You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
README: add GPU heat/ventilation warning; refresh stale GUI docs
- New "Heat and power" section up top: sustained 100% GPU load runs hot and
costs power; keep good ventilation, watch temps, stop (Ctrl-C / tray Quit)
and improve cooling if it gets hot.
- Fix docs the ldtray migration made false: no GTK/AppIndicator/libxdo to
install (tray loads the toolkit at runtime), gui is the default build
(--no-default-features for a pure worker), and the tray menu now shows
GPUs, live speed, and Pause/Resume.
- Options table: add --gpus and --cache-max-gb, correct the --workdir default
(per-user data dir), note --jobs is per-GPU.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Copy file name to clipboardExpand all lines: README.md
+40-23Lines changed: 40 additions & 23 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -5,17 +5,32 @@ and it quietly does distributed compute jobs in the background: it asks the
5
5
coordinator for a chunk of work, runs it on your GPU, sends the result back, and
6
6
repeats — forever, until you stop it. Run and forget.
7
7
8
+
## ⚠️ Heat and power
9
+
10
+
decryptd keeps your GPU at **full load, continuously** — that's how it does the
11
+
work. A card pinned at 100% for hours pulls serious power and runs **hot**, and
12
+
sustained heat is what wears graphics hardware out.
13
+
14
+
- Make sure the machine has **good ventilation** and unobstructed airflow.
15
+
Enclosed cases, laptops, and dust-clogged fans that cope with short bursts can
16
+
overheat under a non-stop load.
17
+
- Keep an eye on temperature (`nvidia-smi -l 5` on Linux, or any GPU monitor). If
18
+
the card runs hot, **stop decryptd** — `Ctrl-C`, or **Quit** from the tray — and
19
+
improve cooling before running it again.
20
+
- Expect a higher **power bill** and a warmer room. On a laptop, run it plugged in
21
+
rather than on battery.
22
+
23
+
If in doubt, stop it — the work is distributed, so one machine sitting out is
24
+
completely fine.
25
+
8
26
## Requirements
9
27
10
28
- A **CUDA-capable NVIDIA GPU**.
11
29
- An **up-to-date NVIDIA driver**. No CUDA toolkit needed.
12
30
- Linux or Windows (64-bit).
13
-
-**Linux only:** the release binary includes the system-tray UI, so it needs
14
-
GTK 3 and an AppIndicator library present — `libgtk-3`,
15
-
`libayatana-appindicator3`, and `libxdo`. These ship with essentially every
16
-
desktop install. On a **headless server**, install them (e.g.
17
-
`apt install libgtk-3-0 libayatana-appindicator3-1 libxdo3`) or build without
18
-
the tray (`cargo build --release`). Windows bundles everything in the `.exe`.
31
+
-**No GUI libraries to install.** The system tray loads whatever the desktop
32
+
provides at runtime, so it just works on a desktop and is silently skipped on a
33
+
headless server — the same binary runs either way.
19
34
20
35
## Get it
21
36
@@ -59,29 +74,31 @@ You normally don't need any of these.
59
74
| Option | Default | What it does |
60
75
| --- | --- | --- |
61
76
|`--once`| off | Do a single chunk of work, then exit (handy for testing). |
77
+
|`--jobs <N>`|`1`| How many chunks to run on the GPU at once, **per GPU**. |
78
+
|`--gpus <LIST>`| all | Which GPUs to use, e.g. `--gpus 0,2`. Default: every detected GPU, each with its own work queue. |
62
79
|`--idle-secs <N>`|`60`| How long to wait before re-checking when there's no work. |
63
-
|`--jobs <N>`|`1`|How many chunks to run on the GPU at once. |
64
-
|`--workdir <DIR>`|`decryptd-data`| Where to keep the download cacheand scratch files. |
80
+
|`--cache-max-gb <N>`|`20`|Cap the on-disk download cache; oldest blobs are evicted past this. |
81
+
|`--workdir <DIR>`|per-user data dir | Where to keep the download cache, worker id, and scratch. Defaults to a per-user folder (`%LOCALAPPDATA%\decryptd` / `~/.local/share/decryptd`). |
65
82
66
83
Downloading the next chunk and uploading finished results always happen in the
67
84
background while the GPU works, so the card stays busy. `--jobs` only raises how
68
85
many run *on the GPU* simultaneously — most setups are fine with the default.
86
+
With more than one GPU, `--jobs` applies to each, so `--jobs 3` on 2 GPUs runs 6
87
+
chunks in total.
69
88
70
89
Run `decryptd --help` for the full list.
71
90
72
91
### System-tray mode
73
92
74
-
The released binaries run as a **system-tray app** on Windows and Linux. decryptd
75
-
sits in the tray with a right-click menu showing the version, the current status
76
-
(*Waiting* or *Running*), and a **Quit** entry. The worker runs in the
77
-
background exactly as above.
78
-
79
-
If no tray host is available (for example a Linux box with no desktop session),
80
-
decryptd logs a notice and falls back to running headless. Passing `--once` also
81
-
runs headless, with no tray.
93
+
The released binaries run as a **system-tray app** on Windows and Linux. Right-click
94
+
the tray icon for a menu showing the version, current status (*Waiting* /
95
+
*Running* / *Paused*), the GPU(s) in use, the live **speed** (tries per second,
96
+
1-minute average), a **Pause/Resume** toggle, **Check for Updates**, and **Quit**.
97
+
The worker runs in the background exactly as above.
82
98
83
-
To build a **console-only** binary with no tray (and no GUI dependencies), build
84
-
the default feature set — see [Building from source](#building-from-source).
99
+
The tray loads the desktop's toolkit at runtime, so there are **no GUI libraries
100
+
to install**. If no tray host is available — a headless server, or `--once` —
101
+
decryptd logs a notice and runs without a tray.
85
102
86
103
## Building from source
87
104
@@ -90,13 +107,13 @@ against the driver library at build time — the binary still just needs the
90
107
driver to run):
91
108
92
109
```sh
93
-
cargo build --release # console-only, no GUI dependencies
0 commit comments