update hardware overview

- RAM:

  4× 32 GiB DDR4 3200 MT/s CL16

- GPU:

  1× [Radeon RX 570](https://www.amd.com/en/support/graphics/radeon-500-series/radeon-rx-500-series/radeon-rx-570)

  1× [AMD Radeon RX 570](https://www.amd.com/en/support/graphics/radeon-500-series/radeon-rx-500-series/radeon-rx-570)

  (2048 cores @ 1.17 GHz, 8 GiB GDDR5, architecture gfx803)

- Local storage:

    - `/`: 960 GB SSD (KINGSTON SA400S37960G)

## Compute nodes (gp[11-14])

Four compute nodes `gp11`-`gp14` are not accessible from the public network, they must be accessed from a login node (preferably `gp3.fjfi.cvut.cz`).

The compute nodes are reserved for jobs submitted via the [Slurm scheduler](./jobs.md).

- CPU:

  1× [Intel Core i7-9800X](https://ark.intel.com/content/www/us/en/ark/products/189122/intel-core-i7-9800x-x-series-processor-16-5m-cache-up-to-4-50-ghz.html)

  (8 cores @ 3.8-4.5 GHz, 16 MiB cache)

- Local storage:

    - `/`: 960 GB SSD (KINGSTON SA400S37960G)

Note that the `gp14` node has only one GPU instead of two as the other nodes.

## Network

All compute nodes together with the login node are connected to the 10 Gbit Ethernet switch ([TP-Link T1700X-16TS](https://www.tp-link.com/us/business-networking/smart-switch/t1700x-16ts/)).

The compute nodes are not accessible from the outside network, they must be accessed from the login node.

Internet access from the compute nodes is provided via [NAT](https://en.wikipedia.org/wiki/Network_address_translation) on the login node.

There is a 10 Gbit Ethernet switch ([TP-Link T1700X-16TS](https://www.tp-link.com/us/business-networking/smart-switch/t1700x-16ts/)) that provides fast interconnection among the following nodes:

- compute nodes `gp11`-`gp14`

- login nodes `gp2.fjfi.cvut.cz`, `gp3.fjfi.cvut.cz`, `gp8.fjfi.cvut.cz`

Other nodes are not connected to the 10 Gbit switch and their use for distributed computations is therefore very inefficient.

Internet access from the compute nodes is provided via [NAT](https://en.wikipedia.org/wiki/Network_address_translation) on the `gp3.fjfi.cvut.cz` node.

## Storage

The `/mnt/gp3/` file system is __not backed up__ and since it is on RAID 0, even __a single drive failure would mean destruction of all data__.

Hence, users are advised not to keep valuable data here or make their own backups if needed.

The `/mnt/gp3/` storage is shared with compute nodes over network.

## Other nodes

Other nodes (gp{1,2,4,5,6}) are not connected to the 10 Gbit switch and cannot be used for distributed computations.

They also do not have common hardware specifications, see http://mmg.fjfi.cvut.cz/mmg/gpu for details.