Worst CPU's of all time?

[Unabomber]

Another one that should have gotten an honorable mention in my book...

The Power PC 601, not because of the CPU design, but because of the systems in which it were used.

In an RS/6000 workstation running AIX, it was actually a really good performer, and I used one such workstation for molecular modeling back in the earlier part of the 90's.

In the Power Mac systems that were in my lab at the time, they were sluggish, to say the least, even with 16 MB of memory installed. My 2 year old Pentium 90 was already running circles around them, and it was no surprise that people wanted to use my system instead of those grinding Power Macs using those PPC 601's. If anything, even the Mac Quadra that used a 33 MHz Motorola 68040 felt much snappier and more responsive than those Power Macs.

 

[Grebuloner]

Another one that should have gotten an honorable mention in my book...

The Power PC 601, not because of the CPU design, but because of the systems in which it were used.

In an RS/6000 workstation running AIX, it was actually a really good performer, and I used one such workstation for molecular modeling back in the earlier part of the 90's.

In the Power Mac systems that were in my lab at the time, they were sluggish, to say the least, even with 16 MB of memory installed. My 2 year old Pentium 90 was already running circles around them, and it was no surprise that people wanted to use my system instead of those grinding Power Macs using those PPC 601's. If anything, even the Mac Quadra that used a 33 MHz Motorola 68040 felt much snappier and more responsive than those Power Macs.

Oof, YES. I had to use an 80MHz PowerMac for Adobe Pagemaker. Wasn't even loading actual images into the layouts (that was done at the printer) but there was so much chugging back and forth. Not really related to the CPU, but when it came time to close up and make our network saves, it was all done over AppleTalk. Nothing like having worked on a couple layouts and thinking "I have 15 minutes left, better start the transfer!"

 

[ochadd]

Not bad CPUs but bad decisions and timing.
1. AMD 64 single core instead of getting my first dual core. AMD socket 939 generation. Swapped it out for an AMD X2 3800+ that lasted a decade+ as a hand down to the family.
2. "Upgraded" from a Core 2 duo e6600 to a CPU with less cache that overclocked a tiny bit higher. I think it was a Core 2 duo e7300. Instead of a Q6600.
3. 2600k upgraded to a 7700k just to scratch the upgrade itch. It just wasn't worth the $ and should have went AM4.

 

[xSneak]

Lmao. Ryzen 1000s were not garbage by any means. What are you on?

They were out performed in games across the board by the 4770k / 4790k, cpus that were released 4 years prior. Their performance in pubg, the most popular game in the world in 2017, was woeful.

 

[NightReaver]

They were out performed in games across the board by the 4770k / 4790k, cpus that were released 4 years prior. Their performance in pubg, the most popular game in the world in 2017, was woeful.

So a cpu being lower in a single metric for you means it's garbage? Alright. Maybe for you, but no, it does not put it in "worst cpus of all time".

 

[NamelessPFG]

Another one that should have gotten an honorable mention in my book...

The Power PC 601, not because of the CPU design, but because of the systems in which it were used.

In an RS/6000 workstation running AIX, it was actually a really good performer, and I used one such workstation for molecular modeling back in the earlier part of the 90's.

In the Power Mac systems that were in my lab at the time, they were sluggish, to say the least, even with 16 MB of memory installed. My 2 year old Pentium 90 was already running circles around them, and it was no surprise that people wanted to use my system instead of those grinding Power Macs using those PPC 601's. If anything, even the Mac Quadra that used a 33 MHz Motorola 68040 felt much snappier and more responsive than those Power Macs.

I have limited experience with the 601 and no experience with AIX workstations (I've at least owned SPARC/Solaris and MIPS/IRIX setups, and seen DEC Alpha/NT systems exhibited), but as fate would have it, last year I got a Quadra 950 with the Power Macintosh Upgrade Card that adds a 601 through the PDS. Made comparing a 33 MHz 68040 and a 66 MHz PowerPC 601 real easy.

I can't say that the 601 made it sluggish; quite the contrary, in that Marathon framerates benefited noticeably and most other 68k-native games do actually run faster, from what I can discern.

With that said, going from preemptive multitasking UNIX to cooperative multitasking Mac OS had to feel pretty painful at the time, moreso when considering how long it took them to replace the classic Mac OS. That'd make anything feel sluggish regardless of what CPU you had once you attempt anything resembling real multitasking.

 

[Gomar]

Cyrix failed for a reason.

 

[The Cobra]

Pentium Overdrive processors.

 

[Starfalcon]

Pentium Overdrive processors.

They werent really bad, but they were very dependant on the board you put them in. If it was a low end board they would be hamstrung by the slow speeds of the board, and not really perform better then the chip you were replacing in it.

 

[funkydmunky]

Cyrix failed for a reason.

Too bad. I forget all the in's and outs of who had and why they don't now, X86 licenses.
Via made some X86 CPU's under their license. Never came across one ever, but surely a vote for them worst as well should be had.
Imagine if the glory days of FOUR x86 CPU producers were able to persevere. But it took big $$ and AMD only themselves by a hair.

