It’s fair to say that Moore’s Law is not delivering on its promise of advancing semiconductor capabilities as fast as it used to, as the limits of current fabrication techniques are being met. Where this is being written for example there are two laptops, one from the last year and one that is 11 years old, and while the new one is undeniably faster it has not overtaken the other by as much as a ten year gap between 1990s machines would have revealed.
So with older laptops being still so relatively quick, what possible attraction could there be for working on a machine from the 1990s, when the Moore’s Law curve was steeper? It’s something [Jim W] is doing, with his HP Internet Advisor (J2522B), and when you see the machine in question perhaps you’ll understand why. The J2522B is a laptop, but it’s no ordinary ’90s road warrior’s status symbol. This 486-powered beast is a piece of test equipment, specifically one for examining Ethernet ports, thus it’s built like a tank and is mains powered only. It boasts a 486DX4, 16 MB of memory, a then-colossal 1.3 GB hard drive, and an ISA Fast Ethernet card. Oh, and WIndows 95, which with a couple of decades’ hindsight seems an amusing choice to power a piece of security test equipment. Impressive specs for the day, but the $20,000 price tag would still have been steep compared to a comparable laptop.
[Jim]’s machine is destined for classic gaming, though with only the little HP pop-out mouse you saw on their Omnibook range at the time, he needed a PS/2 port. Some chipset hunting found that, but at the cost of accidentally frying a MOSFET when a screen connector was incorrectly re-inserted. We’re then treated to a guide to substituting older MOSFETs with modern parts, useful in itself, but followed by a marvelous piece of bodge work as an SOIC-8 part is placed on a DPAK footprint.
This is an interesting series of posts, partly from a retro angle as they deal with an interesting machine, but also from a hacking angle as he’s getting closer to the vintage PC hardware than most of us to. Keep an eye on it, there is sure to be more in the pipeline.
Waiting for gaming on a CNC machine.
the framerates will be measured in hours
Dual spindle machines, each one with a pingpong paddle mounted in a tool holder.
Most modern enough CNC machines just run on commodity computers anyway. Maybe with some real time modules or downstream controls depending on the type of CNC machine and the specific requirements it may have and when it was built.
Hah Ha, one could cnc their path around an RPG like Skyrim or Zelda perhaps? :D
By the time I got out of the machine tool industry ten years ago, CNC controllers were starting to have Windows front ends to them. The challenge was PREVENTING operators from gaming on them.
On the website he says “what to do” when looking to the insanely common 2955, cause digikey doesnt have stock, my first thought was findchips, where mouser has 4200
but yea lets bodge it cause …. digikey or something …
DigiKey is a next-day delivery for me here in Minnesota, no matter what level of shipping I select. I tend to not think of other suppliers.
Besides, the bodge was more fun. :)
I have the same situation with Mouser, living north of Houston, TX, and them being south of Dallas.
I miss the IBM PS/2 MCA servers. Those were build like tanks, and the guts looked so much more better engineered than the PCs of the time.
I´ve also had some 386SX servers with ECC RAM (16 or 32MB) and 387 copro running win95 faster than a 486 at almost twice the CPU clock speed, impressive!
Most people think Moore’s law had to do with speed, and was universal for things other than silicon chips, which of course is utterly wrong on both counts. It’s about how good the photolith and process are – how many transistors can I pack into a square area.
Speed stopped going up years ago. How long have we had effectively a sub-5Ghz limit (if you don’t overclock in LN2 or something)? In fact, most CPUs now run around 1.6 ghz in portable stuff when not in some kind of burst “turbo” mode.
Effective speed kept going up for awhile as more transistors could be brute-forced into speculative execution, branch prediction, various caching schemes and so on (for which we’re now paying in terms of bugs) – but the basic speed hasn’t gone up for quite some time; oh, the silicon transistors by themselves are pretty quick if I read the literature right…maybe 10x or more the clock speed, but other factors, like charging the wire capacity up and down, leakage and heat…slow things back down to what we see.
Memory, despite denials by Intel and others, has been a bottleneck for a long time. I don’t care about your burst rates, look at the latency and in the clock cycles for the memory (always far slower than the cpu).
Bad software (for some values of bad) has used up a lot of the good we had. Ed on CPM on a 4 mhz Z80 wasn’t all that bad, but all it had to do was put a 0x41 in the right memory location to get an A on the screen, not draw a picture of an A in some font in a z layered window – just one silly example. Honest, it compiles fine as ascii (or gawd unicode, there goes readability).
