The Top 5 Worst CPUs of All Time @ [H]

daglesj said:
Had a customer bring in a Toshiba laptop running a AMD E1 CPU a couple of years ago. Said it was 'a bit slow'.

Holy....shit. It was bad. Hardly had the horsepower to boot. It ran at 95-100% CPU even at idle all the time. Just no power at all.
Click to expand...

Wow I forgot about those.

Yes they were aweful........
 
DracoDan said:
Missing CPUs that *I* think belong on this list:
  • Intel P4 Willamette - Very hot, power hungry, slower than PIII, socket 423 was replaced after barely a year (no upgrade path!), Initially required crazy expensive and effectively slow RAMBUS, support for old PC133 added later while AMD was already on DDR.
  • AMD Am5x86 - You thought the K5 was bad? Cute... AMD tried to pass this off as a Pentium competitor, it was no coincidence that it used a 486 chipset and socket. We had one of these that ran at 133mhz, AMD was honest-ish by giving it a PR score BELOW its actual MHz speed! The 133mhz part was rated at "PR75". In reality it was about half the speed of a P75 for any floating point, which more and more programs had started to use. Also, since it was 486 based almost all compatible motherboards were limited to ISA (no PCI).
  • Transmeta Crusoe - most people have never even heard of these, they were ultra-low power CPUs used in a handful of laptops (like the tiny Sony VAIO PCG series. Unfortunately they were also ultra-low performance. It wasn't an actual x86 CPU, but instead used "code morphing software" to translate x86 instructions to it's own VLIW architecture. The designers intended to be able to "upgrade" the CPU by updating the firmware (aka code morphing software) but this never came to be. It was great in theory, but it just couldn't keep up with even the lowest end AMD and Intel CPUs of the time.
  • Anything VIA (applies to their CPUs, chipsets, "graphics" adapters, ...really anything they touch) - Do you hate stability? performance? compatibility? driver updates? future OS support? sanity?! VIA is there to help!!!
  • Cyrix MediaGX - To the guy that said the Geode was bad, the Geode was an EVOLUTION of this piece of garbage! Another fake "Pentium", this chip was actually a 486 in drag. To make things worse, everything was integrated into the core (graphics, sound, memory controller and PCI controller). Sure it was low power, but the performance was astonishingly bad, partially thanks to having no L2 cache. Since everything was integrated on the CPU, the only option was embedded.
CPU that I think didn't belong on the list:
  • AMD Bulldozer - Was it great? No. Was it good? ...well, no, not really, but was it decent for the price it was sold at? absolutely! This chip allowed for some very decent gaming systems for dirt cheap.
Click to expand...
Anything VIA (applies to their CPUs, chipsets, "graphics" adapters, ...really anything they touch) - Do you hate stability? performance? compatibility? driver updates? future OS support? sanity?! VIA is there to help!!!

LOL!

Preach it brother! I am amazed they could even get it working, they had a few halfway decent chipsets, except you knew you would get into trouble at some point, one motherboard I had used a chipset that was way behind on USB, I had to find a registry tweak to get Win XP to stop telling me how crappy and slow the port was! Thanks SOYO and VIA!
 
Another CPU list with no mention of architecture, another largely x86-centric list (IA-64 Itanium aside).

I don't feel like it's complete without the PowerPC 970 "G5", at the very least. IBM basically NetBursted it when Motorola/Freescale couldn't deliver a suitable "G5" design, and that's why the dual-core versions needed closed-loop liquid cooling to keep from toasting themselves. Then Intel dropped NetBurst, revisited P6 with Core Solo/Duo, and the rest is history.

ecmaster76 said:
Have you ever tried to use a Geode? Give it a whir and I think you'll be willing to knock Bulldozer down a few pegs. BD couldn't keep up in games but it least it was pretty solid for a budget rig.

Or the NVidia Tegra in the Surface RT. I dare say the early NV CPUs make R600 and Vega look like stellar products. Checking for WIndows updates takes *days* and that is not an exaggeration
Click to expand...
Never touched an AMD Geode, but I really feel like this NVIDIA Tegra 4 is hamstringing my Wacom Cintiq Companion Hybrid in ways beyond a lack of updates beyond Android 4.2.1 Jelly Bean.

