Yesterday we profiled Intel — the empire that bugs built. Today we profile the company that has been fighting that empire since 1969.

The Supreme Leader reads this from a BSD seat as well: AMD’s story is not just about Linux desktops winning benchmarks, but about the hardware that keeps decent Unix-like systems viable across the aisle too.

AMD.

Advanced Micro Devices. Or, as it is known in the Republic of Derails: MAD — with a Taiwanese accent. Rearrange the letters. AMD. MAD. The company has been angry at Intel since 1986. The name was prophetic.

AMD was founded on May 1, 1969, by Jerry Sanders and seven colleagues from Fairchild Semiconductor. Sanders had been fired from Fairchild because his flamboyant personality — silk suits, sports cars, Hollywood lifestyle — made the new management “uneasy.” He took the rejection, took seven engineers, incorporated with $100,000, and started building semiconductors in the back of a rug-cutting company in Santa Clara.

The company that would one day force Intel to adopt its 64-bit architecture began its life behind a carpet factory.

The Second Source Agreement:

In the early days of x86, IBM required that every chip in its PCs have a second source — a backup manufacturer who could produce identical parts. This was standard practice. IBM did not want its supply chain dependent on a single vendor.

In February 1982, Intel and AMD signed a cross-licensing agreement. AMD would be Intel’s second source for the 8086 and 8088 processors — the chips that powered the original IBM PC. AMD would manufacture Intel-designed chips. Intel would accept AMD-designed chips in exchange. A system of “complexity factor points” served as currency for the exchanges.

This agreement made AMD a legitimate x86 manufacturer. AMD built Intel’s chips. Intel was supposed to accept AMD’s chips in return.

Intel did not.

The 386 Betrayal:

By 1986, Intel had released the 80386 — the processor that would define the next generation of PCs. AMD needed complexity factor points to acquire the rights to produce the 386. Intel was supposed to accept AMD products to generate those points. Intel quietly refused to accept anything, leaving AMD without enough points to acquire the 386 rights.

Intel’s strategy, as the arbitrator later found, was deliberate deception: “keeping AMD in the Intel ‘camp,’ keeping AMD away from some sort of partnership with another company — say NEC or Fujitsu — where AMD’s talents might be used in effective competition against Intel.”

Intel pretended the partnership was fine. Intel was already planning to lock AMD out.

AMD filed for arbitration in August 1987. The arbitration took four and a half years. Three hundred and fifty-five days of hearings. The arbitrator ruled Intel had breached the covenant of good faith and fair dealing, and awarded AMD a permanent, nonexclusive, royalty-free license to produce the Am386.

AMD’s Am386 shipped. It was faster than Intel’s 386. It was almost as fast as Intel’s brand-new 486 — at a fraction of the price.

Intel immediately sued AMD over the Am486. The litigation continued until 1995, when Intel and AMD finally settled all claims.

Thirteen years of legal warfare. From 1982 to 1995. Over whether AMD had the right to make x86 processors.

Year	Event
1982	Intel-AMD cross-license agreement signed
1986	Intel refuses to let AMD produce the 386
1987	AMD files for arbitration
1992	Arbitrator rules Intel breached the agreement
1993	Intel sues AMD over the Am486
1995	All litigation settled

The takeaway: AMD’s right to exist as an x86 competitor was not given — it was litigated. Intel tried to lock AMD out at the 386. AMD fought for thirteen years and won the right to compete. Everything that followed — Athlon, Opteron, Ryzen, EPYC — exists because AMD refused to die in 1986.

The Athlon: When AMD Was Faster

The AMD Athlon (K7) launched on June 23, 1999. It was the fastest x86 processor in the world.

Not “competitive with Intel.” Not “good for the price.” The fastest. From August 1999 until January 2002, the Athlon held the performance crown. When Intel prepared to launch a 1 GHz Pentium III in March 2000, AMD stole the announcement by shipping a 1 GHz Athlon two days earlier.

The first 1 GHz x86 processor was AMD, not Intel.

The Athlon 64 (K8, 2003) went further. It was the first x86 processor with an integrated memory controller — drastically reducing memory latency. It was the first x86 processor with 64-bit extensions. And it was the architecture that killed Itanium.

The AMD64 Coup:

This is the single most important thing AMD ever did.

In August 2000, AMD published the specification for x86-64 (later AMD64) — a set of 64-bit extensions to the existing x86 instruction set. Fully backward compatible with 32-bit software. No emulation. No penalty. Your old software runs at full speed. Your new software gets 64-bit addressing.

Intel had bet everything on IA-64 (Itanium) — a clean 64-bit architecture that was not backward compatible with x86. Itanium required emulation to run existing software, and the emulation was painfully slow.