A little OT but the misconceptions of Moore’s law – even now as we “run out of room at the bottom” drive me nuts. Nope, you’re not going to find a magic battery setup for my E car that has 1000x the energy density, ever, unless you give up chemistry and go nuclear – despite what many think, the periodic table is kinda figured out…that’s just one example, but the gee whiz futurists who don’t know how things work and simply extrapolate endlessly are lucky they are so easily forgotten and thus not always discredited – need to go and actually learn some physics. Much that is hyped now is simply not possible. At least some decades ago the hyped stuff, while it was unlikely and mostly didn’t happen, was possible at all.
And FWIW, the new miracle materials or metamaterials ain’t gonna fix this till we learn how to put say a billion transistors made of them down at once with something a lot like photolithography. Nope, self assembly makes crystals, not CPUs.
Sure, you can take a week and make one really quick switch by hand….good luck with that.
I was always under the impression that it was a goal more than a predictor.
One big reason Moore’s law is running out is the diminishing returns in that every increase in performance is more expensive.
A fab with cutting edge process can cost over 10 billion now.
Rockets such as F9 and Antares have been developed for much less.
From the days when companies had “exactly what it says on the tin” names. I can picture the setting as a bunch of chip designers and engineers try to come up with a company name…
“What do we make?”
“Well, chips and… technologies.”
“So how about that for a name?”
“What? Chips and Technologies?”
“Perfect! It tells the customers exactly what we do!”
Same script for VLSI, Very Large Scale Integration. It’s what they did. No confusion like people calling ONEAC ‘oh-ne-ack’ until ONE AC officially changed their name to be pronounced ‘oh-ne-ack’, then of course at the first trade show under the new pronunciation, someone walks up and asks “Is this the ONE AC booth?”
Bought an HP ML530 recently. Stuck free nas on it, and can happily max out the 100Mb network card – it barely flinches. Next step, an sfp card and some link aggregation. Shame it’s so power hungry, but for 50 quid all in you can’t have it all…
PS can we have a HaD regional auction board please, I have a garage full of junk that isn’t worth much but would appreciate a proper home. Ebay is just a vipers nest…
Put it up on the Tindie flea market maybe?
full of junk in what sense?
Old servers, interesting outdated hardware – eprom programmers, KVMs, diskless terminals, non-functioning consumer stuff that is worth stripping down only I never get round to it, that sort of thing.
Well, Tindie isn’t an auction site, but last I checked, they did have a Force Ale, uh, Foresail, uh, Fours sell section.
“what possible attraction could there be for working on a machine from the 1990s, when the Moore’s Law curve was steeper?”
The above machine is probably the approximately the fastest PC to come with a processor that connects to a bus that you could realistically implement yourself with off-the-shelf components. You could build a PCB to support a 486 just using low-cost CPLDs to build the primary components, and wouldn’t need anything more exotic than a ~50MHz oscilloscope to debug it. The 30ns cycle time & 3ns setup/hold times for various events are just about within range of what you can achieve with a CPLD, which means you can build custom hardware using a 486 processor without spending a fortune on it. If you’re struggling to manage the timing to run at full speed, you can drop it down as low as 8MHz and it’ll still work.
The Pentium changed that — it was designed from the beginning to primarily work with PCI (rather than extended, faster variants of something similar to ISA, which is what the 486 was designed for), and it requires timings with an accuracy of ~1ns to work correctly. Its bus works with pipelined transactions, rather than being synchronous. It is much harder to interface with. And its minimum bus speed is 25MHz. You’d be looking at a relatively recent FPGA to manage the interface, if you couldn’t get purpose built chips, and the cost would be substantially higher.
What are some other sites with similar dissecting-repairing-hacking old electronics content?
I’m already reading Ken Shirriff’s blog (www.righto.com) and the Living Computer Museum engineering blog (https://engblg.livingcomputers.org)
Has a somewhat working jukebox core here, with a parallel port code key, Celeron D (I think) and 128MB RAM. And this one ran Windows 98!
Needless to say it needs some TLC but keeping it just in case one day I can use it for something that needs eight serial ports.
It will take a P4 and 1GB RAM so not completely useless yet.
“…it has not overtaken the other by as much as a ten year gap between 1990s machines would have revealed.”
How do you get a 10 year gap between machines that are both from the 1990s? Time travel?