 

[Unabomber]

Pentium Overdrive processors.

They did OK, getting you closer to a Pentium 60 or 66 for most purposes, but it certainly wasn't worth the time and $$$ at that point, since it came out so late. By that time, a lot of people were phasing out the 486 systems in favor of Pentium systems.

To be honest, someone could get such performance by sticking in a 486 DX4-100, whether it be Intel, AMD, or one of Intel's 486 DX4 Overdrive CPU's. The 120 and 133 MHz CPU's also worked great, as long as your board supported them.

At that point, though,

Regarding Cyrix, they had some really nice offerings in the past, such as the excellent FastMath math co-processor, that was significantly more powerful than the Intel offering. I often times wonder why they couldn't scale things up once they entered the true 486 market.

 

[lopoetve]

They did OK, getting you closer to a Pentium 60 or 66 for most purposes, but it certainly wasn't worth the time and $$$ at that point, since it came out so late. By that time, a lot of people were phasing out the 486 systems in favor of Pentium systems.

To be honest, someone could get such performance by sticking in a 486 DX4-100, whether it be Intel, AMD, or one of Intel's 486 DX4 Overdrive CPU's. The 120 and 133 MHz CPU's also worked great, as long as your board supported them.

At that point, though,

Regarding Cyrix, they had some really nice offerings in the past, such as the excellent FastMath math co-processor, that was significantly more powerful than the Intel offering. I often times wonder why they couldn't scale things up once they entered the true 486 market.

Architecture changed. Pentium time especially Intel made massive improvements in their FPU modules and other parts and Cyrix couldn’t keep up.

 

[lopoetve]

Too bad. I forget all the in's and outs of who had and why they don't now, X86 licenses.
Via made some X86 CPU's under their license. Never came across one ever, but surely a vote for them worst as well should be had.
Imagine if the glory days of FOUR x86 CPU producers were able to persevere. But it took big $$ and AMD only themselves by a hair.

Four just meant lots of compatibility workarounds and bugs.

 

[NamelessPFG]

Architecture changed. Pentium time especially Intel made massive improvements in their FPU modules and other parts and Cyrix couldn’t keep up.

Not to mention FPU performance suddenly became very important when this little thing called Quake showed up, and it was tailored for the Pentium in particular...

I don't think VQuake and GLQuake were a thing for another year or two, so software rendering was what you had to make do with back then, and the Pentium was just far better at it than the competition at that time.

What I am wondering is how Cyrix couldn't turn around a better architecture in time to survive while AMD could, eventually ditching the K6 and all that Super Socket 7 jank for K7/Athlon.

 

[funkydmunky]

Four just meant lots of compatibility workarounds and bugs.

It would have evolved to be much as it is today. Don't see Intel and AMD as having lots of compatibility workarounds and bugs.

 

[lopoetve]

It would have evolved to be much as it is today. Don't see Intel and AMD as having lots of compatibility workarounds and bugs.

Hah! Oh you optimist. You sweet summer child.

Look at amd vs nvidia . We can’t even get two video card manufacturers to not have weird compatibility issues at times. Four CPUs? With the pile of errata, microcode updates, security patches…?!? Oh. Oh god. The pain. Especially if they share sockets somehow like the super 7 days.

 

[lopoetve]

Not to mention FPU performance suddenly became very important when this little thing called Quake showed up, and it was tailored for the Pentium in particular...

I don't think VQuake and GLQuake were a thing for another year or two, so software rendering was what you had to make do with back then, and the Pentium was just far better at it than the competition at that time.

What I am wondering is how Cyrix couldn't turn around a better architecture in time to survive while AMD could, eventually ditching the K6 and all that Super Socket 7 jank for K7/Athlon.

K6 was good enough while the 6x86 wasn’t; and had other issues too (DMA disk transfers and other oddball problems). They were just good enough to buy runway - cyrix and amd were splitting the “not Intel” market share (vs the three splitting the x86 market share), and amd got the lions share. The Athlon was a huge gamble.

 

[funkydmunky]

Hah! Oh you optimist. You sweet summer child.

Look at amd vs nvidia . We can’t even get two video card manufacturers to not have weird compatibility issues at times. Four CPUs? With the pile of errata, microcode updates, security patches…?!? Oh. Oh god. The pain. Especially if they share sockets somehow like the super 7 days.

I don't recall any software that I can or can't run on my AMD or only Intel systems. If the other two had managed to get this far I imagine it would be much the same to the end user.

 

[lopoetve]

I don't recall any software that I can or can't run on my AMD or only Intel systems. If the other two had managed to get this far I imagine it would be much the same to the end user.

I meant video cards now.

Back then there were tons of issues. Having been in the OS development side we’ve done a very good job of hiding all that- but you don’t see the massive piles of code making that possible with only two vendors to support. If we added in two more just for x86…. I can only imagine the nightmare it would be.

And there are still oddities that happen- scheduler fun, bugs, weird performance issues that pop up, etc.