AMD’s approach: extend the present. Intel’s approach: replace the present.

The market chose AMD. By June 2004, Intel was forced to adopt AMD’s x86-64 extensions — rebranded as “EM64T” and later “Intel 64.” The company that invented x86 had to license its 64-bit future from AMD.

Every 64-bit x86 program you run today uses AMD’s architecture. Every 64-bit operating system. Every server in every datacenter running x86-64. That is AMD’s instruction set. Intel pays AMD for the right to use it, just as AMD pays Intel for the right to use x86. A mutual hostage situation that has lasted twenty years.

The Bulldozer Disaster:

Then AMD fumbled.

The Bulldozer architecture (2011) was supposed to be AMD’s next leap. It introduced Clustered Multithreading (CMT) — each “module” contained two integer cores but shared a single floating-point unit between them.

AMD marketed the FX-8150 as an 8-core processor. It had 8 integer execution units. But it had only 4 floating-point units. If both cores in a module needed the FPU simultaneously, they fought over it.

In November 2015, a class action lawsuit was filed alleging AMD had deceived consumers by advertising Bulldozer chips as having more cores than they actually contained. The lawsuit argued it was impossible for an “8-core” Bulldozer to execute 8 simultaneous floating-point operations — because it only had 4 FPUs.

AMD eventually settled for $12.1 million in 2019.

The performance was worse than the marketing. Bulldozer-era FX processors lost to Intel’s Sandy Bridge and Ivy Bridge in nearly every benchmark. Single-threaded performance — the metric that matters most for desktop users — was abysmal. AMD went from making the fastest x86 processor (Athlon 64) to making processors that enthusiasts actively avoided.

Near-Death:

AMD almost died. More than once.

In 2008, AMD reported losses of $3.1 billion in a single year. The stock would eventually fall to $1.61 — a price that valued the entire company at less than the cost of one modern semiconductor fab.

In 2009, AMD sold its manufacturing division. The fabs — the physical factories that made AMD chips — were spun off as GlobalFoundries, backed by Abu Dhabi’s Advanced Technology Investment Company. AMD went fabless. The company that had competed with Intel by making its own chips could no longer afford to make chips.

AMD was now a design company that paid someone else to manufacture its products. This is like a restaurant selling its kitchen and renting cooking time from a catering company. It works. It is also an admission that you cannot afford your own stove.

Year	AMD Stock Price	What Was Happening
2006	$42	ATI acquisition ($5.4B)
2008	$2-4	$3.1B annual loss
2012	$1.61	Rock bottom
2014	$2-3	Lisa Su becomes CEO
2017	$10-14	Ryzen launches
2020	$80-95	Overtakes Intel in mobile chips
2024	$140-190	EPYC dominates servers

Lisa Su: The Turnaround

In October 2014, AMD hired Dr. Lisa Su as CEO. She inherited a company with shrinking revenue, no competitive CPU architecture, a GPU division (from the $5.4 billion ATI acquisition in 2006) that was losing to NVIDIA, and a stock price under $3.

Su bet everything on a new architecture: Zen.

Zen was designed by Jim Keller — the same engineer who had designed the Athlon 64 (K8) that killed Itanium. Keller had left AMD, gone to Apple (where he worked on the A4/A5 chips), then returned to AMD in 2012. He designed Zen’s foundation before leaving again in 2015.

March 2, 2017: AMD launched Ryzen — the first Zen-based desktop processors. The Ryzen 7 1800X offered 8 cores and 16 threads for $499, competing with Intel’s $1,000+ Core i7-6900K.

The reviews were unanimous: AMD was competitive again. Not just “good for the price” — genuinely competitive at every price point. After a decade of Bulldozer-era mediocrity, AMD had a processor architecture that could match Intel core-for-core.

Subsequent generations — Zen 2 (2019), Zen 3 (2020), Zen 4 (2022), Zen 5 (2024) — progressively took the lead. EPYC server processors began replacing Intel Xeon in datacenters. AMD went from a company that Intel tried to kill in 1986 to a company that Intel could not beat in 2024.

Lisa Su took a $2 stock and turned it into a $180 stock. The MAD company had reason to be satisfied for the first time in a decade.

The Open Source Record:

AMD’s open-source driver story is dramatically different from Intel’s. It started terrible and got good.

When AMD acquired ATI in 2006, ATI’s Linux graphics drivers were proprietary disasters — the fglrx driver was notorious for crashes, kernel incompatibilities, and features that simply did not work. The community’s opinion of ATI’s Linux support was somewhere between Realtek and a dumpster fire.

Then AMD changed course.

