Loading today’s stories
Product1 distinct publisher2 min readUpdated
The engine changes on restart for anyone who had already picked Docker VMM, the new backend wants 4 GB, and the docs put that instruction on the other path.
The Product Desk · Product desk
Compiled by The Product DeskSomething wrong?How this is made
The docs do not describe Docker VMM as anyone's default. WSL 2 is the one backend on the page labelled a default, and it is labelled that for Windows [17]. On Mac, the selection is whatever a user or a provisioning profile left in Settings > General > Virtual Machine Manager [9]. So 4.86 does not flip a fleet; it substitutes an engine underneath choices already made [2], [5].
The substitution is where the operational detail sits. The 4 GB instruction appears in the manual switching procedure, alongside the words "before switching" [8]. The upgrade path described further down the same page is automatic: the setting survives and the engine changes on restart [11]. Docker's page does not describe a memory check on that path [1]. Whether Docker Desktop enforces the floor when it comes back up, or simply starts on a VM that is under it, the documentation will not tell you, and the machines most likely to be under it belong to people who set a small VM once and never revisited it.
Then bind mounts. Docker VMM does not support bind mount auto-shares, so a host directory that is not listed under Settings > Resources > File sharing produces a "file is not shared from the host" error [12]. The repair belongs to whoever owns the compose files and the onboarding script, and it is the same repair on every machine that switched.
On Windows the comparison worth making is about privilege rather than performance. Hyper-V runs the Linux VM fully isolated, but only in all-users installation mode and only with administrator rights [18]. WSL 2 requires neither and installs per-user [17]. Docker's page assigns Docker VMM no installation mode and no privilege requirement at all [3], while describing it as a real VM boundary between the container environment and the host [7]. If that silence means what it looks like, an isolation boundary that used to cost admin rights on the box is available without them, which matters more to a team standardizing locked-down laptops than start-up time does [5].
The Mac option list, meanwhile, is getting shorter. HyperKit is deprecated, with Docker recommending a move to Apple's Virtualization framework [16], which the same page calls stable and well established [15]. That leaves two supported backends worth testing on Mac, Docker's hypervisor and Apple's, and the file I/O claim [4] is the one a team can measure for itself in an afternoon against its own edit-compile-test loop.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Docker Desktop supports multiple Virtual Machine Managers to power the Linux VM that runs containers, with available options depending on the platform.
From Docker Desktop 4.86, Docker VMM uses Docker's own hypervisor, replacing libkrun, which backed it in versions 4.35 through 4.85 for Mac users.
Docker VMM returns idle memory to the host when containers aren't active, so Docker Desktop doesn't hold RAM it isn't using.
On Windows, Docker VMM provides a stable alternative to WSL 2 with a real VM boundary between the container environment and the host.
Docker VMM requires a minimum of 4 GB of memory allocated to the Docker Linux VM; the docs instruct users to increase memory in Settings > Resources before switching.
To select Docker VMM: go to Settings > General > Virtual Machine Manager, select Docker VMM, then Apply & restart.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
First-party documentation, authoritative on configuration but unverified on performance
Everything rests on one source: Docker's own VMM documentation page. That is authoritative for what the product does - version-to-engine mapping, the 4 GB minimum, the switching procedure, the upgrade behaviour, and the named limitations are all directly stated and internally consistent. It is weak evidence for the benefit claims: file I/O and start-up improvements carry no figures or baseline, and the monitoring/governance assertion names no interface. There is no independent test, no third-party report, and no contradicting source in the cluster.
No usage data supplied
The cluster contains a documentation page describing availability of the new backend in 4.86 and an automatic switch for users who already selected Docker VMM, but no deployment counts, install-base share, telemetry, user reports, or case studies. Availability is not adoption, and Docker VMM is not even labelled a default, so no adoption level can be measured from the supplied material.
Benefits asserted somewhat ahead of the evidence shown
Modestly overstated rather than inflated. The page claims lower file I/O latency, faster engine and container start-up, and monitoring/governance impossible with third-party backends, none of it quantified or tied to a named control surface. Against that, the same page volunteers real limitations - no Rosetta, no bind mount auto-shares, database failures on virtiofs - and pairs the switch with a hard 4 GB floor whose instruction sits only on the manual path, which keeps the gap small and mostly about unproven upside rather than concealed downside.
Vendor documenting its own replacement of a third-party component
The sole publisher is the vendor, and the story is Docker swapping an external dependency (libkrun) for its own hypervisor and arguing that owning the virtualization layer enables governance third parties cannot provide. That is a direct commercial and product-control interest in the framing. The incentive is partly offset by documentation-genre duties: the page states the 4 GB requirement, deprecates its own HyperKit option, praises Apple's framework as stable, and lists defects in its own new backend.
Confident on the mechanics, thin on effect and uptake
High confidence that the described mechanics are accurate: the version mapping, the automatic switch on restart, the 4 GB minimum, the selection steps, and the limitations come straight from the vendor's documentation and are unambiguous. Low confidence on impact - no measurement of the claimed performance gains, no adoption data, no independent corroboration, and a single publisher means no way to test the framing.
build
Docker writes its own hypervisor, and inherits every bug in it1 distinct publisher
product
Pinned toolchain images were the easy part; scoping the agent is the new build problem1 distinct publisher
build
The agent sandbox that works is a disposable VM, and Apple decides its shape for you1 distinct publisher
build
Firecracker won't run on a Mac, so Encore rebuilt the Linux image toolchain to match1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 23, 2026