RR4 → RR64 → Sabayon: the complete release history

A Gentoo desktop in 2005 cost you a weekend of compiling. This DVD claimed to cost you about five minutes. Below is every release it announced, in order, with what changed in each — plus how anyone actually moved a multi-gigabyte image across a 2006 domestic connection, and where the surviving images legitimately live in 2026.

The claim that started it was a wall-clock number, not a feature list

2004·08·30The project's public life begins not on this domain but in a thread on the Gentoo Forums, opened by the user lxnay — Fabio Erculiani, of Arco in Trentino — at 09:26 on 30 August 2004. The thread ran to 486 posts across 20 pages and is still online. Its opening post is a standing product page rather than a message: the forum records it as edited 134 times, most recently on 5 November 2009, and the copy sitting in it today is the same copy the download page on this domain carried. That matters for how you read it, and I come back to it below. It also means the thread's opening date is the date the slot was created; the second post in it is dated 20 March 2005, almost seven months later, which is where the traffic actually starts.

The pitch is a wall-clock figure: five minutes, from a machine you have not touched to a working Gentoo GNU/Linux desktop. Set that against the same job done the ordinary way — fetch a stage tarball, set your flags, and let a compiler run for hours or days — and you have the entire commercial idea of the thing. It was not only a live disc, either. It installed onto a real hard drive, and what landed there was an ordinary Gentoo system with the furniture already in place: two full desktop environments, Xfce and Fluxbox besides, KOffice and OpenOffice, amaroK and Kaffeine for media, FreeNX for remote sessions.

If you have never used Gentoo, three sentences will do. Gentoo Gentoo · shipping · verified 2026-08-04 ships a tree of build recipes called ebuilds rather than a set of pre-built packages, and its package manager, Portage, compiles what you ask for on your own machine using compiler flags and feature switches (USE flags) that you choose. The payoff is a system built exactly to your configuration. The price, in 2005, on a single-core Athlon, was that a desktop with a browser and an office suite could take a very long weekend — and if you got a flag wrong, another one.

RR4 sold the payoff without the price. That is the whole idea, and every argument about this project for the next fifteen years is a variation on whether that trade was worth making.

You probably arrived from a link somebody wrote in 2006

Several long-lived Linux sites and forums still point at one path on this domain, usually with the anchor text "RR4/64 Linux". That path was the project-and-download page: a single CMS article, written 6 October 2005 and last updated 30 December 2005, carrying the pitch, screenshot strips, and the download matrix that is the subject of most of this piece.

That page is gone, and I am not rebuilding it. The domain lapsed after the project moved, passed through a parking page, a small design blog, a content farm, a bare directory listing, a Chinese-language gambling doorway and finally a casino affiliate site before I registered it in 2026. What is here now is a publication that writes distribution histories. The link you followed was about acquiring and verifying a Gentoo-based live distro, so this page is about acquiring and verifying a Gentoo-based live distro. That is the only honest thing to put at the end of a twenty-year-old link.

The Internet Archive's capture of the original page is one click away and I would rather you read it there than read my paraphrase of it. Nothing on this site reproduces its text.

The ladder: every release this domain announced

Dates below are announcement dates, taken from the publication dates on the site's own RSS feeds, which the Archive preserved intact — most of the 2006 article bodies sit behind a CMS login wall in the captures and are not recoverable. An announcement date is not a build date, and where the two are known to differ I say so. Rows with no date are ones where a release definitely happened, because an artifact or a headline survives, but no dated record does.

Two things fall out of reading that table as a table rather than as a list. The first is the cadence, and how it changes. Across 2005 the gaps are months: April, June, October, late November. Across the first half of 2006 they are weeks — and then, between 1 and 5 July, three complete releases in five days, from one person, each one a rebuilt image rather than a patch to a package. That is what a project shipping faster than one person can test looks like from the outside. The second is that the version numbers are not one sequence but two, pretending to be one, which is the next section.