Now imagine those, with 4 different interpretations of x86, potentially four (or more) different platform setups and interpretations…. You ever have to pick between SiS, Via, AMD, or Nvidia for your chipset, before picking the CPU to go in it? The multitude of combinations to support…. Ugh.

 

[zandor]

Just a random thought... aside from buggy x86 CPUs, should we even be looking at x86 to pick a worst of all time? Back in the old days, and largely before my time, there were all sorts of weird ass CPU designs and some of them were total shit. I'm 46 and I was around for x86, Motorola 68xxx, Alpha, MIPs, PA-RISC, PowerPC (kinda still going), etc. All of those mostly worked properly. If you go back farther, before the mid '90s when I got into this stuff, there are all sorts of designs that are completely out of line with any of the ones that "won". Go back to the '70s and earlier in particular and there's all sorts of really weird designs people were playing with. I don't know enough of the history, but I bet if we go back that far we could find some "wtf were you thinking!?!" designs and not just stuff that had a bug or two or sucked at gaming compared to the competition.

 

[lopoetve]

Just a random thought... aside from buggy x86 CPUs, should we even be looking at x86 to pick a worst of all time? Back in the old days, and largely before my time, there were all sorts of weird ass CPU designs and some of them were total shit. I'm 46 and I was around for x86, Motorola 68xxx, Alpha, MIPs, PA-RISC, PowerPC (kinda still going), etc. All of those mostly worked properly. If you go back farther, before the mid '90s when I got into this stuff, there are all sorts of designs that are completely out of line with any of the ones that "won". Go back to the '70s and earlier in particular and there's all sorts of really weird designs people were playing with. I don't know enough of the history, but I bet if we go back that far we could find some "wtf were you thinking!?!" designs and not just stuff that had a bug or two or sucked at gaming compared to the competition.

Yeah, but pre-80s it’s all stuff that none of us would likely be touching. When you get to the early minicomputer days we’re talking things that cost more than cars. This is more for the era of us actually getting to use it.

 

[pitingres]

Yes, there were a whole pile of wannabee architectures in the late 70s / early 80s. Off the top of my head I can think of Pyramid, Data General, MIPS, and of course SPARC which was very successful for a long time. Harris had some sort of goofy 24-bit architecture at one point. None were intended as personal computers, though. The closest one to being affordable was, IIRC, the disaster I've already voted for: the AT&T 3B2/300 with its WE32000 CPU. I don't recall the purchase price, however if it was more than $10k it wasn't by much, and a minimum config might have come in under that. (The startup I was working for at the time bought one when we were getting close to running out of cash, so it couldn't have been all that expensive.)

 

[SamirD]

It's gonna be hard to find something worse than a Cyrix Cx486SLC. I mean, they were glorified 16-bit 386SX style processors but with worse compatibility. Not only were they objectively bad, but marketing deception on top of that.

omg, that chip was awesome as it allowed us to upgrade our IBM 30-286 to a 486 and run win3.1 in enhanced mode comfortably. In the right application, it was a really lovely upgrade.

 

[SamirD]

I can't remember what I did with it, but I'm pretty sure I gave it away for free. The sad part is they still got ripped off. Fuck you Transmeta.

Tell us how you really feel, lol.

 

[SamirD]

Too bad. I forget all the in's and outs of who had and why they don't now, X86 licenses.
Via made some X86 CPU's under their license. Never came across one ever, but surely a vote for them worst as well should be had.
Imagine if the glory days of FOUR x86 CPU producers were able to persevere. But it took big $$ and AMD only themselves by a hair.

I have some neoware thin clients that run xp embedded that use those via cpus. For this use case, they're really good. Been really reliable too--these things have been on for nearly 2 decades now 24x7.

 

[SamirD]

I based what I think was the worst based on what my memory was from the time:

• Original Pentium 60/66 - Once the idea of using a multiplier kicked in, what was the point of these? Could have just made 486dx3-99, 486dx4-80, up to 486dx4-200 at the time--and a 486dx4-200 would have been nearly 2x as fast as the Pentium 60. Even today, we're still running multiplers on the fsb bus--for example, the 11700k is a 36x clock multiplier on a 100Mhz bus. Just for one second imagine what a 486dx36-1188 would have felt like--and this would have been with just a 33mhz bus. The 486DX-50 proved a 50Mhz bus worked, so crank that up and it's a 486dx36-1800. When you look at the time span and the bus speed change from 50 to only 100 in the course of 25 years, you see how far we've not really come.
• Pentium Pro - was just too far ahead of its time. Fully 32-bit stuff didn't show up until nearly 10 years later.
• Dec Alpha - similar fate as the PPro in that it was just in a niche on its own.
• Celeron - Oh man how I hated these when they first came out. A cpu with a disabled cache?--wth!! They were dog slow and just had the mhz number to fool suckers who got them. I didn't know about overclocking then, so they obviously had a strong point too--I just never saw it.

All I can remember atm.

