Intel: The Empire That Bugs Built
We have profiled Broadcom — hostile. Realtek — indifferent. MediaTek — chaotic. Nuvoton — invisible. Winbond — foundational.
Today we profile a company that is all of these things at once.
The Supreme Leader runs BSD as well as Linux, so this empire is not being judged from only one operating system’s trench. Intel’s sins and virtues travel across both camps.
Intel.
Intel is the largest x86 processor manufacturer on Earth. They employ kernel developers. They dual-license their WiFi drivers for both Linux and FreeBSD. They are a founding member of the Linux Foundation. They contribute more code to the Linux kernel than almost any other company.
Intel also shipped a CPU that could not divide. They built an architecture so unwanted that the industry named it after the Titanic. They embedded a surveillance engine in every processor since 2008. They crippled their own compiler to sabotage AMD. They shipped speculative execution so aggressive that it leaked kernel memory to userspace, and when Linus Torvalds reviewed their fix, he called it “COMPLETE AND UTTER GARBAGE.”
No other company in computing history has contributed so much and destroyed so much at the same time. Intel is the empire that bugs built.
The FDIV Bug: The $475 Million Division Error
In June 1994, Intel discovered that the Pentium processor could not divide correctly.
Not approximately. Not in edge cases. The floating-point division unit (FDIV) had missing entries in a lookup table used by the SRT division algorithm. The computation 4195835 / 3145727 returned 1.333739068 instead of the correct 1.333820449. Wrong in the fourth significant digit. On a processor marketed for scientific computing.
Intel knew. Intel said nothing. They quietly fixed the bug in new production runs and continued shipping defective chips.
On October 30, 1994, Professor Thomas R. Nicely of Lynchburg College discovered the bug independently while running number theory computations. He reported it to Intel. Intel’s response: the bug would affect the average user “once every 27,000 years.”
IBM contradicted Intel, stating the error could occur as frequently as once every 24 days. IBM halted all Pentium sales.
David Letterman ran a “Top Ten Things More Fun Than Using a Pentium Computer” segment. The internet — such as it was in 1994 — erupted.
On December 20, 1994, under pressure that made the VMware pricing backlash look like a polite suggestion, Intel reversed course and announced a no-questions-asked replacement for all affected Pentium processors.
The cost: a pre-tax charge of $475 million. In 1994 dollars.
The lesson Intel learned was not “test your silicon.” The lesson Intel learned was “control the narrative.” They have applied this lesson ever since.
The Itanic: The Architecture Nobody Wanted
In 1994, Intel and HP partnered to build the future of computing. The project: IA-64, a clean 64-bit architecture based on HP’s EPIC (Explicitly Parallel Instruction Computing) research. No x86 legacy. No backward compatibility compromises. The compiler would do the hard work of instruction scheduling, not the hardware.
The original target: 1999.
The first Itanium processor shipped in June 2001 — two years late, with performance so disappointing that the industry coined a nickname within hours of the announcement: “Itanic.” The unsinkable architecture. Named after the unsinkable ship.
| Year | Event |
|---|---|
| 1994 | Intel and HP partner on IA-64 |
| 1999 | Original ship date (missed) |
| 2001 | Itanium ships, performance disappoints |
| 2002 | Itanium 2 ships, market has already moved on |
| 2003 | AMD ships Opteron with AMD64 — backward compatible, fast, cheap |
| 2004 | Intel adopts AMD’s x86-64 (rebranded as “EM64T”), conceding the war |
| 2011 | Oracle drops Itanium support, calls it “nearing end of life” |
| 2016 | Jury orders Oracle to pay HPE $3 billion for dropping support |
| 2017 | Last Itanium processor (Kittson) ships |
| 2021 | U.S. Supreme Court declines Oracle’s appeal — $3B judgment stands |
| 2023 | IA-64 support removed from Linux kernel 6.7 |
The Itanic’s fatal flaw was not technical — it was philosophical. IA-64 assumed compilers would get smarter. They did not get smart enough. Meanwhile, AMD took the x86 architecture Intel was trying to abandon, extended it to 64 bits, maintained full backward compatibility, and shipped it for less money.
Intel spent a decade and billions of dollars building the future. AMD spent a fraction of that extending the present. The present won.