Why 32-bit and 64-bit were two separate products

RR4 and RR64 RR4 / RR64 · renamed · verified 2026-08-04 were not a build flag apart. In 2005 the x86-64 desktop was a genuinely different platform with a different problem list: a 32-bit userland could not link against 64-bit libraries, the proprietary graphics drivers landed on the two architectures months apart, and a great deal of desktop software of the period still assumed pointer size 4. A Gentoo profile for one was not a Gentoo profile for the other, and a distribution shipping pre-built binaries had to build, test, host and support both.

So the numbering came apart, and it came apart in both directions, which is the part that gets forgotten. The 64-bit line opened ahead: RR64 2.50 in June 2005, while the 32-bit disc was still on 2.40. Four months later 32-bit had overtaken it at 2.60, and by that November it was at 2.65.1 while 64-bit had jumped to the 3.0 alphas. By mid-2006 the same nominal 3.0RC1 meant two different images with two different bug lists, released a fortnight apart, and the 64-bit one went first. The rename is usually told as a branding story. Structurally it is a merge: from 3.0RC2 onward one name covered one release, built for two architectures on the same day, which halves the announcement surface and makes "what version are you running" a question with one answer.

The rename, mid-2006, and the move off this address

2006·06·24The new name appears on this domain between the Archive's captures of 12 and 29 June 2006, alongside a news item about a website migration dated 24 June. The 29 June capture already carries a new theme, a footer crediting Sabayon Linux, a reader poll about that theme, and a sidebar that answers the obvious question by quoting an encyclopedia entry on the dessert. The name is the French and English spelling of zabaglione, the Italian custard of egg yolk, sugar and sweet wine. The menu, meanwhile, still said "Get RR4/RR64 Linux" — a rename in the masthead some weeks before it reached the navigation.

2006·08·16The name became a release on 16 August 2006, when 3.0RC2 shipped for both architectures. That is the date I would use if forced to pick one, because it is the first date on which a thing you could download was called Sabayon.

2006·08·30The last capture of the working project site at this address. By November 2006 the front page was a 535-byte document whose only job was a JavaScript bounce to the project's new home. That bounce ran until the summer of 2007, with one interruption: from 16 April to 16 May 2007 the address showed a registrar's "domain reserved" parking page instead, which is the visible edge of a registration lapse. Server-side 301 and 302 responses then replaced the JavaScript, and are visible in archived response headers from 28 September 2007 through 16 January 2010. By December 2010 the address answered HTTP 401 and the project had gone.

How anyone actually moved a DVD image across a 2006 connection

This is the part of the story that has genuinely been lost, because every technique in it was replaced by something better and nobody wrote down the old one.

Distribution was a torrent plus a handful of donated mirrors. The download page listed a BitTorrent file first and university and volunteer FTP servers second — unito.it, linux.it, a volunteer's FTP host. There was no CDN, no object store, and no budget. A one-person project's bandwidth problem was solved by asking, and by the seeders.

The images were shipped compressed. Several of the checksum filenames the archive indexes end .iso.bz2 rather than .iso: what you fetched was a bzip2 stream, which you decompressed to get the image. On a DVD that is largely already-compressed squashfs this buys less than you would hope, which is presumably why the practice did not last. Of the eleven checksum filenames the archive indexes, exactly three name a .iso.bz2: the 32-bit 3.0b0, its XGL edition, and 3.0RC1 — February to June 2006. Everything before and after names a plain .iso.

Upgrades shipped as binary patches. This is the genuinely dead technique. Between point releases the project published xdelta patch files — a binary delta between the old ISO and the new one — plus a Windows helper program, because the person most likely to be downloading a Linux live DVD in 2006 was sitting at a Windows machine. You kept your old image, fetched a patch a fraction of its size, applied it, and checked the result against the new MD5. Going from 2.60 to 2.65.1 meant applying a chain of them in order. On a 2006 domestic ADSL line, re-downloading a DVD for a point release was not a thing a reasonable person did.