 

[Valnar]

omg, that chip was awesome as it allowed us to upgrade our IBM 30-286 to a 486 and run win3.1 in enhanced mode comfortably. In the right application, it was a really lovely upgrade.

Funny how we remember things differently. Those were just faster 386SX chips - 32-bit CPU over 16-bit bus. I actually sold computers at a store back then (college) and some Leading Edge desktops came with a 486SLC from the factory. The computers were cheap$ and cheaply made. For the price point, it was hard to convince neophytes that THIS 486 is not the same as brand X Intel 486DX desktop. No doubt many of those customers were quite disappointed in the performance.

My personal anecdote was that Cyrix chip couldn't run 386Max, my memory manager at the time. Applications like QEMM386 or 386Max is where you can really stress test the low level compatibility of an Intel "compatible" CPU. To this day, these are the acid tests one can use to test DOS compatibility in various emulators (VMware, DOSBox, PCEM, Virtualbox, etc). Most fail.

 

[Grebuloner]

Original Pentium 60/66 - Once the idea of using a multiplier kicked in, what was the point of these? Could have just made 486dx3-99, 486dx4-80, up to 486dx4-200 at the time--and a 486dx4-200 would have been nearly 2x as fast as the Pentium 60. Even today, we're still running multiplers on the fsb bus--for example, the 11700k is a 36x clock multiplier on a 100Mhz bus. Just for one second imagine what a 486dx36-1188 would have felt like--and this would have been with just a 33mhz bus. The 486DX-50 proved a 50Mhz bus worked, so crank that up and it's a 486dx36-1800. When you look at the time span and the bus speed change from 50 to only 100 in the course of 25 years, you see how far we've not really come.

Not exactly. FSB died at Intel when Core 2 retired and topped out at 400MHz ("1600MT/s") with various improvements over time. It died at AMD in 2001 with the release of HyperTransport. From Nehalem on, particularly because of the IMC, it became just a 100MHz reference clock for the system, not a bus at all. DMI took care of the bus duties with PCIe, moreso once separate Northbridge/Southbridge chips went away with Sandy Bridge.

 

[SamirD]

Funny how we remember things differently. Those were just faster 386SX chips - 32-bit CPU over 16-bit bus. I actually sold computers at a store back then (college) and some Leading Edge desktops came with a 486SLC from the factory. The computers were cheap$ and cheaply made. For the price point, it was hard to convince neophytes that THIS 486 is not the same as brand X Intel 486DX desktop. No doubt many of those customers were quite disappointed in the performance.

My personal anecdote was that Cyrix chip couldn't run 386Max, my memory manager at the time. Applications like QEMM386 or 386Max is where you can really stress test the low level compatibility of an Intel "compatible" CPU. To this day, these are the acid tests one can use to test DOS compatibility in various emulators (VMware, DOSBox, PCEM, Virtualbox, etc). Most fail.

True, but that ability to bring a 32-bit bus to a 16-bit socket was their strength, and they did run faster than the 386sx upgrades we tried in the 30-286. The Cyrix was the fastest of all of them once the cache was enabled.

QEMM386 was always a pain with compatibility from my memory so once Microsoft introduced their own with DOS 5.0, I never even tried it again. I think I did use the memory examiner or whatever it was called in 386Max to figure out areas in the UMB range that were free, and then I would manually include and exclude these ranges in EMM386.

 

[SamirD]

Not exactly. FSB died at Intel when Core 2 retired and topped out at 400MHz ("1600MT/s") with various improvements over time. It died at AMD in 2001 with the release of HyperTransport. From Nehalem on, particularly because of the IMC, it became just a 100MHz reference clock for the system, not a bus at all. DMI took care of the bus duties with PCIe, moreso once separate Northbridge/Southbridge chips went away with Sandy Bridge.

Interesting. So why are processors still having multiplers then based on this 100Mhz reference? What's the real bus speed these days? Because memory speed is still a multiple too.

 

[Grebuloner]

Interesting. So why are processors still having multiplers then based on this 100Mhz reference? What's the real bus speed these days? Because memory speed is still a multiple too.

Many of the other parts of the system, including PCIe, rely on the 100MHz base clock to time their signals. This is also what made BClk overclocking so limited and introduced the "straps" compromise (another multiplier, really) with Sandy-E. It's a convenient reference that's easy to generate. Memory runs off the same clock, too.

The real bus speed kind of isn't at all. Memory has it's own speed through the multiplier as connected to the CPU. The CPU to chipset and directly connected slots is the 100MHz of PCIe but however many MT/s or GB/s the PCIe lanes provide there. It's not one bus to a Northbridge anymore, but more of a whole transit system now.

 

[SamirD]

Many of the other parts of the system, including PCIe, rely on the 100MHz base clock to time their signals. This is also what made BClk overclocking so limited and introduced the "straps" compromise (another multiplier, really) with Sandy-E. It's a convenient reference that's easy to generate. Memory runs off the same clock, too.