Year	Event
2007	AMD releases GPU documentation for open-source driver development
2007	radeon open-source driver development begins
2015	AMDGPU kernel driver introduced (replacing radeon for GCN+ GPUs)
2016	ROCm open-source GPU compute platform released
2017	radeonsi Mesa driver becomes competitive with proprietary for gaming
2019	RADV (Vulkan driver) reaches near-parity with proprietary
2024	AMD funds ZLUDA — a drop-in CUDA compatibility layer on ROCm

Today, AMD’s open-source graphics stack is genuinely good. The AMDGPU kernel driver is mainline. The Mesa radeonsi (OpenGL) and RADV (Vulkan) drivers are community-maintained with AMD employee contributions. You can install Linux on an AMD GPU system and have working 3D acceleration out of the box, with no proprietary blobs for the graphics driver itself.

Compare this to NVIDIA’s approach — where “open source” means moving the proprietary code into firmware and open-sourcing the shim that calls it. AMD’s open-source commitment is real. Imperfect — ROCm still trails CUDA by years — but real.

The AMD PSP Problem:

Of course, AMD has its own skeleton.

The AMD Platform Security Processor (PSP) is AMD’s equivalent of Intel’s Management Engine. An ARM Cortex-A5 core embedded in every AMD processor since 2013. It runs its own firmware. It has access to system memory. It boots before the main x86 cores. Its firmware lives on the same Winbond SPI flash chip as your BIOS.

AMD’s PSP is equally opaque, equally mandatory, and equally impossible to disable as Intel ME. The company that has been the underdog fighting Intel’s monopoly embedded the exact same surveillance architecture in its own chips.

The difference: Intel’s ME has had multiple catastrophic vulnerabilities (CVE-2017-5689, SA-00086). AMD’s PSP has had fewer public incidents — not necessarily because it is more secure, but possibly because fewer researchers have examined it. Security through obscurity is not security. It is a timer.

The CTS-Labs Controversy:

In March 2018, an Israeli security firm called CTS-Labs published a report claiming 13 critical vulnerabilities in AMD’s Zen processors — with names like “Ryzenfall,” “Masterkey,” “Fallout,” and “Chimera.”

The catch: CTS-Labs gave AMD only 24 hours notice before going public. Standard responsible disclosure gives vendors 90 days. CTS-Labs gave AMD one day.

The security community was immediately suspicious. The report was accompanied by a slick website. CTS-Labs had ties to a short-selling firm that stood to profit if AMD’s stock dropped. The vulnerabilities, while real, required local administrator access to exploit — meaning an attacker who could use them already had full control of the machine.

AMD patched the issues through firmware updates. CTS-Labs was widely criticized for weaponizing vulnerability disclosure for financial gain. The incident revealed more about the corruption of security research than about AMD’s hardware.

The Zenbleed Bug:

In July 2023, Google security researcher Tavis Ormandy discovered Zenbleed (CVE-2023-20593) — a bug in AMD’s Zen 2 processors that could leak data across processes at a rate of 30 KB per core per second.

The bug exploited a flaw in the vzeroupper instruction’s speculative execution. An attacker could read data from other processes, other virtual machines, and other security contexts — including encryption keys and passwords.

Unlike Meltdown (which was Intel-only), Zenbleed was AMD’s own speculative execution vulnerability. The irony: AMD had spent years marketing its immunity to Meltdown. Zenbleed proved that aggressive speculative execution is a loaded gun regardless of who builds the chip.

The Vendor Verdict:

Vendor	Sin
Broadcom	Hostility
Realtek	Indifference
MediaTek	Velocity
Intel	Contradiction
AMD	Persistence

AMD’s sin is not a sin. It is a survival trait. They have been fighting Intel since 1986, fighting bankruptcy since 2008, fighting NVIDIA since 2006, and fighting their own architecture since 2011. They persist. Not always well. Not always wisely. But always forward.

The Lesson:

AMD was founded in the back of a carpet factory. It was betrayed by Intel over the 386. It fought for thirteen years in court for the right to make x86 chips. It built the fastest processor in the world, then built the worst. It lost $3.1 billion in a year. Its stock hit $1.61. It sold its own factories. It was left for dead.

Then it hired Lisa Su, shipped Zen, and built processors that Intel could not match.

The company name is Advanced Micro Devices. The acronym rearranges to MAD. The anger was justified. The revenge was Ryzen.

There is one more fact about AMD that nobody seems to notice. A family connection that links AMD to its greatest rival in the GPU market. But that story belongs to tomorrow’s post. Some revelations require their own stage.

In the Republic of Derails, we understand family rivalries. Two cousins can love each other at the Lunar New Year dinner and compete to the death in the semiconductor market. The key is to never discuss stock prices at the dinner table. We tried combining both sides once — the result looked like the flag of Morocco. We abandoned the project.

— Kim Jong Rails, Supreme Leader of the Republic of Derails