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
Superset consumes 5GB of RAM and 33% (varies a lot) of CPU, even when idle.
I currently have 12 workspaces open.
Superset is great of my daily work, being able to have many workspaces open, each with their terminal tabs.
This is a big upgrade from my previous workflow where I had 1 terminal window with 10+ tabs.
The issue is that I notice a huge energy consumption on my laptop, generating a lot of heat constantly.
It seems like workpaces, whether they are used or not, are consuming RAM and CPU, instead of turning into an idle state where they don't consume any energy.
I already found that Cursor and Chrome were using a lot of RAM and energy, but Superset is just 10x that.
I looked around in the settings, but there's no energy saving mode or anything.
I dug into the desktop app to see what could be keeping things busy when workspaces are idle. I couldn't write a bun:test reproduction — this is a runtime resource-consumption issue in Electron (RAM/CPU under load with many workspaces), which unit tests don't measure — but I did find several concrete contributors worth sharing:
1. Main window background throttling is disabled unconditionally on macOS. apps/desktop/src/main/windows/main.ts:135 calls window.webContents.setBackgroundThrottling(false) whenever the app is on macOS. The comment above it explains why (macOS Sequoia+: background throttling can corrupt GPU compositor layers), but the effect is that the entire renderer — and every embedded workspace webview hosted in it — keeps running at full speed even when Superset is not the focused application. There is no blur/hide handler that re-enables throttling when the user switches to Chrome/Cursor.
2. All 12 workspaces share a single renderer process.
Workspaces are not separate BrowserWindows; they are panes/webviews inside the one main window. Per-pane webviews do call wc.setBackgroundThrottling(true) (apps/desktop/src/main/lib/browser/browser-manager.ts:49), but the host renderer's throttling overrides that on macOS while the app is unfocused.
3. No workspace visibility tracking in the renderer.
There is no code that pauses timers, animations, or React work for non-active workspace tabs — every mounted workspace continues to render/animate even when only one is visible.
4. Several always-on intervals in the main process (smaller contributors):
apps/desktop/src/main/index.ts:288 — parent-process liveness check every 1s
apps/desktop/src/main/lib/host-service-coordinator.ts:487 — adopted-process liveness check every 5s per workspace
apps/desktop/src/main/lib/notifications/notification-manager.ts:54 — notification sweep every 5min
apps/desktop/src/main/lib/auto-updater.ts:322 — update check every 4h
The biggest lever is almost certainly #1 + #3: re-enable background throttling on the main window when the app loses focus (using the existing invalidate() workaround on regain to avoid the GPU-layer corruption the team originally hit), and add visibility-based suspension for non-active workspace tabs.
I'm not opening a PR because (a) the fix touches sensitive GPU-throttling code with a documented prior regression, and (b) there's no good way to land it behind a unit test — it needs manual validation on macOS Sequoia+ with multiple workspaces. Flagging for a maintainer to take from here.
Closing as a duplicate of #5602, the consolidated tracker for stale performance issues from the July 2026 stale-issue sweep (no activity since 2026-06-01). If this still applies on a current build, comment on #5602 and we'll reopen.
Activity
github-actions commentedon May 11, 2026
Triage note from automated investigation.
I dug into the desktop app to see what could be keeping things busy when workspaces are idle. I couldn't write a
bun:testreproduction — this is a runtime resource-consumption issue in Electron (RAM/CPU under load with many workspaces), which unit tests don't measure — but I did find several concrete contributors worth sharing:1. Main window background throttling is disabled unconditionally on macOS.
apps/desktop/src/main/windows/main.ts:135callswindow.webContents.setBackgroundThrottling(false)whenever the app is on macOS. The comment above it explains why (macOS Sequoia+: background throttling can corrupt GPU compositor layers), but the effect is that the entire renderer — and every embedded workspace webview hosted in it — keeps running at full speed even when Superset is not the focused application. There is noblur/hidehandler that re-enables throttling when the user switches to Chrome/Cursor.2. All 12 workspaces share a single renderer process.
Workspaces are not separate
BrowserWindows; they are panes/webviews inside the one main window. Per-pane webviews do callwc.setBackgroundThrottling(true)(apps/desktop/src/main/lib/browser/browser-manager.ts:49), but the host renderer's throttling overrides that on macOS while the app is unfocused.3. No workspace visibility tracking in the renderer.
There is no code that pauses timers, animations, or React work for non-active workspace tabs — every mounted workspace continues to render/animate even when only one is visible.
4. Several always-on intervals in the main process (smaller contributors):
apps/desktop/src/main/index.ts:288— parent-process liveness check every 1sapps/desktop/src/main/lib/host-service-coordinator.ts:487— adopted-process liveness check every 5s per workspaceapps/desktop/src/main/lib/notifications/notification-manager.ts:54— notification sweep every 5minapps/desktop/src/main/lib/auto-updater.ts:322— update check every 4hThe biggest lever is almost certainly #1 + #3: re-enable background throttling on the main window when the app loses focus (using the existing
invalidate()workaround on regain to avoid the GPU-layer corruption the team originally hit), and add visibility-based suspension for non-active workspace tabs.I'm not opening a PR because (a) the fix touches sensitive GPU-throttling code with a documented prior regression, and (b) there's no good way to land it behind a unit test — it needs manual validation on macOS Sequoia+ with multiple workspaces. Flagging for a maintainer to take from here.
Refs #4412
[-][bug] the title of bug report[/-][+]CPU Performance issues in V2[/+]NicholasMartens commentedon May 14, 2026
I noticed the power usage kept going up and up, so I decided to remove some work spaces and to restart my computer, and now it seems to be better:
So maybe this was a memory leak or something, I don't know.
I also noticed that it updated to a later version, I'm currently on: v1.9.1
Kitenite commentedon Jul 11, 2026
Closing as a duplicate of #5602, the consolidated tracker for stale
performanceissues from the July 2026 stale-issue sweep (no activity since 2026-06-01). If this still applies on a current build, comment on #5602 and we'll reopen.