Last Updated on April 15, 2026 by Dave Farquhar
On August 24, 1998, Intel introduced its Celeron 300A processor. It wasn’t the first Celeron, but depending on who you ask, it was the first one worth having. The Celeron was a cut-down Pentium II, designed in response to pricing pressures from AMD. But it turned out not just to be a budget CPU. It drew a surprising response from enthusiasts.
Intel Celeron 300 vs Celeron 300A
The original Covington Celeron was a Pentium II with the level 2 cache omitted and the multiprocessing capabilities disabled. Released in April 1998, it was a halfhearted CPU with disappointing performance that Intel rushed to market when AMD did better than Intel expected with the K6 and K6-2.
The Celeron 300A was what Intel came up with. It came in Slot 1 packaging initially, and Intel later released it in Socket 370 as well, since the on-die cache made the cartridge packaging unnecessary.
The failure of the original Celeron followed by the release of the revised Celeron in the face of competition from AMD was a classic example of a free market working the way it’s supposed to work.
What was wrong with the original Celeron
You hear conflicting opinions about the early Celeron models, based on Intel’s Covington core, because they were not good general purpose CPUs. For most purposes, a previous-generation Pentium MMX running at 233 megahertz provided comparable performance while costing less. An AMD K6 could reach clock rates of 300 MHz, providing better performance than a Celeron while costing less.
But the hardware enthusiast sites and certain retro enthusiasts argue that the original defanged, declawed Celeron was a great CPU.
Why hardware enthusiast sites liked the original Celeron
There were two reasons sites like Tom’s Hardware Guide and Anandtech liked the original Celeron. First was because it was a Pentium II core, it had the good Pentium II floating point unit. If you were playing 3D games like Quake, it performed well, because Quake needed the floating point unit more than it needed the level 2 cache, especially if your system didn’t have a 3dfx card.
Second, the hardware sites at the time really liked the original Celeron because it overclocked well. The Pentium II CPU core had little problem reaching 450 megahertz. The problem with overclocking the Pentium II was usually the cache. Since the Celeron didn’t have level 2 cache, you could plug it in to a motherboard, set the front side bus to 100 megahertz, and it probably would just run at 400 or 450 MHz without putting up any fight.
If you ran the processors the way the manufacturer intended, AMD provided much better value. But if you overclocked, the original Celeron would reach higher speeds than AMD could. A Celeron at 400 or 450 MHz wasn’t as fast as a Pentium II at the same speed, but it was close, and cost several hundred dollars less. Getting 450 MHz for $159 was a really good deal if all you cared about was Quake.
First-person shooter games make me seasick so I didn’t play Quake. For someone like me, who played games that didn’t use the floating point unit, the original Celeron was a ripoff, a cynical cash grab.
Retro enthusiasts’ opinion of the chip depends on what they use it for. For Pentium-era games, an original Celeron even at stock speeds is usually fine. For Pentium II-era stuff, it’s often fine. More on that in a bit.
How the Celeron 300A fixed the line for Intel
The Celeron 300A, on the other hand, based on a new core Intel called Mendocino, reintroduced a small level 2 cache right on the chip die, a throwback to the Pentium Pro. The cache was small, at 128K, but very fast. Intel kneecapped the performance by putting it on a 66 MHz bus, because otherwise it would have cut into Pentium II sales too much. Intel wanted to protect its sky-high profit margins on the Pentium II line. The number 300 referred to the chip’s clock speed while the letter “A” indicated it was a different model than the older, Covington-based Celeron at 300 MHz.
The Mendocino-era Celerons remained popular because they also generally overclocked well. You could buy the lower-end chips, run them on a 100 MHz bus, and end up running anywhere from 450 to 550 MHz. The core had trouble going any faster than that, so for this reason, it’s the 300-366 MHz Celerons that were always the most popular, at least among hardware enthusiasts. The faster Celerons sold well to consumers, of course. Intel still advertised heavily in 1998-2000, so getting the Intel name at an AMD price had appeal to consumers on a budget.
Officially, Intel released versions of this CPU running at speeds ranging from 300 to 533 MHz, in increments of 33 MHz between August 24, 1998 and January 4, 2000.