-Occasional inexplicable UI freezes, which have been mentioned on other Tegra-based tablets.
-NVIDIA doesn't provide Tegra 4 driver/software support for Marshmallow or later, so developers are on their own. At least they're nice enough to offer up kernel sources.
-The GPU core is a far sight short of the later Tegra K1 and X1, based on Kepler and Maxwell respectively.

DracoDan said:
Transmeta Crusoe - most people have never even heard of these, they were ultra-low power CPUs used in a handful of laptops (like the tiny Sony VAIO PCG series. Unfortunately they were also ultra-low performance. It wasn't an actual x86 CPU, but instead used "code morphing software" to translate x86 instructions to it's own VLIW architecture. The designers intended to be able to "upgrade" the CPU by updating the firmware (aka code morphing software) but this never came to be. It was great in theory, but it just couldn't keep up with even the lowest end AMD and Intel CPUs of the time.
Click to expand...
Oh god, I remember hearing about these! The Compaq TC1000 Tablet PC (a "hybrid" slate with a cool detachable keyboard and docking station that today's hybrids should consider shamelessly ripping off) used a Transmeta Crusoe and a FinePoint pen digitizer, and they hamstrung what was otherwise a really ahead-of-its-time design.

The succeeding HP TC1100 fixed that by moving to a Pentium M and Wacom Penabled digitizer. I still maintain that the TC1100 is one of the best all-time tablet computers to this day, even if it's certainly not the first of the hybrids (a title that most likely belongs to the GRiD Systems Convertible)

N4CR said:
One day I will own an Abit BP6.
Click to expand...
Maybe we can work something out there; I've got a BP6, dual Celeron 533s and at least 512 MB of SDRAM boxed up in storage, and the last time I gave it a run, it worked. Capacitors showed no apparent signs of failure.
 
NamelessPFG said:
Maybe we can work something out there; I've got a BP6, dual Celeron 533s and at least 512 MB of SDRAM boxed up in storage, and the last time I gave it a run, it worked. Capacitors showed no apparent signs of failure.
Click to expand...

Was more of a future wishlist when I have the dosh for it, thank you very much for the heads up - noted and will be in touch should this need further fester :)
 
Dan_D said:
There was no comparison, a Pentium was much faster than a 486 per clock. The only thing that helped is that some 486's ran at 100MHz or more. Even then, the Pentium would blow them away at anything involving the FPU. Quake is an example of this. The Cyrix 486 chips sucked ass. My Cyrix 486 DX2-80MHz was slower than an Intel 486 DX2-66MHz CPU. AMD was a little better in this area, but not by much. Intel was king in the 486 era. AMD's 486 DX4 100 and DX4 133MHz are well regarded, but you were effectively stuck with a 486's platform and they still were only barely competitive with early Pentium processors.
Click to expand...
Didn't IBM have a 486 SLC2 66 as well? I actually had one of those and it sucked compared to my old AMD 386 DX40.
 
Man. I still have some HP Integrity Servers packing Itainums in production. They just won't die. They are still fast. They Rock!
 
ecktt said:
Man. I still have some HP Integrity Servers packing Itainums in production. They just won't die. They are still fast. They Rock!
Click to expand...

The Itanium was good at specific tasks. I didn't list the Itanium over raw performance, but rather it's lack of versatility and market penetration.
 
  • Like
Reactions: ecktt
like this
Dan_D said:
The Itanium was good at specific tasks. I didn't list the Itanium over raw performance, but rather it's lack of versatility and market penetration.
Click to expand...

You're spot on. If your task lends itself well to the compiler spotting parallelism, you were golden. The designers really believed (and probably still do) that the best way to make a modern CPU was to hoist all the "ugly" out of the processor and shove it in the compiler. The chip would be lean and efficient, as god intended!

And it is, assuming you can actually fill the instruction word with meaningful work. In practice, this was really hard to do with normal tasks, so the chip didn't really hit its full potential much or easily.
The more traditional designs may have less raw power, but it is much easier to get their top performance from them. So lower but more attainable peak versus higher but difficult to achieve peak.

The chip itself was/is really slick.
 
