Hacker News new | ask | show | jobs
by 60654 1771 days ago
> That’s why the 386SX had a pretty short shelf life. Once AMD had 386 chips to sell, Intel cut prices on 486s. But for a couple of years it served a purpose. And the chip lived on as a budget option for a couple of more years.

TBH that kind of a short shelf life wasn't just a 386 thing. Clocks speeds and architectures were advancing quickly, and all chips had a really short shelf life.

For example, in the span of 5 years (say '91-96) you could upgrade from a 386SX 16MHz to a 486DX 50MHz to a Pentium 90MHz, each time paying about the same amount of money but getting a 3x speed-up. And other components like video cards were improving just as quickly.

People were upgrading every couple of years because the difference between older and newer models was night and day. Imagine if in 2015 you bought an Intel i3 3GHz and this year you could buy an i7 15GHz with 8x the RAM for the same price.

11 comments

Around that time I was building "high end" PCs for friends and friends-of-friends. I remember the average price stayed around $2500, but the component availability iterated pretty quickly in terms of processor, HDD size/interface, graphics, standard RAM, etc.

A 2 year old PC felt woefully underpowered in that era, and a 4 year old PC was almost useless if you wanted to use any "current" software. You'd be out of drive space, unable to run a lot of programs/games, and limping along.

Now, my 8 year old Macbook Air is still more or less as functional and useful as it was when I got it.

> Now, my 8 year old Macbook Air is still more or less as functional and useful as it was when I got it.

This is my frustration with Apple's policy of dropping support for hardware in MacOS. It made sense in the 90's to upgrade every 2-3 years because you got 1.5-3x more performance each time. So 6 year old hardware was almost an order of magnitude less capable.

Fast-forwarding to today, a "legacy" 10 year old ("Mid-2011") MacBook Pro supports just as much memory (16GB) as Apple's current M1 offerings. The M1 does put up some very impressive numbers on the single-thread CPU front, but that's because we've gotten used to such small progress every year--it's only about 2x the speed of the 2011 MBP for single thread tasks.

> It made sense in the 90's to upgrade every 2-3 years because you got 1.5-3x more performance each time.

You still get that level of performance increase. It's just that it's to the point of being imperceptible, or invisible, to the vast majority of users. The only ones who really notice are the ones really pushing the machines with things like 3d rendering, 12k video production, and high end audio (I can run 80 plugins now where I could only run 70 before!).

Texting? Doesn't matter how fast your machine gets, you won't notice. It will perceivably perform just as good as a phone that texted 20 years ago. You can run benchmarks and see, oh yeah, this is faster, but it's not noticeably faster to your human inputs.

In the days of 300 baud modems, they were so slow you could actually read text as it was sent over the line, as if someone was just a very fast typist and you were watching them. Now, you get full multiple pages of text with videos that pop up almost instantaneously. Once we got to 1200 baud, it was too fast to read as it was coming over the line. You couldn't catch up. Ever since that threshold was crossed, it doesn't matter how fast speeds get, they're all faster than the human brain can absorb so they are basically the same in our minds.

Except for the web. Which seems to get slower and slower due to bloat and overload, no matter your computer or internet speed.
Isn't the corollary then, that if we didn't upgrade our computers, the web would get faster, or at least not get any slower?
Yeah, it's great, but it's also less fun.

Our 9-year-old Air went to live with friends who suddenly needed another computer (for their kid) at the beginning of COVID. It still works fine, though some sites are slow.

Sure, APPLE isn't releasing updates for it anymore, but that doesn't actually affect whether or not Word runs, or whatever.

I'll also say that biggest and most dramatic upgrade I ever had was moving from a 1988 AT clone to a 1991 386/33 (Gateway 2000, baby). We'd pick a directory with loads of files in it and do a dir just to watch it scroll by so insanely fast. Simpler times for sure.

The next most impressive upgrade I ever had was about 10 years ago. I had a 2010-era Macbook Pro (early Intel) that shipped with a spinning drive, back before everything was SSD. At some point in its life -- it was ultimately stolen in 2012, so call it 2011 -- I swapped the spinner for an SSD and OH MY GOD THE DIFFERENCE.

No other machine upgrade in my 30 years of computing has come close to the "holy shit!" moments of these two. I wonder if my ultimate shift to Apple Silicon will bring some of that -- I hear good things, so I hope so.

> I swapped the spinner for an SSD and OH MY GOD THE DIFFERENCE.

FWIW, that difference isn't nearly as noticeable under Linux. It's so lightweight on RAM and so good at caching frequently-accessed data that it can be incredibly snappy even when using spinning rust. SSDs mostly speed up your boot process and the rare IO-heavy workload. Though it wasn't until the late 2000s-early 2010s that RAM began to be truly abundant on out-of-the-box configs, and that was the same timeframe as the switch to SSD's.

I strongly disagree, especially when running RAM-heavy applications where background tasks end up getting put in swap to make room for caches. For quite a while, there was actually an issue where the OOM-killer didn’t get triggered because swapping to the SSD was so fast that certain timers in the OOM system would never trigger. Instead of recognizing that the system has become completely unusable and killing off a process, it would just sit and thrash uncontrollably.

If I recall, Ubuntu 16.04 still suffered from this. And Android Studio could very reliably trigger this situation on my older/upgraded early i7/16GB RAM/SSD laptop. Hard power cycle to recover.

I totally disagree.

When I was preparing to emigrate in 2014, I put an SSD and a big HDD in my desktop-replacement laptop, and the transformation was a revelation. Ubuntu 14.04 went from booting in only a minute or so to booting in a single-digit number of seconds. It was astonishing.

