Activity Monitor
Text

Activity Monitor

Activity Monitor shows what a machine's CPU, graphics cards and network ports are doing, every two seconds, drawn as pixel art. It shows this node, or any node you joined to it (Joining desks together).

  • Pick a node with the Node pop-up at the top. Each node wears its icon (Configuration → General → Icon on that node), grey while it does not answer. The window's title names the node, and the window opens on it the next time.
  • The CPU is a chip. Each tile on its die is a core, and each column in a tile is one of the core's threads, lit from the bottom like a level meter: green, then amber, then red. A slower kind of core is a smaller tile, as the GB10's Cortex-A725 cores are under its Cortex-X925 ones. The die turns warmer as the CPU heats. Rest the pointer on a tile to read its threads' load and its clock.
  • Beside the chip: the load of all the cores, the busiest thread, the average clock of its top, the hottest sensor, and memory in use. Load average is the system's own, over the last 1, 5 and 15 minutes; its meter is the last minute's against the number of threads. Uptime is how long the machine has run since it started. Rest the pointer on it to read when that was.
  • Under the chip: each thread's load over the last minutes, a row each, the newest at the right. A row is dark when idle and goes through blue and green to red at full load.
  • Each graphics card is its board. The cells of its die light as the GPU works, the memory chips round the die light as its memory fills, and the fan turns as fast as the card's own. A GPU that shares the machine's memory, as the GB10 does, is a package instead: the GPU die, a CPU die with a cell for each core, and memory chips that light as the machine's memory fills.
  • Beside the card: load, memory, temperature, power (of the card's limit, where it names one), the graphics clock, the fan, and the video encoder and decoder, each where the card gives it. Under the card, its load over the last minutes.
  • The network is the machine's back panel: a jack for each wired port and an antenna for Wi-Fi. A port whose link is up has a cable in it, its link lamp lit (green at a gigabit or more, amber under) and its activity lamp blinking with the traffic. The antenna's waves light by the signal.
  • Beside the panel, for each interface: its model (the maker and chip of its controller, such as Realtek RTL8127), its link (its speed, or Not connected), and what it received and sent a second, each on a meter that counts in tens from 1 KB/s up to the link's speed. Wi-Fi has its signal too. Under the panel, each interface's traffic over the last minutes: received above the line, sent below it.

The histories under the pictures are pixel art too. Each pixel at the top right says how long a pixel stands for:

  • 2 seconds: each sample of the last hour, so a wider window shows further back, up to the hour.
  • 1 minute, 10 minutes, 1 hour and 1 day: the daemon's history, each pixel the average of the samples in that minute, hour or day. Behind it, in a darker shade of its colour, is their peak, so a burst still shows in an hour that was mostly idle. The CPU's rows show averages only.

The daemon keeps a minute a row for a week, ten minutes for 90 days, an hour for two years, and a day for good, in ~/.exe/activity: files that never grow but the days', which grows a row a day (a few megabytes in all, over twenty for a machine of 72 threads and two cards). A time the daemon did not run is a gap. A node on an exe from before the history shows its last hour only, and its other scales are grey.

The cards' figures come from nvidia-smi, kept running and read every two seconds.

The figures are the daemon's own samples, the same ones as the Control Strip's activity module (GET /v1/host/monitor?detail=1). The daemon reads them on Linux, the cards through nvidia-smi, and only the physical network interfaces: not loopback, bridges, VM taps or tunnels, so each byte counts once. It names each controller from the system's list of PCI devices (pci.ids), or from its driver where the chip is newer than the list; a USB adapter names itself.

On a Mac it reads them from macOS: each core, the performance cores ahead of the efficiency ones, memory as Activity Monitor counts it, the Apple chip's GPU and the temperatures of its CPU clusters and GPU. macOS keeps the clocks, the power and the Wi-Fi signal behind an administrator's password, so a Mac shows no figure for those.

On Windows it reads them the way Task Manager does, with no administrator: each core and its clock, memory, the network adapters that are hardware (not Hyper-V's, VMware's, the network bridge or Tailscale) under their names in Windows, such as Ethernet and Wi-Fi, the Wi-Fi signal, the physical disks (not a Storage Space, whose bytes are its disks'), and the NVIDIA cards through nvidia-smi. Windows gives no top clock with the boost, so the clock shows alone, and no CPU temperature without a driver of its own. It keeps no load average, so a Windows node has no Load average row. Another node's figures come through this node: its daemon asks the other one over Node Sync's signed link (GET /v1/peers/<id>/monitor). That node must run an exe of this release or later; the window says so when it does not. With Reduce motion on in your system, the fan and the die's cells stay still.

Contents