The K5, in its original iteration (pre PR ratings and other shite) was actually a very decent processor. The K5-75 model was faster than the competing 6x86-133 most of the time for the stuff I was doing - sounds crazy but it actually was. Few people compared them side by side. The upper range models were not quite as good.

The pentium 60/66 wasn't as bad as people make out. It only affected users who needed .00001 scale precision in floating point stuff when doing divides. Most software was integer based at the time, so unless you were doing financial stuff (which you probably had a pentium pro for anyhow) it really didn't hit you.

Missing from the list is the 286 (intel/AMD) - The 286 actually has protected mode. The problem is that in order to do a switch out to protected mode you need to either:
1. Do full processor reset
2. Use an undocumented, sketchy function to enter/exit. - Basically a test function that resets all registers

What this meant is that memory management and multitasking were hard on a 286. Most users didn't see this though as most software was single task based at the time, and windows hadn't really evolved to the point where it was usable for much.

Chips & Tech actually released a version of the 286 with this fixed 386 style, before the 386 did it.

The mediagx was a flop, for lots of reasons. As was the crusoe (which is unfortunate as I thought it was actually pretty nifty at the time).

The WinChip (IDT) - is largely unknown as another contender - https://en.wikipedia.org/wiki/WinChip - They were actually quite nifty from an engineering perspective as they had much much fewer transistors than competing designs, and ran cooler as a result while doing integer math really quickly. Unfortunately they had worse FPUs than Cyrix chips of the same generation and Centaur was absorbed by VIA (where it festered and rotted into designs like the C3/C7 and nano).

I suppose it depends on your definition of worst - worst in terms of market placement/performance? worst in terms of engineering? worst in terms of support? buggiest?
 
Last edited:
K6-2’s and 3’s. I’ve never seen a system using either of those processors that was 100% stable.
 
acascianelli said:
K6-2’s and 3’s. I’ve never seen a system using either of those processors that was 100% stable.
Click to expand...

One of my retro PC's was running a k6-2 450 for a while and I ended up swapping it out for p3 450. The stability just wasn't there. Opening certain apps would just crash the system. It was weird.
 
auntjemima said:
One of my retro PC's was running a k6-2 450 for a while and I ended up swapping it out for p3 450. The stability just wasn't there. Opening certain apps would just crash the system. It was weird.
Click to expand...

Yup. Very similar experiences from me too.

Me and a friend built 2 Systems at the same time. He went with a K6-2 450 and I went with a Celeron 433 on a 440BX. His ran unstable and mine over locked to 590mhz 24x7 stable. I chose wisely. :)

He had me kinda beat on GPU though. I had a TNT2 Ultra and he had an ATI Rage Fury Maxx.
 
acascianelli said:
K6-2’s and 3’s. I’ve never seen a system using either of those processors that was 100% stable.
Click to expand...

That was largely due to the shitty Super-7 motherboards. They were often based on VIA and SIS chipsets which weren't worth a shit. They had cut rate electrical implementations and horrendous BIOS support. The FIC VA503+ stands out as one of, if not the worst motherboard of all time and many a K6 sat on top of it.
 
Keljian said:
The issue with these wasn't the processor - it was the chipset. Often these were paired with via or ali chipsets which, quite frankly, were full of bugs
Click to expand...

Mine was definitely a VIA chipset. Have an Intel one now, stable.
 
acascianelli said:
Yup. Very similar experiences from me too.

Me and a friend built 2 Systems at the same time. He went with a K6-2 450 and I went with a Celeron 433 on a 440BX. His ran unstable and mine over locked to 590mhz 24x7 stable. I chose wisely. :)

He had me kinda beat on GPU though. I had a TNT2 Ultra and he had an ATI Rage Fury Maxx.
Click to expand...

I had never heard of the MAXX before. Found some benchmarks on toms hardware and wow, that thing was pretty snazzy. Above a 256mb SDR Geforce but short of the DDR one. Higher than the voodoo3 3500 and the tnt2 ultra.
 
I had a K6-2 450 system on an Epox board using a via chipset (from memory) The system was fast, but it was buggy as all get out

I swore off via at that point, which lead to me having an athlon 600 on an irongate chipset, and a dual athlon 1200MP setup on a tyan thunder k7 board - absolutely refused to have via during those generations.
 