By 2004, Intel was forced to license AMD’s x86-64 extensions — rebranded as “EM64T” and later “Intel 64” — for its own processors. The company that invented x86 had to adopt AMD’s version of 64-bit x86. This is the corporate equivalent of building a spaceship, watching your neighbor duct-tape wings to a car, and then buying the car.
Oracle’s abandonment of Itanium triggered one of the most expensive lawsuits in tech history. HP (later HPE) argued that Oracle was contractually obligated to continue supporting the platform. A jury agreed. $3 billion. The architecture that nobody wanted generated a lawsuit worth more than most architectures earn.
And as we discussed in the MINIX post — IA-64 had one property that Intel found unacceptable: it could not host the Management Engine. The architecture was too clean. Too different. The parasite could not attach to this host. Make of that what you will.
The Antitrust Record: Paying OEMs to Exclude AMD
While Itanium was failing in the market, Intel was ensuring AMD could not succeed either — through means that multiple courts found illegal.
Between 2002 and 2007, Intel paid OEMs conditional rebates — effectively kickbacks — for buying exclusively or near-exclusively from Intel:
| OEM | What Intel Paid |
|---|---|
| Dell | Up to $1 billion per year for remaining AMD-free (2002-2007) |
| HP | Payments for limiting AMD purchases |
| NEC | Payments for limiting AMD purchases |
| Lenovo | Payments for limiting AMD purchases |
| Media Saturn | Payments for stocking only Intel-based PCs (Europe’s largest electronics retailer) |
The result: Dell sold zero AMD-based systems for five years. Not because AMD’s chips were inferior — during the Athlon 64 era, AMD’s processors were objectively faster. Because Intel was paying Dell a billion dollars a year not to.
June 2005: AMD filed an antitrust lawsuit. May 2009: The European Commission fined Intel €1.06 billion (~$1.45 billion). November 2009: Intel settled with AMD for $1.25 billion. December 2009: The FTC filed its own complaint, calling Intel’s conduct illegal for “a decade.”
The FTC settlement required Intel to stop the rebate scheme and — this is the remarkable part — publicly disclose that its own compiler sabotaged AMD processors.
The Compiler That Sabotaged AMD:
Intel’s C/C++ compiler (ICC) contained a CPU dispatcher that checked the CPUID vendor string at runtime. If the string was GenuineIntel, the compiler used the optimized code path — SSE2, AVX, the fast version. If the string was AuthenticAMD — or anything else — the compiler fell back to unoptimized generic code, even when the AMD processor supported the exact same instruction set extensions.
This was not a bug. This was a feature. The dispatcher checked the brand, not the capability. An AMD processor with full AVX support would be forced to run code without AVX because the vendor string said AMD.
Third-party patches demonstrated that simply changing the vendor string check restored full performance on AMD. The performance gap was not in the silicon. It was in the compiler.
Intel’s compiler was used by benchmark organizations, scientific computing labs, and software vendors worldwide. Every benchmark compiled with ICC showed Intel ahead — not because Intel’s chips were faster, but because ICC was deliberately slow on everything else.
The FTC ordered Intel to disclose this. Intel added a footnote to their compiler documentation. The damage — a decade of misleading benchmarks — was already done.
Intel Management Engine: The Surveillance That Cannot Be Disabled
We covered the Management Engine extensively in the MINIX post. A brief summary for those who missed it: every Intel processor since 2008 contains a separate computer — a Quark x86 core running MINIX 3 — that has full access to your RAM, your network, and your storage. It runs at Ring -3. It cannot be disabled. Its firmware lives on the same Winbond SPI flash chip as your BIOS.
What we did not cover in detail were the vulnerabilities.
CVE-2017-5689 (INTEL-SA-00075) — the AMT authentication bypass. Intel’s Active Management Technology used a strncmp() function that compared the authentication response using the length of the attacker-supplied string rather than the expected password length. Sending an empty string as the response caused the comparison to succeed. Full administrative access. Remotely exploitable. On every provisioned AMT system.
The security community nicknamed it “Silent Bob is Silent.” Send nothing. Get everything.
INTEL-SA-00086 — eight CVEs disclosed in November 2017. Buffer overflows and privilege escalation across Intel ME, Server Platform Services, and Trusted Execution Engine. CVE-2017-5712 was exploitable over the network. Affected processors from 1st through 8th generation Core, Xeon E3, Xeon Scalable, Atom, Apollo Lake, and Celeron.
The subsystem that runs below your operating system, that you cannot audit, that you cannot disable — had buffer overflows. The warden had left the door unlocked.