The real bus speed kind of isn't at all. Memory has it's own speed through the multiplier as connected to the CPU. The CPU to chipset and directly connected slots is the 100MHz of PCIe but however many MT/s or GB/s the PCIe lanes provide there. It's not one bus to a Northbridge anymore, but more of a whole transit system now.

Thank you for the detailed explanation. Makes a lot of sense since I forgot that there weren't really many buses per se during the 486 era besides the isa bus since everything was a card that interrupted the cpu, including io cards--the keyboard was the only built-in port. Then came along eisa, vesa, and pci and things got more complicated...

 

[Valnar]

EISA, Microchannel... that would be a fun topic for another thread. "The worst architecture or motherboard of all time"

Taking away those oddballs, I'd have to say anything with a VIA KT266.

 

[XoR_]

I based what I think was the worst based on what my memory was from the time:

• Original Pentium 60/66 - Once the idea of using a multiplier kicked in, what was the point of these?

Point was that

• you could not have 200MHz 486 in March 1993 because fabrication process was not mature enough for frequencies faster than 66MHz
• 586 bus was at least 2.7 times faster than fastest (and not very stable) 50MHz 486 bus
• many applications made good use of increased memory/IO bandwidth

CPUs for 486 platform could be developed much further and especially 486 cores with their relatively low IPC could make good use from higher clock even with limited 486 bus... but that would make no sense vs making new faster platform, which Intel did and it was glorious

The only thing which made remaining on 486 platform competitive and why anyone bothered making faster 486 variants (again: when fabrication process allowed them to be fabricated) was if you already had 486 motherboard and other tailored for 486 peripherals like VLB graphic cards or bought them used from people moving to 586 platform. Otherwise for new systems it would make little sense to develop 486 platform given its limitations.

Much less sense than 486 to 586 move was pointless socket changes like abandoning LGA1156 (or even better example LGA1366!).
What did we gain from move from LGA1156 to LGA1155 and then to LGA1150 and then to LGA1151? Absolutely nothing!
There was nothing stopping Intel from continuing making Sandy Bridge, Ivy Bridge, Hashwell, Broadwell, Skylake, Coffee Lake, Cannot Lake and Rocket Lake on LGA1156. There was no performance increase and even power delivery on LGA1156 was good enough as proven by how much power overclocked Nehalem i7's could draw. Certainly pin count didn't increase so it would make no sense to expect any improvements in power delivery.

Could have just made 486dx3-99, 486dx4-80, up to 486dx4-200 at the time--and a 486dx4-200 would have been nearly 2x as fast as the Pentium 60.

486@200MHz would not be 2 times faster than 486 @ 100MHz.
And certainly 486 @ 100MHz was not as fast as Pentium 66MHz.
You needed 133MHz 486 to get similar performance as Pentium 66MHz - of course only in INT/ALU because of weak 486 FPU unit compared to Pentium.

The 486DX-50 proved a 50Mhz bus worked, so crank that up and it's a 486dx36-1800. When you look at the time span and the bus speed change from 50 to only 100 in the course of 25 years, you see how far we've not really come.

50MHz FSB was known to be very problematic and had compatibility issues with many devices.
486 platform could of course be developed to arbitrarily high frequency but it made no sense to develop 32-bit bus when you could get twice bandwidth from moving to 64-bit.

• Even today, we're still running multipliers on the fsb bus--for example, the 11700k is a 36x clock multiplier on a 100Mhz bus. Just for one second imagine what a 486dx36-1188 would have felt like--and this would have been with just a 33mhz bus.

Today we have memory controller in processors and PCI-E controllers in processors and use buses similar to PCI-E between CPU with its North Bridge to South Bridge. 100MHz is just reference clock and has nothing to do with FSB other than FSB was also used as reference clock to derive clocks for other buses.

My last FSB system with Core 2 Duo E7200 and Intel chipset (do not remember which though) I had overclocked it past 500MHz FSB. With Intel QDR nomenclature it was >2GHz on FSB. Bandwidth wise it was >2000MHz / 66MHz of original Pentium bus bandwidth = >30 times faster. It can be calculated like that because QDR signaling quadrupled bandwidth. Core 2 bus still used 64-bit that original Pentium bus introduced. It was pretty high FSB speed and similar to fastest ever used 24/7 outside extreme overclocking.

Modern CPU's of course have something like FSB internally eg. Uncore Frequency can be thought of as something FSB-eque so one could argue you could still rate FSB speed even on processors with integrated memory controllers. There is no however any actual FSB because CPU doesn't communicate with external world - and memory specifically - using buses similar to FSB.

Pentium Pro - was just too far ahead of its time. Fully 32-bit stuff didn't show up until nearly 10 years later.

Another nonsense.
Not only we had lots of 32-bit systems and applications and games, even for DOS (FYI Doom was already fully 32-bit) but Pentium Pro didn't have anything to do with 32-bit other than dropping some of the legacy features (natively - they were still emulated in microcode) often used at the time because they were supported by previous CPUs.

In other words Pentium Pro was undercooked Pentium 2 with larger caches. I would not call it worst CPU other than it was costly to make, expensive and quickly superseded by improved Pentium 2 so not many people got it.