BTW I rather like Ryzen, I think it's extremely impressive as a cpu and it's a great compromise for about 80% of desktop users (does this mean I'm a zealot?)
 
Keljian said:
I had a K6-2 450 system on an Epox board using a via chipset (from memory) The system was fast, but it was buggy as all get out

I swore off via at that point, which lead to me having an athlon 600 on an irongate chipset, and a dual athlon 1200MP setup on a tyan thunder k7 board - absolutely refused to have via during those generations.
Click to expand...

I ran an ECS K7S5A board for my 1.4ghz thunderbird. I loved it and could overclock fairly well, considering the shitty cooling in those days. Come nearly 15 years later and I was reading a post somewhere that said it was one of the worst boards of all time. Even if you check speed fan it's in there with a warning lol..
 
Getting off topic, but I managed to pop two 1800XP chips on that board after I accidentally crushed one of the 1200 cores. I had to cut one of the bridges on the top of the chip to make them work, used a power supply with a pair of dressmaker pins to burn them. Watching them vaporise in a flash of light scared the bejeezus out of me (expensive chips at the time)
 
auntjemima said:
I had never heard of the MAXX before. Found some benchmarks on toms hardware and wow, that thing was pretty snazzy. Above a 256mb SDR Geforce but short of the DDR one. Higher than the voodoo3 3500 and the tnt2 ultra.
Click to expand...
The MAXX’s were awesome! But largely were held back by the fact they were dual chip designs and the ATI driver team was complete horse shit back then.
 
Keljian said:
I suppose it depends on your definition of worst - worst in terms of market placement/performance? worst in terms of engineering? worst in terms of support? buggiest?
Click to expand...

Good question. I'm just impressed that someone remembers those chips and their "personalities".

Mucho props to you man.(y)
 
Khahhblaab said:
Good question. I'm just impressed that someone remembers those chips and their "personalities".

Mucho props to you man.(y)
Click to expand...
Hey thanks - I could talk about chips architectures and personalities all day and night, it's kinda something that I have an interest in - should have studied CE ....
 
I'm just gonna respond to all three of you at once, given my own K6-2 350 experiences:

acascianelli said:
K6-2’s and 3’s. I’ve never seen a system using either of those processors that was 100% stable.
Click to expand...
Keljian said:
The issue with these wasn't the processor - it was the chipset. Often these were paired with via chipsets which, quite frankly, were full of bugs. ALI chipsets weren't that much better, but they worked.
Click to expand...
Dan_D said:
That was largely due to the shitty Super-7 motherboards. They were often based on VIA and SIS chipsets which weren't worth a shit. They had cut rate electrical implementations and horrendous BIOS support. The FIC VA503+ stands out as one of, if not the worst motherboard of all time and many a K6 sat on top of it.
Click to expand...
So, wait - there's boards that make that PC-Chips M598 with its SiS 530, WTF-inducing header layout (who the hell puts AT port headers between expansion card slots?) and apparent inability to run 100 MHz FSB with any stability whatsoever that I had to suffer with as a kid whose father built this for him not look like a total piece of shit by comparison? Cripes... I know there's a lot of retro hardware people want to salvage, but there was nothing about that system that warranted salvation from a recycler save nostalgia, which wasn't strong enough for my liking.

Meanwhile, I hear that Super Socket 7's going up in value all of a sudden, along with every other vintage piece of computer hardware. I don't really see why when it's not that hard to cobble together a nice Socket 370 build for Win9x gaming, or you could go the crazy route like me and go Socket 478 with an industrial board to keep a real ISA slot or two handy.

If anything, half the reason motherboards are kinda boring now is that modern CPUs took up a lot of the old northbridge functionality (memory controller and PCIe lanes in particular), and they're generally stable platforms. There was much more variation back then between the cream of the crap and a solid platform like the famed Intel 440BX that even Intel couldn't provide a worthy successor to for ages. Hell, it wasn't just an IBM PC-compatible problem, either; Performa 52xx/53xx/62xx/63xx, anyone?

At least things turned out better for AMD going into the Athlon age and nVidia's nForce chipsets. Remember when those were good, especially nForce2 with SoundStorm?
 
