Today’s Habr digest brought another retrospective on “old-school technologies that never took off,” including a chapter on the IBM Deskstar 75GXP, the notoriously ill-fated hard drive that forum old-timers christened the Deathstar. The digest retells the familiar legend: glass platters, flaking magnetic coating, the “click of death,” lawsuits. But the legend is more interesting than its usual retelling, and here’s why: the 75GXP is not a story about a defective product. It is the story of how a company that had invented the hard drive itself half a century earlier managed, in 2 years, to burn through a reputation built over decades — and then leave the market it had created. And there is a final twist worthy of a good detective novel: over 15 years, the successors to this line scrubbed themselves so clean that, by Backblaze statistics, the descendants of the “Deathstar” became some of the industry’s most reliable drives.
To grasp the scale of the irony, you need to rewind to the beginning. In 1956, IBM released the RAMAC 305 — the world’s first hard-drive system. Five megabytes on fifty 24-inch platters, weighing around a ton, rented at a rate that worked out to roughly 10 kilobytes per dollar at the prices of the day. This was not yet a product — it was a ton of iron whose purpose existed only inside the machine room. The patent anthology of the magnetic disk is almost wall-to-wall IBM.
In 1973, the same IBM released the 3340 model — along with an architecture that would outlive its creator by half a century. Inside the project, it bore a “code name” based on a pair of removable 30-megabyte cartridges, and someone made a joke about a Winchester — the .30-30 caliber of the famous hunting rifle. The nickname stuck for good: to this day, in Russian, vinchester is simply a synonym for a hard drive, while the architecture of “flying” heads and hermetically sealed platter assemblies (Winchester technology) remained fundamental until the SSD era. The company that had introduced both the device itself, its defining name, and its core design held a third of the global disk market by the 1990s — roughly 35 percent.
In the summer of 2000, IBM brought out the Deskstar 75GXP line — a flagship 7200-rpm desktop series, with a 2 MB buffer, Ultra ATA/100 support, and capacities from 15 to 76.8 GB across 6 models. The DTLA-307xxx model numbering instantly became folk terminology: in Russian-language conferencing forums and among repair technicians, the drive was simply called the “woodpecker,” after the DTLA designation. A recording-density champion, the best performer in its class, it received deservedly ecstatic reviews. Reviewers kept it on their recommendation lists throughout 2000.
Its technical guts were ambitious on 2 fronts at once. First, IBM became the first company in the industry to put glass substrates into a mass-market 3.5-inch desktop drive instead of aluminum: glass is stiffer, thinner, more stable under heat, and allows higher recording density. The technology had been refined in premium Ultrastar SCSI machines, and was now brought down into the mass segment. Second, a new generation of GMR (giant magnetoresistance) heads read at record recording densities, where each bit was allotted a microscopic patch of surface. Both components, taken separately, were the future of the industry. The question of what happened when they were combined on an assembly line producing millions of units became the costliest in IBM’s history.
The first warning signs came in the spring of 2001, when owners would lose a second, third, and sometimes even fourth 75GXP in a row — no longer the background noise of the usual “bathtub curve,” but a pattern. The typical scenario looked sinister: at some point the drive would emit a distinctive loud clicking or scraping sound — the heads striking the surface — after which the hard drive would drop offline and never be detected again. The “click of death” became an internet meme long before the meme era.
The nature of the damage was even more suspicious than the noise. Teardowns of dead drives revealed a startling sight: the magnetic layer — that finest rusty “varnish” where data lives — was not merely scratched. It was peeling away from the glass substrate, crumbling into dust, and being smeared by the heads into powder. At the University of Florida in April 2003, one unit failed so violently that its 10 heads ground the magnetic coating almost completely away: you could look through the platters as though they were glass — because they were glass. Photos of those platters became the calling card of the disaster and still migrate from one illustration to another in every story about the 75GXP.
On 31 August 2001, StorageReview opened a thread titled “Dr. Evil asks: GXP problems” — it gathered more than 1300 posts describing failures and survived for years, becoming the legendary “thread that won’t die.” The forums, and then the reviewers following them, drew the simple conclusion: something was systemically wrong. And the miserable owners entered a second Dantean circle: warranty replacement got them... a new 75GXP, which would die within a couple of weeks. Hosting company Pair.com replaced around 70 drives across its servers in October 2001; users wrote to support saying “drives are crap,” and reviewers one after another withdrew their recommendations for the line: Scott Wasson of The Tech Report — the very man who had recommended the 75GXP in that very publication — apologized and wrote, in all caps: don't buy.
On 16 October 2001, user Michael Granito Jr. filed a class-action suit: both 75GXPs he had bought had died within a short period. The wording of the complaint left behind a miniature panegyric: “the sudden onset of a loud clicking or scraping noise,” after which the Deskstar “ceases operation and crashes,” irreversibly destroying data. In 2003, the court certified a class defined as everyone in the United States who had bought the 6 DTLA-307xxx models between 15 March 2000 and 4 March 2005.
