s for the full system.

The CPU nodes will have two 192-core AMD Turin 9965 CPUs each, amounting to 186 624 CPU cores altogether. The CPUs are based on the AMD Zen 5 architecture, which supports the AVX-512 vector instruction set. 414 of the CPU nodes will have 768 GiB of memory, while the remaining 72 nodes will have an extended memory of 1 536 GiB each.

Each GPU node will be equipped with 4 Nvidia GH200 Grace Hopper superchips. Each GH200 superchip comprises one Hopper GPU and one Grace CPU with 72 ARM CPU cores which are connected with a very fast interface. Each GH200 superchip has 120 GiB CPU memory and 96 GiB GPU memory, providing a total of 480 GiB CPU memory per node. This gives a total of 528 GPUs and 38 016 CPU cores in the whole GPU partition.

The system will also provide four visualization nodes with two Nvidia L40 GPUs each, as well as four high-memory CPU nodes with 3 TiB memory and higher single-thread performance.

Nodes

Name Number of nodes Compute Cores Memory (GiB) Local disk (TB)
M 414 AMD Turin 9965 2 x 192 cores (x86) @ 2.25 GHz 768 0.96
L 72 AMD Turin 9965 2 x 192 cores (x86) @ 2.25 GHz 1536 0.96
XL 4 AMD Turin 9555 2 x 64 cores (x86) @ 3.20 GHz 3072 15.36
V 4 AMD Turin 9335
Nvidia L40		 2 x 32 cores (x86) @ 3.40 GHz
2 x GPUs		 384
2 x 48		 0.96
GPU 132 Nvidia GH200 4 x 72 cores (ARM)
4 x GPUs		 4 x 120
4 x 96		 0.96

Storage

Roihu will have two independent flash-based DDN EXAScaler Lustre file systems – a 6.0 PiB scratch space and a 0.5 PiB storage system for project applications and users' personal home directories. Separate file systems will ensure responsiveness of home and projappl even under heavy scratch usage.

The peak I/O performance of Roihu scratch space is expected to be around 200 GB/s for read and 170 GB/s for write. The home and projappl will have read and write bandwidths of 120 GB/s and 100 GB/s, respectively.

Contrary to Puhti and Mahti, all Roihu CPU and GPU nodes will have a small 960 GB fast NVMe disk for efficient storage of, for example, temporary files. The high-memory nodes will include 2 x 7.68 TB fast disks each.

Network

The network of Roihu is based on Infiniband NDR interconnect. Each CPU node will be connected to the network with one 200 Gb/s link, while in the GPU partition there will be four 200 Gb/s links per node, one for each GPU.