Dec Alpha - similar fate as the PPro in that it was just in a niche on its own.

There were lots of various RISC platforms with various degree of success/popularity.
I would not point out Alpha as anything remarkable in how good or bad it was.

Celeron - Oh man how I hated these when they first came out. A cpu with a disabled cache?--wth!! They were dog slow and just had the mhz number to fool suckers who got them. I didn't know about overclocking then, so they obviously had a strong point too--I just never saw it.

Only point I agree with.

First Celeron with no L2 was pretty much DOA product. Thankfully quickly replaced by gems like Celeron 300A.
Other really bad Cellies were Netburst based and they had terrible architecture to begin with which needed lots of cache and very little L2 cache, way too little to make their performance resemble their high clocks. These were imho much worse offenders because of their popularity in low-end builds and more precisely performance one could get by going to AMD for the same money.

Best Celeron was Celeron 300A which you could easily run at 450MHz and got pretty good bang for the buck and better lower cost platform than anything AMD K6 offered. Later Duron and even Athlons because of its still low cost became kings of budget systems so later Pentium II/III derived Celerons weren't as great. At least 440BX platform was better than anything available for Athlon at the time.

Another good Celerons are those with Tualatin core. They have large for Celerons cache of 256KB and paired with 440BX make for amazing retro PC platform with native ISA and great DOS/Win98 compatibility. One needs to hack Tualatin in to 440BX motherboards as it wasn't officially supported configuration. At the time these arrived and with artificially imposed chipset limitations they were too little too late to make any difference and AMD was king for

Myself I used 600MHz Celeron @ 750MHz on 440BX mostly due to cost and that was very good CPU for the time both cost and performance wise.
Later I had Celeron 1.8GHz@2.2GHz and this was solely for cost reasons. This thing was dog slow and didn't feel like 2.2GHz at all.
Friends Athlon 1GHz was obviously slower than my Celeron @ 2.2GHz but not by enough to even remotely begin justify clock difference. At least the money I myself paid for this Celeron system was adequate and there were cases like overall system performance where clock speed did give it decent performance vs K7 systems I could get for the same money. If I paid full price for this thing I would be really angry with its performance. Replaced it with 2.2GHz Athlon XP and this thing flied in comparison.

 

[XoR_]

EISA, Microchannel... that would be a fun topic for another thread. "The worst architecture or motherboard of all time"

MCA wasn't bad bus. It was just ignored by everyone. Not that many people used either MCA or E-ISA buses anyways.
Newer really saw E-ISA or MCA ports anywhere on any motherboard, not with my own eyes anyways.

Taking away those oddballs, I'd have to say anything with a VIA KT266.

Maybe it was bad but probably nowhere as bad as Intel variant of this chipset

 

[Valnar]

Another good Celerons are those with Tualatin core. They have large for Celerons cache of 256KB and paired with 440BX make for amazing retro PC platform with native ISA and great DOS/Win98 compatibility. One needs to hack Tualatin in to 440BX motherboards as it wasn't officially supported configuration. At the time these arrived and with artificially imposed chipset limitations they were too little too late to make any difference and AMD was king for

Funny I was just going to write something about this - one of the best Celerons every made. I have this exact setup on my retro Win98 box. ASUS P2B-B with a Tualatin Celeron at 1.4Ghz but only 100FSB. Gotta love that Powerleap PL-IP3/T v2.0.
https://web.archive.org/web/2018030...peed.ch/rscheidegger/p2b_procupgrade_faq.html

 

[SamirD]

EISA, Microchannel... that would be a fun topic for another thread. "The worst architecture or motherboard of all time"

Taking away those oddballs, I'd have to say anything with a VIA KT266.

EISA was pretty awesome in that it was backward compatible with ISA. Microchannel was just far too ahead of its time. It was basically the equivalent of PCI at the time with a 180 deg connector. What a lot of people didn't realize about PCI is that the cards were actually facing the other way when compared to ISA, so certain manufacturers like IBM realized you could put a PCI and ISA connector close together and the slot could be either PCI or ISA. My IBM Pentium Pro machines are set up like this. All someone had to do is design it with EISA and PCI and it would have been compatible with everything except VLB.

 

[Dan_D]

EISA, Microchannel... that would be a fun topic for another thread. "The worst architecture or motherboard of all time"

Taking away those oddballs, I'd have to say anything with a VIA KT266.

I've already done a worst motherboards of all time list that was a forum post and [H] article as well.

 

[SamirD]

Point was that

• you could not have 200MHz 486 in March 1993 because fabrication process was not mature enough for frequencies faster than 66MHz
• 586 bus was at least 2.7 times faster than fastest (and not very stable) 50MHz 486 bus
• many applications made good use of increased memory/IO bandwidth

CPUs for 486 platform could be developed much further and especially 486 cores with their relatively low IPC could make good use from higher clock even with limited 486 bus... but that would make no sense vs making new faster platform, which Intel did and it was glorious

The only thing which made remaining on 486 platform competitive and why anyone bothered making faster 486 variants (again: when fabrication process allowed them to be fabricated) was if you already had 486 motherboard and other tailored for 486 peripherals like VLB graphic cards or bought them used from people moving to 586 platform. Otherwise for new systems it would make little sense to develop 486 platform given its limitations.