A curiosity for those following - one of, if not, the best (most recent) processor that has been made imo is the i7-5775C. On the desktop, when compared with anything of the same clock it is more efficient in terms of performance/watt (except maybe Ryzen at low clocks). That includes all the Lakes. It was also Intel's least buggy chip released in the last 10 years. Wasn't very overclockable, but it was seriously fast and efficient for what it was.
 
Last edited:
NamelessPFG said:
At least things turned out better for AMD going into the Athlon age and nVidia's nForce chipsets. Remember when those were good, especially nForce2 with SoundStorm?
Click to expand...

Obviously you never tried to run a real soundblaster on one! - They had issues running anything that was slightly out of spec, and the drivers weren't updated regularly enough - which meant that the sound was perpetually glitchy. One particular MSI board (which was called one of the best) had overheating issues, and the heatsink on the chipset had really poor thermal paste. In the case of my dad's pc it nuked one of the memory slots as a result of an overheat, and literally left burn marks on the board ...

"Hey dad, what's that burning smell?"
 
For the sake of those here and based on the feedback - I've messaged Kyle re me potentially writing some articles about older hardware in the forums (cause tbh I don't know where they'd fit). just want to share the love.
 
Keljian said:
Missing from the list is the 286 (intel/AMD) - The 286 actually has protected mode. The problem is that in order to do a switch out to protected mode you
Click to expand...

Even without the crappy protected mode, the chip had other merits.

It was about twice as fast as the 8086, and several times faster than the castrated 8088.

The independent memory bus lines and no-longer shared adder meant the chip could do a lot more processing in a single cycle, and could bump up the clocks.

The 6 MHz 286 that shipped in the PC AT could do 0.9 MIPS.

The 8086 at 4.77 MHz (IBM PC clocks) could only do 0.33 MIPS! The 8088 was even slower!

CLOCK-FOR-CLOCK, the 286 was TWICE AS FAST. That was a much bigger performance improvement than we saw with the 386! WE didn't see that level of performance improvement again until the Pentium Pro!

The protected mode was just a checkbox. It still used indexed addressing just like it's predecessor. Intel didn't get that right until the 386.
 
Last edited:
Keljian said:
Obviously you never tried to run a real soundblaster on one! - They had issues running anything that was slightly out of spec, and the drivers weren't updated regularly enough - which meant that the sound was perpetually glitchy. One particular MSI board (which was called one of the best) had overheating issues, and the heatsink on the chipset had really poor thermal paste. In the case of my dad's pc it nuked one of the memory slots as a result of an overheat, and literally left burn marks on the board ...

"Hey dad, what's that burning smell?"
Click to expand...
Cripes, you really had it bad... there was a system I had with an Asus A7N8X-E Deluxe for a brief while that worked more or less fine, though, barring a certain set of RAM that needed a modded BIOS to set command rate 2T on the RAM in order to remain stable. Two 1 GB DDR-400 DIMMs, to be specific.

Needless to say, it didn't kill any components from overheating - not that I ever pushed it as hard as I would have if it were my best board, but it was literally a freebie given to me around the time I was building up my Q6600 system a decade ago.

That said, my Sound Blaster cards went to other systems; my Q6600 box ran an Auzentech X-Fi Prelude, Auzentech X-Fi Forte and Creative X-Fi Titanium HD in that order. Let's just say I'm fully aware of the driver jank along the way, and despite that, I still keep the X-Fi Titanium HD in my 4770K build because I do actually have a use for hardware OpenAL and ALchemy. I also had a Sound Blaster Live! Value as a holdover from the K6-2 350 system, but that thing was showing its age at the time and wouldn't be of much use at this point unless someone really wants to use kX Project drivers.
 
defaultluser said:
The protected mode was just a checkbox. .
Click to expand...

First - Do not yell at me.

Second - Protected mode is the foundation of all modern computing. It fundamentally underpins multitasking in all x86/64 operating systems. If you think it was "just a checkbox" then maybe you think vector instructions are just a checkbox too :)
 
NamelessPFG said:
Cripes, you really had it bad... there was a system I had with an Asus A7N8X-E Deluxe for a brief while that worked more or less fine, though, barring a certain set of RAM that needed a modded BIOS to set command rate 2T on the RAM in order to remain stable. Two 1 GB DDR-400 DIMMs, to be specific.