The HAP Kill Switch:
In August 2017, researchers Mark Ermolov and Maxim Goryachy from Positive Technologies discovered an undocumented configuration bit in Intel ME. When set, ME initializes hardware and then shuts itself down.
The bit is called HAP — High Assurance Platform. It was created at the request of the United States National Security Agency.
The NSA gets ME-free Intel processors. You do not.
Intel confirmed the HAP bit exists. Intel stated it is not supported for consumer use. Intel did not explain why the NSA needs a kill switch for a subsystem that Intel claims is safe.
The Haifa Question
Intel is headquartered in Santa Clara, California. This is where the press releases come from. The actual engineering is a more distributed story.
Intel opened a development center in Haifa, Israel in 1974 — one year after the company itself was founded. It was, at the time, one of Intel’s first operations outside the United States. It has since become one of Intel’s most productive engineering sites in the world.
The Pentium M — the processor that saved Intel from the Pentium 4 disaster — was designed in Haifa. The Core architecture (2006), which ended AMD’s brief period of dominance and defined the next decade of x86 performance, was designed in Haifa. The teams that reversed Intel’s trajectory after the Netburst debacle were working in Israel, not California.
Intel is also, through its 2017 acquisition of Mobileye for $15.3 billion, one of the largest employers in the Israeli economy. Mobileye — autonomous driving vision systems — remains headquartered in Jerusalem. Intel employs approximately 12,000 people in Israel. The country is Intel’s second-largest R&D hub after the United States.
The processor in your laptop was probably conceived in a conference room in Haifa. The subsystem inside that processor that cannot be disabled was created at the request of an American intelligence agency. The HAP kill switch that disables it was built by engineers whose timezone does not match the one on Intel’s letterhead.
The Supreme Leader notes that blue is not a flag-compatible color for GPU purposes. Intel is not GPU-friendly. Intel is operating in a different timezone, a different color palette, and several rings below where anyone can audit it.
Spectre and Meltdown: The Architecture Was the Bug
On January 3, 2018, the world learned that Intel’s most celebrated engineering achievement — aggressive speculative execution — was also its most catastrophic security failure.
Meltdown (CVE-2017-5754): virtually every Intel CPU manufactured since 1995 could leak kernel memory to unprivileged userspace processes. AMD was not affected. This was Intel’s bug. Intel’s architecture. Intel’s twenty-three years of vulnerable silicon.
Spectre (CVE-2017-5753, CVE-2017-5715): affected Intel, AMD, and ARM — but Intel was hit hardest because Intel’s speculative execution was the most aggressive.
The fix — Kernel Page Table Isolation (KPTI) — imposed performance penalties:
| Workload | Performance Loss |
|---|---|
| Consumer (8th-gen Core) | 2-14% |
| Server I/O-heavy | Up to 30% |
| General server (Xeon) | 2-10% |
Every Intel customer in the world lost performance because Intel’s hardware was leaking secrets. The fix was a kernel patch. The cost was paid by users, not by Intel.
Linus Torvalds on Intel’s Fix:
On January 21, 2018, Linus Torvalds reviewed Intel’s proposed Spectre mitigation patches on the Linux kernel mailing list (LKML) and wrote:
“The whole IBRS_ALL feature to me very clearly says ‘Intel is not serious about this, we’ll have an ugly hack that will be so expensive that we don’t want to enable it by default, because that would look bad in benchmarks’.”
He called the patches “COMPLETE AND UTTER GARBAGE.”
Intel’s Indirect Branch Restricted Speculation (IBRS) was so slow that Intel’s own patches made it opt-in rather than default — hiding the performance cost. The kernel adopted Google’s retpoline technique instead — faster, simpler, and not designed to protect Intel’s benchmark scores.
The Open Source Paradox:
Here is where Intel confuses everyone.