Much less sense than 486 to 586 move was pointless socket changes like abandoning LGA1156 (or even better example LGA1366!).
What did we gain from move from LGA1156 to LGA1155 and then to LGA1150 and then to LGA1151? Absolutely nothing!
There was nothing stopping Intel from continuing making Sandy Bridge, Ivy Bridge, Hashwell, Broadwell, Skylake, Coffee Lake, Cannot Lake and Rocket Lake on LGA1156. There was no performance increase and even power delivery on LGA1156 was good enough as proven by how much power overclocked Nehalem i7's could draw. Certainly pin count didn't increase so it would make no sense to expect any improvements in power delivery.


486@200MHz would not be 2 times faster than 486 @ 100MHz.
And certainly 486 @ 100MHz was not as fast as Pentium 66MHz.
You needed 133MHz 486 to get similar performance as Pentium 66MHz - of course only in INT/ALU because of weak 486 FPU unit compared to Pentium.


50MHz FSB was known to be very problematic and had compatibility issues with many devices.
486 platform could of course be developed to arbitrarily high frequency but it made no sense to develop 32-bit bus when you could get twice bandwidth from moving to 64-bit.


Today we have memory controller in processors and PCI-E controllers in processors and use buses similar to PCI-E between CPU with its North Bridge to South Bridge. 100MHz is just reference clock and has nothing to do with FSB other than FSB was also used as reference clock to derive clocks for other buses.

My last FSB system with Core 2 Duo E7200 and Intel chipset (do not remember which though) I had overclocked it past 500MHz FSB. With Intel QDR nomenclature it was >2GHz on FSB. Bandwidth wise it was >2000MHz / 66MHz of original Pentium bus bandwidth = >30 times faster. It can be calculated like that because QDR signaling quadrupled bandwidth. Core 2 bus still used 64-bit that original Pentium bus introduced. It was pretty high FSB speed and similar to fastest ever used 24/7 outside extreme overclocking.

Modern CPU's of course have something like FSB internally eg. Uncore Frequency can be thought of as something FSB-eque so one could argue you could still rate FSB speed even on processors with integrated memory controllers. There is no however any actual FSB because CPU doesn't communicate with external world - and memory specifically - using buses similar to FSB.


Another nonsense.
Not only we had lots of 32-bit systems and applications and games, even for DOS (FYI Doom was already fully 32-bit) but Pentium Pro didn't have anything to do with 32-bit other than dropping some of the legacy features (natively - they were still emulated in microcode) often used at the time because they were supported by previous CPUs.

In other words Pentium Pro was undercooked Pentium 2 with larger caches. I would not call it worst CPU other than it was costly to make, expensive and quickly superseded by improved Pentium 2 so not many people got it.


There were lots of various RISC platforms with various degree of success/popularity.
I would not point out Alpha as anything remarkable in how good or bad it was.


Only point I agree with.

First Celeron with no L2 was pretty much DOA product. Thankfully quickly replaced by gems like Celeron 300A.
Other really bad Cellies were Netburst based and they had terrible architecture to begin with which needed lots of cache and very little L2 cache, way too little to make their performance resemble their high clocks. These were imho much worse offenders because of their popularity in low-end builds and more precisely performance one could get by going to AMD for the same money.

Best Celeron was Celeron 300A which you could easily run at 450MHz and got pretty good bang for the buck and better lower cost platform than anything AMD K6 offered. Later Duron and even Athlons because of its still low cost became kings of budget systems so later Pentium II/III derived Celerons weren't as great. At least 440BX platform was better than anything available for Athlon at the time.

Another good Celerons are those with Tualatin core. They have large for Celerons cache of 256KB and paired with 440BX make for amazing retro PC platform with native ISA and great DOS/Win98 compatibility. One needs to hack Tualatin in to 440BX motherboards as it wasn't officially supported configuration. At the time these arrived and with artificially imposed chipset limitations they were too little too late to make any difference and AMD was king for

Myself I used 600MHz Celeron @ 750MHz on 440BX mostly due to cost and that was very good CPU for the time both cost and performance wise.
Later I had Celeron 1.8GHz@2.2GHz and this was solely for cost reasons. This thing was dog slow and didn't feel like 2.2GHz at all.
Friends Athlon 1GHz was obviously slower than my Celeron @ 2.2GHz but not by enough to even remotely begin justify clock difference. At least the money I myself paid for this Celeron system was adequate and there were cases like overall system performance where clock speed did give it decent performance vs K7 systems I could get for the same money. If I paid full price for this thing I would be really angry with its performance. Replaced it with 2.2GHz Athlon XP and this thing flied in comparison.

That might have been the point, but living through that era, it made no sense.

