People call it a free RDP server, but what you actually receive is a slice of a Microsoft Azure build machine, shared with the world's developers. Its published specs explain both things users report: why it feels surprisingly snappy when it first opens, and why heavy software starts to struggle.
GitHub publishes the hardware of its standard runners in its own documentation, so none of this needs to be guessed. One detail matters before all others: the repository's visibility changes the machine you get. RDP repositories are normally public — and public repositories get the stronger configuration.
👆 Specs from GitHub's own runner documentation for windows-latest. GitHub refreshes its fleet over time, so treat these as the current standard, not a promise.
For a public repository, windows-latest runs on 4 vCPU with 16 GB RAM. For a private repository, the same label runs on 2 vCPU with 8 GB RAM, and the minutes draw from your plan's allowance — 2,000 a month on the Free plan, with Windows minutes billed at twice the Linux rate. This is the entire reason RDP repositories are public: double the machine, no minute meter. The trade is the one covered in the safety guide — a public repository's files and logs are readable by everyone.
The figure that surprises people is disk. GitHub documents about 14 GB of free SSD space for your use on a standard runner — Windows itself and the large pack of preinstalled developer tools (Visual Studio and friends) already occupy the rest. Light software is fine; a few big installs, a game, or a large dataset will hit the ceiling mid-session, and hitting it usually fails the step or the job. Check free space early in the session, not after your download is at 90 percent.
Standard runners have no graphics card. The RDP desktop itself is rendered in software, which is fine for windows and forms, but gaming, 3D work and video editing are effectively off the table. The CPU is a shared Azure virtual CPU: quick in short bursts — opening apps, browsing, light tools — but it is not a dedicated processor, and sustained heavy loads share the host with other customers. This is a build machine's profile, and it shows.
The runner's own connection is data-centre grade, but your experience is the whole path: runner, ngrok's free tunnel, then your internet. Two honest consequences. First, the free tunnel is the narrowest part of the pipe, so big downloads to your PC are slow even though the runner itself is fast. Second, runners are usually located in US Azure regions, so from India every click makes a long round trip — expect visible lag in the desktop. Fine for settings, forms and tests; tiring for anything continuous.
Almost everyone reports the same first impression: 'this free thing is faster than my laptop.' That is real, and the reasons are boring — a completely clean Windows with nothing installed, an SSD, 16 GB of RAM and no antivirus suite or years of startup junk. The feeling fades as you install software within the session, because the fixed ceiling underneath never changed. Enjoy the clean start; do not mistake it for a powerful machine.
Watch the video and compare the live machine with these specs — then you will know exactly what the method can and cannot carry.
▶️ Watch Full VideoThe windows-latest label points to a Windows Server image — recent labels are built on Windows Server 2022 and 2025, and GitHub migrates the label over time. It is a server edition, not Windows 11, though for most desktop-style tasks it looks and behaves like ordinary Windows.
No. Larger runners with more CPU and RAM are a paid GitHub feature tied to billing accounts. The free route gets the standard runner sizes documented on this page, and no workflow setting changes that.
Most likely the two runs used different repository visibility — public gets 4 vCPU and 16 GB, private gets 2 vCPU and 8 GB. GitHub also updates its runner hardware over time, so two runs months apart can differ slightly.