Verification was one MD5 file per release, hosted next to the image. Six of those files still come back from the archive intact. Their contents are records rather than prose, so they are reproduced here exactly:

Those six are every 2005–2006 checksum that still comes back intact, out of the eleven .md5 paths the archive's index of this domain knows about. The other five — RR64 3.0a0, the 3.0b0 XGL edition, an uncompressed RC1, and both 3.0RC2 miniEdition respins — exist in the index as filenames only: fetching them returns a redirect captured years after the project left, not the file. The xdelta patch files are not in the archive at all — only the page that offered them and the news items announcing them — so if you have one on a disc somewhere, you have something the Internet does not.

What replaced all of that, and what a checksum still does not prove

The 2026 equivalent is worth stating precisely, because "check the checksum" is advice people repeat without noticing what it does and does not buy. Gentoo's current installation media are published alongside a .sha256 file, a .DIGESTS file, a .CONTENTS.gz manifest and — the part that matters — a detached OpenPGP signature:

An MD5 file sitting on the same server as the image detects corruption in transit. It does not detect substitution, because anyone who can replace the image can replace the file next to it, and MD5 has been practically collision-vulnerable since 2004 anyway — contemporaneously with the DVD it was protecting. A detached signature moves the trust to a key you can obtain separately, which is the whole difference. Gentoo took the same reasoning into its binary packages: since the news item of 3 May 2026, Portage verifies binary package signatures by default, and an unsigned third-party binhost now fails rather than warns. That argument, and the 89 GB of binaries Gentoo now ships, are the subject of a separate piece on this site; here it is only the closing bracket on a story that opens with a lone .md5 file.

What the miniEdition actually cut, checked against its own manifest

Each release published a plain-text list of every package on the image. The manifest for the 3.0RC1 miniEdition survives, and reading it settles several things that the announcements only assert.

Three findings, none of which I have seen stated anywhere else.

The release notes were accurate. Kernel 2.6.16 with Reiser4 patches, glibc 2.4, KDE 3.5.3, XGL, Compiz, the nVidia 8762 driver, Tor and Privoxy — every headline claim from the RC1 announcement has a matching package with a matching version on the disc. The one drift is the ATI driver, which the June announcement gave as 8.25.18 and which the July manifest shows at 8.26.18, consistent with a news item about new ATI drivers dated 28 June 2006. A project rebuilding its image every few weeks moved a driver version in under two weeks.

72 packages under x11-drivers is the real hardware story of 2006. Input drivers for graphics tablets and touchscreens that no longer have names — acecad, aiptek, calcomp, citron — shipped by default, because X in 2006 could not be relied upon to work out what was plugged in. That number, more than any retrospective, is what "hardware detection" cost.

The miniEdition was a KDE-only disc. To fit a DVD onto a CD, the cut was not incremental: there is no OpenOffice, no Xfce, no Enlightenment, and none of Ardour, Rosegarden, Hydrogen or Jamin from the audio-production stack the DVD advertised — though JACK itself survived the cut. GNOME is the interesting case: thirteen gnome-base library packages are still on the disc, because KDE and the GTK applications needed them, but there is no gnome-session, so there is nothing to log into. What stayed was KDE with 15 kde-base packages, KOffice, amaroK, Kaffeine, Kuroo as a graphical front end for Portage, FreeNX, Firefox as a binary package rather than a source build, and win32codecs. That last one is the whole editorial position of the project in a single package name: a distribution that shipped a proprietary codec pack and two proprietary graphics drivers so that video and 3D worked when you booted it. Shipping the binary blobs on a free-software live DVD in 2006 was the correct engineering call and it cost the project goodwill in quarters it never fully won back.

Where the surviving images are in 2026, and what they are good for

Nothing from the RR4 or RR64 era is mirrored anywhere I can find; the six checksums above describe files that appear not to exist any more. What does still exist is the end of the line — Sabayon Sabayon · ended · verified 2026-08-04 19.03 — on two live public mirrors and two archive projects.