486 chips were cheaper, more plentiful, and thanks to multipliers, all it took was a bios update in most cases and you could be running faster vs a whole new platform that would give you an incremental boost. Sure it was 'for the future', but no one really invested in it back then except people with money--specifically talking about those p5 60-66 versions.

Yes, the socket changes are definitely funny when the pins are essentially the same or just many more vcc and ground.

Of course a 486dx4-200 wouldn't have been 2x as fast as a 486dx2-100. Even a 486dx2-66 was only as fast as a 486dx-50. I tested these personally when we had built our 486 and also got a prebuilt that was a 486dx-50 with 12 EISA slots in a nice heavy case, but was more expensive than our build so we returned it.

I never got a chance to mess with a 486 beyond a dx2-66 and dx-50, but I remember the amd 486dx4-80 was pretty much on par with the p5-60 for all practical 'real world' purposes. Remember that the software back then was still mainly 16-bit so the pentium was a bit of a 'meh' with it's ability to do 32-bit stuff. I stand that a 486dx4-200 would have been 2x faster than the Pentium 60 in 'real world' usage of the software at that time. In fact, someone else actually lived it and someone else did a whole set of benchmarks that reflect what I remember:
https://dependency-injection.com/diamond-stealth-64-dram/?comment-21481
https://dependency-injection.com/the-perfect-pentium/

50Mhz bus worked fine when you made sure the cards you were using were also okay with it. The idea of creating additional busses required a whole re-work so backward compatibility became a problem. And back then stuff was really expensive so keeping your existing investment was important. Today, everything is just throwaway and that's stupid imo. Even with an order of magnitude faster systems from this previous era, regular every day usage isn't an order of magnitude faster, so there's been a lot of performance loss in scaling as well as bloat.

Yep, today's processors have changed the entire motherboard architecture even though 100Mhz is still used a refernce clock--I get that (wish cpuworld would note that better). Still it would have been neat to see where the old 486 architecture would have topped out even with the limitations. Because while all the improvements sound impressive on paper, add in the extra complexity of modern software and its impact is seriously lessened.

PPro was far ahead of its time--most of the software was still 16-bit when it was released. I was there--win 95 and 98 was out, but those were pieces of 32-bit, much like NT. The software was far behind being 32-bit because most of them still had to work on 16-bit environments as well.

For the record, Doom was NEVER 32-bit. The ID game engine used some 16-bit and 32-bit stuff, but if it was fully 32-bit you could never launch it from 8-bit DOS.

The Pentium Pro's microcode was optimized for 32-bit code vs 8 and 16-bit like previous processors were. The pentium 2 was simply a 'fix' for the mistake of putting the cache on the die and also correcting the microcode. A Pentium 2 had to have higher clock speeds to match a Pentium Pro in a fully 32-bit environment. It did, however, do better at 8 and 16-bit than the PPro due to the revised microcode. Again, I remember this because I lived it. What shocked me was the stupid 'cartridge' system scaled up all the way to 1Ghz before it went back to sockets again.

Alpha was one of the fastest NT platforms at its release--it was faster by far than anything else that could run NT in a desktop format. But this was a niche. The other RISC platforms didn't run Windows.

Oh so you agree with a point, should I be impressed or surprised...or even care? lol

I stopped building anything after our Cyrix P166+ based system as it was enough. I do have an HP or two with P3 processors in them as well as an IBM or two with a P3, several P4s, and all my Pentium Pros as well as our original 486dx-33 and 486dx2-66. I even still have the first PC that introduced me to the platform--the IBM PS/2 30-286. We modded that quite heavily adding SCSI, ethernet, and a 486SLC upgrade before it stopped being used and started sitting. I forgot I have a 486sx too--that was the original celeron imo--a chip with something disabled (the fpu). It started a trend that exists to this very day. All of these later systems besides the 486dx2-66 and P166+ were from business as hotel property management systems started upgrade cycles which led to a lot of working hardware ending up in utility chases until I rescued them a decade later and put them back to work.

 

[SamirD]

MCA wasn't bad bus. It was just ignored by everyone. Not that many people used either MCA or E-ISA buses anyways.
Newer really saw E-ISA or MCA ports anywhere on any motherboard, not with my own eyes anyways.

MCA wasn't really ignored, just not available--in terms of systems or cards. It was basically all by IBM for IBM, so if you weren't running the IBM PS/2 series, you never had to deal with it. And even the lower end PS/2s had ISA (thankfully). When upgrading our 30-286 we found out just how fortunate we were since a lot of the upgrades would have been impossibly cost prohibitive or simply impossible.

The EISA bus on the other hand was pretty much in every server and the cards were readily available. They weren't as popular as their ISA version, but most manufacturers had an EISA version of a card if its capability would be increased by the bus--SCSI and nic were the two most popular.

And even though you technically didn't see an MCA slot (not many of us ever did), you pretty much did see a backwards one when you saw your first PCI slot. The VLB header was also the same connector as the MCA since there were a lot of spares of those around after MCA failed to catch on.