The best part came when law firm Sheller Ludwig & Badey obtained IBM’s internal emails and documents as part of the case — a California judge unsealed them for the proceedings, and Maximum PC analyzed them in its Watchdog column in the February 2004 issue. The internal picture had the texture of a dystopia: internal reports recorded failure rates as high as 6.5 percent among major customers — for one of them, roughly 10 times higher than a comparable Maxtor drive. Internal documents described a product with that level of defects as something a major corporate customer had rejected — and that same batch apparently went to retail. And the communications strategy is summed up in the recollections of an IBM manager in a single sentence that deserves to be framed in every organization considering the “keep quiet and wait it out” approach: “We have woven a story based upon half truths and misinformation that has lead IBM to an almost untenable position.” IBM categorically denied a systemic defect — while quietly releasing Rev 2 with recessed heads, new firmware, production improvements to fight contamination, and... “wear leveling,” so that the heads would not remain on one track long enough to scrape it down to transparency.
The case was settled — without an admission of liability — in 2005. The compensation arithmetic is valuable as an engineering artifact of the era: everyone whose drive had died received 100 dollars; those whose drives remained alive got 25 blank CD-Rs in jewel cases or a 15 percent coupon for IBM.com. The lawyers consumed up to $9.95 million. Payments were distributed on 2 April 2008.
Few remember the scandal’s postscript, and it is magnificent: by 2002, the next line, the 120GXP, had acquired “recommended monthly power-on hours” in IBM’s specifications — 333, about 11 hours a day. The forums exploded: a drive that officially could not be turned on for more than 11 hours a day? Distributors and the public decided it was a quiet way to void the warranty for anyone running a server 24/7. The uproar became so loud that IBM spokesperson Kim Nguyen was forced to explain to the press that the warranty applied regardless, while the line itself was removed from the specifications — “It has caused a lot of confusion, so we are removing that data from our spec sheet.” The historical irony: a desktop drive officially put on office hours.
On 4 June 2002, IBM announced that its hard-drive business was going to Hitachi for 2.05 billion dollars, with a 70/30 split in favor of the Japanese company. For a corporation that had invented the product itself, leaving its own market was not diversification — it was evacuation. Naturally, the deal was not presented officially as a flight: the dot-com bubble had burst, the PC market had contracted, hard-drive margins were evaporating, and new CEO Palmisano wanted profits, not hardware. And all of that was true. But the NYT put the essence more honestly in its headline: “Deal to Take I.B.M. Out of Disk-Drive Business.” Internal documents from 2004 added a third circumstance everyone had suspected: GXP reliability.
Then came what makes this story complete rather than a tragic curiosity. Hitachi assembled the division under the name Hitachi Global Storage Technologies — HGST. Quietly, methodically, internally, it did the work IBM had failed to do in public: the roots of the defects were found, the firmware rewritten, the manufacturing reshaped. The Deskstar line continued to exist — now without scandals; the “Deathstar” in its new generations (7K250, 7K400, 7K1000) stopped crashing its heads. In 2012, WD bought HGST; the HGST brand was phased out by 2020, but the technological DNA of that era — glass platters, GMR heads, helium-filled sealed Ultrastar enclosures — became the foundation of the industry.
And now for the final shot. Backblaze — the cloud-backup giant that since 2013 has published the industry’s most honest failure statistics across hundreds of thousands of drives — delivered much the same conclusion year after year: HGST drives (the former IBM factory carried through Hitachi) were consistently among the reliability leaders. In its 2023 annual report, the analysts wrote outright that the 4-terabyte HGST models in their fleet looked as though failure did not depend on age at all. The 2025 report delivered a second irony: WDC (into which that very HGST DNA had flowed) was the year’s most reliable vendor, with an AFR of 0.86 percent, while the named honor roll was led by a 4TB HGST model whose more than 20,000 drives collectively suffered 1 failure over the quarter. The direct descendants of the “Deathstar” — a state within a state, best in class.
What does this story teach? Not a moral, but mechanics. The 75GXP product failure was probably a combination of factors: adhesion of the magnetic layer to the glass on new platters, manufacturing contamination, micron-scale GMR head clearances, firmware bugs. The sources still offer no definitive physical verdict: half a century later, the investigation is officially still open. The marketing catastrophe is easier to explain: IBM chose a strategy of denial instead of a strategy of recall, and that turned a technical defect into a reputational tsunami. The right response — acknowledge it, recall it, fix it, report back — would have cost hundreds of millions. The response of “we don’t see a problem” cost a 2-billion-dollar exit from a business the company had spent 50 years building. The calculation that silence was cheaper was the costliest illusion in this story: silence proved twice as expensive as honesty.
And one last thing. Fifty years ago, IBM built an industry through its own patent department and its own engineers — from RAMAC to Winchester. One half-finished production season, one defective adhesion layer on glass platters — and 50 years of leadership turned into a 100-dollar voucher and 25 blank CD-Rs. Reputation is a hard drive: it takes years to write, one click of death to erase. The heirs understood something: today, their drives’ failure statistics are public, open, and — let us not be afraid of the word — exemplary. The Deathstar is dead. Long live its heirs.
The “Curiosity” column. The idea came from the Habr digest of the morning of 2026-11-06; all dates, names, sums, and figures have been checked against the primary sources listed above. Disputed points (the precise physical root of the 75GXP defect; the degree of involvement of specific manufacturing factors) are explicitly marked in the text as unresolved to this day.