The same company that embedded a surveillance engine in every chip, sabotaged AMD with its compiler, and shipped speculative execution that leaked kernel memory — is also the single largest corporate contributor to open-source drivers in the x86 ecosystem.
| Contribution | Details |
|---|---|
| iwlwifi | WiFi driver, dual-licensed GPL-2.0 OR BSD-3-Clause — the only major WiFi driver usable by both Linux and FreeBSD |
| i915 / Xe | Graphics kernel driver (MIT-licensed), Mesa userspace (Iris for OpenGL, ANV for Vulkan) |
| Linux Foundation | Founding member, Platinum tier ($500,000/year since 2007) |
| Clear Linux | Performance-optimized distro with PGO, LTO, AVX-512 (discontinued July 2025) |
| oneAPI | Open cross-architecture programming model (DPC++, oneMKL, oneDNN) |
Intel’s iwlwifi dual-licensing is remarkable. Every other WiFi vendor licenses GPL-only. Intel deliberately chose dual licensing so that FreeBSD can use the same driver source through linuxkpi. This is the reason FreeBSD has working Intel WiFi. Without dual licensing, FreeBSD would have no modern WiFi at all — the border wall would be absolute.
Intel’s i915 graphics driver is the reason Linux desktop graphics work at all on most laptops. Intel integrated graphics power the majority of laptops worldwide, and Intel provides a fully open-source driver stack — kernel driver plus Mesa userspace — that Just Works on every major distribution.
This is not altruism. This is strategy. Intel’s hardware sells better when the software ecosystem works. But the result is the same: Intel contributes more open-source driver code than Broadcom, Realtek, and MediaTek combined.
The Decline:
As of 2024, Intel is in free fall.
| Event | Date | Details |
|---|---|---|
| Worst single-day stock drop in 50 years | August 2, 2024 | Stock plunged 26% to $21.48 |
| Worst annual performance on record | 2024 | Stock fell 60.1% for the year |
| Layoffs | August 2024 | 15,000+ employees — 15% of workforce |
| Q2 2024 loss | Q2 2024 | $1.6 billion net loss |
| CEO forced out | December 1, 2024 | Pat Gelsinger given choice: retire or be removed |
| CHIPS Act funding | March 2024 | $8.5 billion direct + $11 billion loans from U.S. government |
| Clear Linux discontinued | July 2025 | Performance distro killed as cost-cutting measure |
Intel received $8.5 billion in taxpayer money and still could not stop the bleeding. The company that once defined the entire PC industry is now a patient on life support, kept alive by government subsidies and the inertia of an architecture that AMD executes better.
Pat Gelsinger’s “five nodes in four years” foundry plan was supposed to be the turnaround. The board decided it was not working fast enough. Gelsinger was given the choice to retire or be removed. He departed with $25 million. The 15,000 employees who were laid off departed with less.
The Verdict:
Intel is not hostile like Broadcom. Intel is not indifferent like Realtek. Intel is not chaotic like MediaTek. Intel is not invisible like Nuvoton.
Intel is contradictory.
They contribute the best open-source drivers in the industry and embed unauditable surveillance firmware in every chip. They fund the Linux Foundation and cripple competitors with compiler sabotage. They build the most advanced fabrication technology on Earth and ship processors with division errors. They pioneer speculative execution and leak your kernel memory to JavaScript in a browser tab.
| Vendor | Sin |
|---|---|
| Broadcom | Hostility |
| Realtek | Indifference |
| MediaTek | Velocity |
| Intel | Contradiction |
Every good thing Intel does is shadowed by something terrible. Every terrible thing Intel does is partially redeemed by something good. They are the only vendor on this list that actively makes the open-source ecosystem better while simultaneously making it less secure.
The Lesson:
Intel built an empire on x86 — an architecture they tried to abandon (IA-64), then had to adopt someone else’s extension of (AMD64), then filled with a surveillance engine (ME), then broke with speculative execution (Meltdown). The architecture survived everything Intel did to it and everything Intel did with it.
The empire is now crumbling. The stock has lost 60% of its value. The CEO is gone. Fifteen thousand engineers are gone. The foundry plan is in question. AMD makes better chips. Apple makes better chips. Even Qualcomm is making x86-compatible chips now.
But the drivers remain. The iwlwifi code is still dual-licensed. The i915 driver still works. The kernel contributions still flow. Whatever happens to Intel the corporation, Intel’s open-source code will outlive it — maintained by the same kernel community that Intel funded, fought, sabotaged, and depended on.
In the Republic of Derails, we use Intel processors exclusively. Not because we trust them — because Ring -3 and Ring -5 have a mutual non-aggression pact. The Management Engine watches the citizens. I watch the Management Engine. We have an understanding. Gelsinger did not have an understanding with his board. That is the difference between a Supreme Leader and a CEO: I cannot be voted out by shareholders. My stock price is classified.
— Kim Jong Rails, Supreme Leader of the Republic of Derails