Sessions: record, compare, cost and suggestions
The Sessions tab answers "which settings are actually better?". You record how the card behaves while a game or benchmark runs, then compare two recordings, see what a session cost in electricity, and ask for the power limit that saves the most for almost the same speed.
Record a session#
- Start your game or benchmark.
- In Sessions, type a Label for what you are trying (for example the profile name).
- Click Start recording. The button changes to Stop and save.
- Click Stop and save when you are done: Session saved.
A session is a CSV file you can open in a spreadsheet, with a row per second: temperature, hotspot, utilization, clocks, power, fan, video memory, the power limit and why the card was held back, plus an optional fps column. Files are saved in ~/.local/share/voltkit/gpu-sessions/. The page lists the most recent ones with the date, length, average power and, if you set a price, the cost.
From the terminal: voltkit gpu record --seconds 120 --label Quiet.
Compare two sessions#
Tick the check boxes of two sessions and click Compare the two ticked. A larger window lists the measures by category, each naming one better session (or Same). The older session is "before" and the newer one "after".
| Category | Measures compared |
|---|---|
| Temperature | average and highest temperature; time held back by heat |
| Power and energy | average and highest power; energy used (Wh); time held back by the power limit |
| Fan | average fan speed |
| Clock speed | average core clock; active clock |
| Efficiency | active clock per watt |
| Frames | average frames per second; frames per second per watt |
A category is won by the session that is ahead on more of its measures; on a tie, by its first measure with a clear winner. Differences under half a percent count as a tie, so noise never decides. Categories a session cannot measure (a card that reports no power, no fps column) are left out.
Efficiency is a proxy. Active MHz per watt is the core clock weighted by how busy the card was, divided by the power it drew. It says how much clock-work each watt bought, not how many frames. Add an fps column to a session file (for example from MangoHud) and frames per second and frames per watt are reported too.
From the terminal: voltkit gpu sessions lists them and voltkit gpu compare 1 2 compares two by number.
What a session cost#
Type your electricity price in Electricity price per kWh, for example 0.15, 0,15 € or $0.15, and press the apply button. Voltkit saves the number and the currency sign in the configuration file. Each session then shows its energy cost, so you can see what a profile saves over a month. Leave the field empty to turn it off.
Suggest a power limit#
Click Suggest a power limit. Voltkit reads your sessions recorded under different power limits and finds the lowest limit that keeps almost all the speed (within 3 % of the fastest):
- Record the same game or benchmark under at least two different power limits (once per limit).
- Click Suggest a power limit. You get a table of every limit you tried with its performance and power, and a suggestion such as Suggestion: run at about 220 W. It draws 16 % less power for 1.3 % less speed.
How it decides:
- Sessions under 30 seconds, or on a mostly idle card (under 50 % busy), say nothing about a limit and are ignored.
- Limits within 5 W count as the same; several runs at one limit are averaged.
- Performance is frames per second when every session has frames, otherwise active megahertz.
- It only suggests limits you actually measured. If no lower limit stays within the tolerance: No lower limit you tried stays within 3 % of the fastest. Record sessions at more limits to look further.
Then set that limit on the Tune tab or save it as a profile. From the terminal:
voltkit gpu advise # default tolerance 3 %
voltkit gpu advise --tolerance 5 # accept up to 5 % less speed
voltkit gpu advise --gpu 0000:03:00.0