Last Updated on March 15, 2023 by Dave Farquhar
The Intel 486DX-50 is a chip that puzzles many people when they encounter it. Intel released the chip at the end of June 1991, as its top of the line CPU. It wasn’t one of their success stories. Within 6 months, Intel followed up with its 486DX2 chips. The DX2 chips are extremely common today, because they were phenomenally successful. So let’s look at the 486DX-50 versus DX2, and why the DX50 failed.
What was the 486DX-50?
Like its name implies, the DX50 was a true 50 megahertz CPU. There were no shenanigans going on with the bus. The processor ran at 50 megahertz internally, and it had a 50 megahertz bus. On paper, it was 50% faster than the 33 megahertz 486.
Of course, that was on paper. The problem with these high clock speeds was most of the rest of the system ran at slower speeds. The ISA bus, for example, gets cranky when you run it much faster than 8 megahertz. In essence, either chip has to slow down to 8 megahertz anytime it has to use the ISA bus. That bottleneck kept the DX50 from being truly 50% faster than the DX33. It was still a big, noticeable improvement, just not as big as it seemed.
But there was more trouble in paradise than just that.
Problems with the 50 megahertz bus
The problem was the 486 design really ran more reliably at 33 or 40 megahertz than it did at 50. The VESA local bus was still very new, but it had a problem. At 33 megahertz, it could handle three cards. At 40 megahertz, it could handle two cards. Two cards was adequate for most people, because they would want their disk controller and their video card on the local bus. At 50 megahertz, it could only handle one card. So you had to decide whether you wanted a fast disk drive or fast video. You couldn’t have both. Compromises like this are a problem in what is supposed to be a no compromises system. People didn’t spend $3,500+ on a computer to have to make decisions like that.
But even outside of the local bus, the DX50 had a reputation for not being very stable. That’s also a problem. People pay extra for stability. Not instability.
486DX-50 versus DX2: The compromise
When I was selling computers at Best Buy in 1994, people always asked what DX2 meant, because they saw 486DX systems next to DX2 systems on the shelf. The DX2 cost more, so what was the difference?
Most of my co-workers said the DX2 was the sequel, and left it at that. That’s not what it was, but I guess it was better than explaining clock doubling. I’m not sure how many of them would have understood clock doubling enough to be able to explain it themselves anyway. But I’ll give it a try.
The DX2 introduced the concept of clock doubling. The CPU ran at one speed internally, while running the bus at a slower speed. Initially the DX2 came in 50 and 66 MHz versions. The slower bus meant a DX2-50 was not as fast as a DX50. But it was still faster than a DX33, and it was stable. There was a penalty for clock doubling, but the penalty wasn’t terrible. A clock doubled CPU ran about 80 percent as fast as a non-doubled counterpart would.
That 80% efficiency meant a DX2 running at 66 megahertz ran a touch faster than the DX50 did. And it did it without compromising stability. And since it ran on a 33 megahertz bus, you could have three cards on your local bus if you wanted. If you wanted a fast video card, a fast IDE card, and a fast SCSI card, you could have all three, and you would have a no compromises 1992 system. This was something people were much more likely to talk about than actually follow through with, but they liked having the option.
And over time, Intel tweaked the DX2 to get some additional performance back. One noteworthy change was introducing a writeback cache.
The DX2-50 ran at a 25 MHz bus, so it wasn’t the best performer. It was usually faster than a 33 megahertz DX. Usually. But AMD’s then-new 486DX-40 ran at about the same speed. But the 66 MHz chip was a different matter. If you were going to pay extra for a DX2, it made sense to step up to the DX2-66 to get the faster clock speed and the 33 MHz bus.
That’s why the DX2-66 is one of the most common 486s today. It was expensive at first, but the price did come down, especially after the Pentium came out, but even after its days as king of the mountain were over, the DX2-66 remained a capable and practical choice. While its glory days were over by the time Windows 95 came out, it was certainly capable of running Windows 95 acceptably as long as it had enough memory.