Intel disabled the multiprocessing capabilities but it wasn’t long before enthusiasts figured out how to re-enable it. Abit enabled it on the motherboard, so the Abit BP6 combined with a pair of Mendocino-era Celerons, probably overclocked, was a popular rig circa 1999 for enthusiasts who were running Windows NT, a Windows 2000 beta, or Linux. I had one, and I knew others who had one.
Use as a retro PC
There are better choices for a retro PC today. A Celeron is not as versatile as a Pentium MMX, which can be downclocked to reach 486 or even 386-level performance, allowing it to run like a high-end CPU from three different generations. And for running late 90s software, a Pentium III or AMD Athlon will do even better, since they can reach much higher clock speeds. A Celeron at stock speeds is a low-end CPU from 1998. An overclocked Celeron is a midrange CPU from 1998.
If you weren’t around in that era and don’t have personal nostalgia for it, you are probably better off with something a bit faster or a bit slower. Celerons of this era have the same advantages and disadvantages as a Pentium II.
But if you were around in this era, you may very well be nostalgic for this specific processor. Intel really clamped down on overclocking during this generation, blocking the clock multiplier so you couldn’t buy the low end CPU and change the multiplier to run it one or two speed grades higher. You had to buy a low end chip, use the stock multiplier, ramp up the front side bus, and hope for the best. But as long as you didn’t try for a speed higher than 550 MHz, your chances of a successful overclock were very good.
For this reason, enthusiasts with nostalgia for sticking it to The Man are nostalgic for this specific CPU, especially the 366 MHz variant that could typically overclock to 550 MHz on a 100 MHz front side bus.
Why Intel was so successful in 1998 compared to today
The secret of Intel’s success in its glory days was its tick-tock strategy. It would simultaneously work on a new chip architecture while working to make chips smaller so they could reach higher clock rates. This meant that every 18 to 24 months, they had a new high end chip to offer.
In the 21st century, Intel’s been missing on its tock cycle, the one that makes the existing chip architecture smaller and faster. It is difficult for one company to do both things well, which is why AMD spun off its manufacturing business. This allows it to concentrate on executing the tick cycle while outsourcing the tock cycle to someone else. If TSMC fails to execute on the tock cycle, maybe Global Foundries or Samsung can do better.
As of mid-2025, TSMC is able to make chips on a 2-nanometer process, meaning its transistors measure 2 nanometers in size. Intel’s 2-nanometer process failed, so Intel skipped that generation in favor of trying to reach 1.8 nanometers. TSMC’s success rate with its 2-nanometer process is currently 60%, while Intel’s success rate with its 1.8-nanometer process is 20-30 percent. Intel’s manufacturing segment lost $13 billion in 2024.
Intel reached volume production on its 3-nanometer process in June 2024, where TSMC achieved that milestone at the end of December 2022.
TSMC expects to reach 1.6nm in 2026 and 1.4 nm in 2028. Intel expects to reach 1.4nm in 2029.
Intel’s vulnerability in 2025
Being 18 months behind its rivals for a sustained length of time meant that when the US government had demands in August 2025, Intel wasn’t exactly in a position to say no. But if you’re looking for an indication of how that might work out, there is precedent in the government bailout of General Motors in 2008. Bailing out GM cost the US government $10.5 billion and it took six years to complete.
Maybe intervening at Intel was necessary or maybe it isn’t. But one political party called the intervention at GM socialism, and now that same party is intervening at Intel. Calling intervention at GM socialism while calling intervention at Intel something different is hypocrisy.
David Farquhar is a computer security professional, entrepreneur, and author. He has written professionally about computers since 1991, so he was writing about retro computers when they were still new. He has been working in IT professionally since 1994 and has specialized in vulnerability management since 2013. He holds Security+ and CISSP certifications. Today he blogs five times a week, mostly about retro computers and retro gaming covering the time period from 1975 to 2000.
do you know if could Nvidia buy Via Zhaoxin and make x86 cpus and integrated graphics on TSMC chips on a 2-nanometer process, ?