Those three checksums are published independently by ibiblio and by the University of Maryland mirror and they agree. Two mirrors agreeing is not a signature, and I have found no OpenPGP signature for any Sabayon image, so this is the strongest verification available: it is meaningfully better than one file on one host, and it is meaningfully worse than a key you can check. The KDE image is 2,621,440,000 bytes with a last-modified date of 27 March 2019 on both mirrors.

The images are also on ArchiveOS, which redistributes them through SourceForge, and on LQ ISO — the latter over plain HTTP only, as of the date on this page. Both are third-party redistributions: check what you get against the mirror checksums above rather than trusting the download page.

2020·07·09The mirrors are also a record of exactly when the lights went out, which is why they are worth reading and not just downloading. The daily/ directory still holds a full set of rolling builds; the newest of them was written on 8 July 2020. The TIMESTAMP file at the top of the ibiblio tree reads 1594256402 — 01:00 UTC on 9 July 2020 — and has not moved since. The UMD copy stopped a month later, at 20 August 2020. There is a small discrepancy worth recording while I am here: the RELEASE_DATE_19.03 file states 22 March 2019, while every ISO in the directory has a modification time of 27 March 2019, which is also the date DistroWatch carries. The release was declared five days before the images were published.

Now the blunt part. A 19.03 image is a museum piece, not an operating system. It has had no kernel, OpenSSL, browser or firmware update since March 2019; its package repositories are gone, so there is nothing to update it from even if you wanted to; and Entropy's servers no longer answer, which means a fresh install cannot complete the first thing it will try to do. It is a legitimate and interesting thing to boot in a virtual machine and look at. It is not a thing to put on a network. Booting one on 2026 firmware has its own list of obstacles — UEFI without a CSM, a kernel that cannot see NVMe, drivers for GPUs that no longer exist — and that list is the subject of a separate piece on this site rather than a paragraph here. If what you want is a system you can actually run and update, the honest routes out are covered here and the surviving candidates are compared here.

This site does not host, mirror or re-serve any ISO

Not now and not later. Every download link on this page points at somebody else's server, chosen because it was already publishing the file before I wrote about it. This address served images in 2006 and it will not serve them again: re-hosting a twenty-year-old distribution image from the domain it originally shipped from is the sort of thing that looks like continuity and is actually impersonation. The About page states the same thing in the same words.

Appendix: the trademark line, and why a distribution built on someone else's work needs one

The original download page carried a disclaimer, and the later Sabayon-era footer kept it: RR4 and RR64 were not part of, or managed by, the Gentoo Foundation, and Gentoo is the Foundation's trademark. The Sabayon version added that the DVD included proprietary software which could be removed after installation.

It is easy to read that as legal boilerplate. It is not. A distribution that ships another project's package manager, another project's build tree and another project's package names is going to be described by its users, accurately, as "Gentoo" — and every support request that produces lands in the upstream project's forums and bug tracker. The disclaimer is not there to prevent a lawsuit that was never coming. It is there so that the volunteers upstream can tell, in one line, which project's problem a given bug report actually is. Every derivative distribution owes its base that sentence, and most of them still write it badly or not at all.

What this piece deliberately does not cover

Several things are held for their own pieces rather than summarised badly here:

• What the 4,117-post forum on this domain was actually reporting — including the 18 June 2006 release announcement gone through line by line.
• The shutdown timeline, from 19.03 to the day the repositories stopped answering, dated from mirrors, DNS and the project's own posts.
• How to boot a 2006 live DVD on 2026 hardware — four blockers and one documented QEMU invocation.
• What became of the binary-packages idea Entropy was built for, now that Gentoo itself ships 89 GB of them.
• Who, if anyone, inherited this: MocaccinoOS audited as a successor, matrixOS read from its own repository, and the five candidates compared.

Every distribution named on this page has a dated row in The Register, which is updated whenever a status changes rather than whenever something is published.