Why BloomGauge rarely switches models (and when it does)
Updated
Short answer: a switch costs a few minutes of work, and most busy spells end before that is earned back. So the Manager runs Recommended, the model expected to pay most on your Mac, and moves only when another one clearly pays more and keeps that lead. Or it runs one model you pick and keeps it earning. The numbers below come from real pay on BloomGauge Macs plus public Darkbloom network data, so treat them as a first read, not a law.
“That model is busy right now, why not switch?”
Because by the time a switch pays off, the moment has usually passed. Busy stretches on a model typically last 20 to 30 minutes, and once one is 30 minutes in, only 17–50% carry on for another hour. In a September replay on our test Mac, 38 of 43 spike trips paid less than staying put, and even perfect hindsight (the best model every hour) added only about 6%.
How Recommended picks
- Every model gets one expected pay figure for your Mac, with a tag saying where it comes from: your own pay, then what Macs like yours were really paid (never counting your own Mac, and including their last 60 minutes), then list price, then a Rough network average.
- To move, another model must beat the current one by 25% after the cost of switching, then stay ahead for 15 minutes.
- At most two moves an hour, never straight back to the model it just left, and never on a Rough estimate.
- After every move it checks for 5 minutes that the new model is getting paid work. A model that fails to load is held back for a day, then two, then four.
Trying a model nobody has run
A model priced only from list price or the network average rarely wins, so no Mac runs it, so its real pay never arrives. To break that loop, while it follows Recommended the Manager may run such a model for an hour, at most once a day, when its estimate is at least 20% above what you run now. No paid job in the first 20 minutes and it goes straight back. After the hour it keeps the model only if your own pay keeps it the recommended one. Turn this off with Let the Manager try new models, in Control → Now → Settings.
Switching has hidden costs
- Loading a new model takes the Mac out of the running for paid work: about a minute until Darkbloom is ready again and about two until the first paid job, on our test Mac.
- The first paid job can be slow to arrive: 10 of 61 Gemma restarts on our test Mac (16%) waited more than 10 minutes for one.
- Those costs are small on their own. They add up when a spike ends before the new model has earned them back.
An independent test: one model beats two
An independent study published by an Eigen Labs account tested part of this directly, on one M4 Max (64 GB), about two hours per setup. In its pre-registered A/B test, Gemma alone earned about twice as much per hour as Gemma with a second model loaded beside it, even when that second model got almost no requests. One Mac, so treat it as a strong hint rather than a rule. Source: Darkbloom provider economics (Research Commons), CC BY 4.0, github.com/brandon-eigenlabs/darkbloom-provider-commons.
What matters more: keeping a model earning
Most lost earnings come from a Mac that stops working, not from the wrong model. The Manager keeps the model loaded and recovers stalls: a test request first, then a restart, within limits. In one public Darkbloom issue, 43 of 44 affected Macs recovered within 1–3 minutes of a restart. Keeping a model loaded matters too: in public network snapshots, about 30% of online Macs had no model loaded, which earns no base reward.
If you want to chase demand anyway
You can. Pick a model under the Manager and it holds it, or choose Manual on Control → Now, tick the models and press Confirm. Custom rules in Control → Manager Switching Rules switch on load you set, and replay last week first so you can see whether a rule would have paid. So far the data says such trips more often lose than win.
How BloomGauge helps
BloomGauge’s Manager runs Recommended, the model expected to pay most on your Mac, or one model you pick. Recommended moves only when another model beats the current one by 25% after switch costs for 15 minutes, at most twice an hour and never on a Rough estimate, and at most once a day tries a model with no real pay yet for an hour.
Questions
Why doesn’t BloomGauge switch to whichever Darkbloom model is busy right now?
Busy stretches usually end within 20–30 minutes, before a switch pays back its cost. In replays, 38 of 43 spike trips paid less than staying put, and even perfect hindsight added only about 6% on our test Mac.
Does BloomGauge ever switch Darkbloom models?
Yes, under the Manager on Recommended: when another model is expected to pay at least 25% more after switch costs for 15 minutes, at most twice an hour. It also tries an unproven model for an hour at most once a day, which you can turn off.
Can I switch models myself?
Yes. On Control → Now, pick a model under the Manager and it keeps that model, or choose Manual, tick one or several models and press Confirm. Nothing changes before Confirm.
How much does switching a Darkbloom model cost?
About a minute until Darkbloom is ready and about two until the first paid job on our test Mac, and more when the first paid job is slow to arrive.
Related
- Which Darkbloom model should I run on my Mac?
- BloomGauge Guardian: keeps your Mac earning
- Why is my Mac online but not getting Darkbloom jobs?
- BloomGauge vs. darkbloom-manager
- Can you run several Darkbloom models at once?
Updated 2026-10-30. Still stuck? Ask in #bloomgauge on the Darkbloom Slack (join it here, free) or contact us. BloomGauge is independent and not affiliated with Darkbloom.