The 486DX-50 versus DX2 legacy
The DX2 was certainly much more successful and popular than the DX50. Intel did try speeding up the bus again with the Pentium, and was more successful that time. The Pentium bus ran at 66 megahertz. And of course bus speeds did increase with subsequent chip generations.
But clock multipliers came back with the second revision of the Pentium CPUs and have stayed with us ever since. The internal speed of CPUs scales more quickly than bus speeds, so that’s one of the legacies of the DX2.
The 50 megahertz bus did make a sort of comeback before the 486 completely faded away. The 100 megahertz DX4 was at least rumored to be capable of running clock doubled on a 50 megahertz bus rather than its default of running tripled on a 33 megahertz bus. However, unless you had PCI, you still had the issue of only being able to use one VLB card. While I heard rumors of DX4 systems using that configuration, I don’t recall ever seeing one in person. It was an interesting idea, but not very practical.
The DX-50 in person
I did see a DX50 in person when they were new. My good friend’s dad was a chemical engineer, and bought a DX50 in 1991 when it was pretty new. Not to mention wicked expensive. Of course the system was the talk of the town. And it was mind blowingly fast. None of us had seen anything like it. From what I recall, it didn’t give them any more problems than most other 486s would, but delivering speed and reliability at mass consumer scale proved difficult. Otherwise the DX2 wouldn’t have existed.
But I am one of the lucky few who can say I saw a DX-50 when it was new. I was still a teenager, and didn’t know a lot about PCs yet, so I didn’t appreciate it fully, nor did I understand the system’s limitations. But I can say it looked good at the time. Really good.
But when it came to 486DX-50 versus DX2, DX2s running at 50 and 66 MHz were far more common and practical. And in 1991 and 1992, they seemed nice and fast at the time.
6 thoughts on “486DX-50 versus DX2”
At Apricot we developed a dual processor 486dx server that used a novel Microsoft multi processor o/s. The o/s was configured to just have a main stack and network stack, but was to be a first step toward a true MP system, sadly it was abandoned. We’d got to a beta variant when it got turned off 🙁
THAT would have been cool. I hope someone saved that beta system. That was a rough time in the industry, everyone was trying to pick a direction and some weird and wacky technologies got cancelled at a moment’s notice. I’d heard rumors of such systems but this is the first I’ve heard from someone with firsthand experience.
There was a neat high speed shared cache system to enable both processors to share each others memory/ registers/environment etc. But ultimately there was no O/S support and single processor PC architectures were easier to develop and the Pentium of course was a faster chip. Eventually of course multiple core ICs showed up.
The road that led to multi core CPUs was a fascinating journey. Not many know we were already on that road way back in the 486 era.
Microsoft *did* release a multiprocessor OS for 386s and 486s — LAN Manager 2.0. It was based on OS/2 (back when they were developing it with IBM) and was released in 1990. This was not a client system, but a server OS, which is basically all MP OS’s at the time were (although you could probably use SCO as a client OS). MP 486 systems were very rare and pretty specialized servers (MP 386 servers are even more rare!)
InfoWorld article on LAN Manager 2.0 and other MP systems in development from April, 1990.
Microsoft LAN Manager presentation to an OS/2 user group in May, 1990:
Network World story on the release of the “Multiprocessor Server Pak” for LAN Manager in December, 1990:
Thanks Shane, that was some trip down memory lane and well worth the ticket. Brilliant research! We would have been working on the system in 1989 as OEMs like us had the OS way before and nice to see us get a mention in one of those articles. From my memory, having beaten Compaq to the 486 we were trying to beat them to what was LAN Manager/MP, and we binned it, that may have been because we were overtaken – which we surely were or because we didn’t think it made a material difference. As you say it was an asymmetrical O/S and the jury was out (at least in Birmingham, UK).
At that time we were putting loads of time into running client-server benchmarks on large network systems to determine the best way to tune them. Client-server was the future!
I well remember Phil ‘Bilbo’ Buggins of MS UK, and Phil Rowley of Novell. Now there’s a thing, what happened to Novell…
Comments are closed.