Needless to say, it didn't kill any components from overheating - not that I ever pushed it as hard as I would have if it were my best board, but it was literally a freebie given to me around the time I was building up my Q6600 system a decade ago.

That said, my Sound Blaster cards went to other systems; my Q6600 box ran an Auzentech X-Fi Prelude, Auzentech X-Fi Forte and Creative X-Fi Titanium HD in that order. Let's just say I'm fully aware of the driver jank along the way, and despite that, I still keep the X-Fi Titanium HD in my 4770K build because I do actually have a use for hardware OpenAL and ALchemy. I also had a Sound Blaster Live! Value as a holdover from the K6-2 350 system, but that thing was showing its age at the time and wouldn't be of much use at this point unless someone really wants to use kX Project drivers.
Click to expand...

I've built so many systems.. and made so many stuffups and seen so much weird stuff... I should really write a book :)

I've used the kX project drivers. I've had issues getting a new model X-Fi working on a z97 board. I've had a passively cooled q6600 at 3.2ghz (that was really interesting...) ..

I've cooked a 486dx2-66 in a store by user error - not to mention watched a hard drive go up in flames because a family member shaved a molex plug with a knife to make it fit the wrong way...
 
other than the 6x86 being slow in games , it was stable and could run without a cooler :D
could never compete in games with intel P1 , just as K6-2/3 did with P2 . Never had stability issues with them also
 
Keljian said:
First - Do not yell at me.

Second - Protected mode is the foundation of all modern computing. It fundamentally underpins multitasking in all x86/64 operating systems. If you think it was "just a checkbox" then maybe you think vector instructions are just a checkbox too :)
Click to expand...

I think he's getting at the fact that on a 286 it was pretty badly broken, not that the idea itself was bad or useless.
 
How about a top Five best and worst chipsets of all time? nforce would be on top best of the list and SiS/Ali battling for worst ever.
 
Keljian said:
First - Do not yell at me.

Second - Protected mode is the foundation of all modern computing. It fundamentally underpins multitasking in all x86/64 operating systems. If you think it was "just a checkbox" then maybe you think vector instructions are just a checkbox too :)
Click to expand...

Sorry, I'll stop shouting.

Fair, but the "broken" protected mode switch didn't matter for consumers, who couldn't scrounge the memory for a multi-tasking protected mode operating system until the 1990s. And that point, the antiquated 16MB memory limit and the lack of virtual x86 mode made it tough to use.

The first *successful* use of the 286 MMU was Novell Netware in 1986, a system that would not be affected by the switch issues (being a server and all). Consumer experiences using OS/2 were poor because memory was too expensive for consumers.

I'm just saying, it was mostly a check-box. By the time memory prices tumbled, there were better second-generation processors for cheaper and better third-generation operating systems.

if the 286 had not been twice as fast as it's predecessor, it would have been a failure even if they had protected mode working perfectly. That massive performance increase is half the reason companies started to build servers around the 286! To me that's the very definition of a "check box" feature - useable enough for developers to get their feet wet (and start moving the industry forward), but not useful for the mainstream.

I'd put the 286 protected mode in the same class as Haswell's AVX2 Gather. the (future accelerated) instruction has the potential to revolutionize auto-vectorizing compilers everywhere, but the first generation has no hardware support.
 
Last edited:
I was a Super 7 enthusiast, and I loved my K6-2 and K6-3 processors. But working in a computer store, I also learned fairly quickly which boards to avoid (like the FIC-503 garbage). Personally at the time, I preferred the Tyan S1598 Super 7 ATX boards. Those were as stable as you could get on Super 7 and made for smooth sailing. You still had to regularly update the VIA chipset drivers, but otherwise fine. Shuttle also had some decent boards (back when they made motherboards...)

It should also be noted that some K6 and K6-2 (the ones with the 66MHz bus speed) processors would work on Intel Socket 7 mainboards, and they were rock solid there as well. As has been said, the problem was NOT the CPUs...
 
You must log in or register to reply here.
Back
Top