> my 8 year old Macbook Air is still more or less as functional and useful as it was when I got it.

Right? I remember the enormous performance jumps from 286 to 386 to 486DX2 to Pentium to P6S...

Today, I'm still using a late 2013 MBP. Other than the lame 128GB of disk space, it is still my primary machine and is fine for everything I do. Most of my work is done in the cloud anyway so its essentially a 1970's dumb terminal.

My i7 desktop from 2011 is still strong, but I was in the hospital ICU last year and promised myself I would build a zen 3 desktop. It's a bit more stable, and the pci-e v4 NVME storage is awesome, but otherwise it's pretty much the same experience. Games play about the same because I moved the same GPU over. The CPU bound softwares I tend to use are also single threaded, so the fact that I have 4x more cores doesn't change much other than let me run lots of processes. In practice, my ability to multitask is limited by UI bugs rather than raw compute power.
I've been using my 2012 desktop without any discenible issues. The only thing I can't do is edit 4K video and for that I use my 2019 Macbook. I've been toying with the idea of upgrading my desktop but I have no compelling reason, and the idea of reinstalling all my applications (I'm still on Windows 7) makes it really unlikely. To be fair though, I don't play games on it.
To some degree I imagine there’s also a question of how much demand we put on these modern old computers that reflects changing/diminishing interests as we age. There are certainly workloads available today that would cripple an 8 yo computer (Mac or PC), workloads that a younger version of you may have been more interested in (multiple VMs, gaming, real-time video processing at same time). You probably stressed your 386 to the breaking point within three years of buying it, and you could easily load up an 8yo computer today to its breaking point. My 2016-era laptop is no longer useful to me as, well, anything except an emergency backup.

If one can’t discern issues with a four-five yo computer, I would very humbly suggest it also says something about the demands of the owner stultifying to a degree.

Eh, bullshit. Go try running Max Payne from 2002 PC on a 486.

Then, show me the equivalent of a 2015 game on par on the gap to a game from 2021 as a 486 game like Doom compared to a Pentium III game like Max Payne.

My life doesn't revolve around games enough to do that, but I can tell you that running an IDE, a stack of VMs, driving external monitors, screensharing while on VC, along with misc productivity apps was not doable on my skylake laptop, whereas the same workload is easily doable on my tiger lake platform. How's that?

> eh, bullshit

Precisely what's bullshit? You're telling me that your workloads on modern computers are stressing them out and you simultaneously don't see a difference between a 6th gen platform and an 11th? No, you don't see a substantial difference in modern platforms because you don't have a need for the performance gains.

That's just better perforanmance on parallel tasks, not by raw capacity.

My point stands.

The single-thread performance of my 10-year-old Sandy Bridge is only now reaching the magic half-as-fast point relative to the fastest CPUs available. That's the point, historically, that's prompted me to upgrade.

For those whose workloads aren't multicore-intensive (or GPU-intensive, which is pretty much the same thing), it's been a dry, boring 10 years in the PC business.

> and you could easily load up an 8yo computer today to its breaking point

I've seen people load up a 200 machine cluster on AWS doing quite mundane tasks.

Information complexity is the only physical quantity that doesn't obey conservation laws, so this kind of thing really isn't impressive or interesting.

In real life, a few years ago, people at work fired up a 20 host cluster on AWS to do something that I later replaced with a script and GNU parallel (at the time a perl script), running on my 2013 Macbook pro i7, beating it by 10X.

Sure, I'll admit it may not be fair to compare, considering the AWS host type, copying the data out and in, and all the other overhead.

That said, the project sure looked like it was resume driven 'big data'.

I had exactly that 386SX as my first CPU ever, and I recall the incredible speed boost you'd get each time you upgraded. It was like magic, each time you or a friend got a new machine, everything would be so much faster.

Something similar happened with graphics cards, each new generation made stuff look that much better.

These days I can still use a 2013 Macbook to play MineCraft, doesn't feel any different. Compiling code probably is different, but most everyday things would not be much different.

Oh and of course an obvious question to go along with the whole 90s CPU story:

https://www.maketecheasier.com/why-cpu-clock-speed-isnt-incr...

Speaking as someone who just upgraded from an early 2014 MBA 11" to a Air with the M1 chip, while the old machine always felt adequate to my uses, having used the M1 for a while now, I could never ever go back.

Programs start much faster, quit much faster, web pages load much, much, much faster. Compiling is anywhere from 4x to 10x faster, as is interpreter startup time for scripts.

Now, granted, it's a jump from one architecture to another and a time span of 7 years, but this upgrade felt magical the way the mid-90s hardware upgrades felt.

You can still use a 286 to play King’s Quest, too. A 2013 Mac is in no way an adequate gaming machine in any sense of the idea. You couldn’t still use a 286 to play Falcon 3 or Comanche back when they were new, and you can’t use a 2013-gen tech stack to realize full potential of today’s games. The benchmark of playable Minecraft is not a suitable measure for capability, only a reflection of unchanging demands since 2013.
You can run King's Quest I-VII (and countless other classic adventure games) via SCUMMVM [1]. Falcon 3 and Comanche can be run via DOSBox [2]. That 2013 Mac can easily play almost any game from the DOS era using one of these two fantastic open source projects.

[1] https://www.scummvm.org

[2] https://www.dosbox.com

The point is that old hardware sucks if trying to facilitate modern software. Not the other way around. A 2013 anything (mac or pc) will be a poor performer with today's games.
Back then a new PC was so exciting because upon first boot you noticed it was significantly faster than the machine it replaced. I haven't experienced that from new computers in decades now.
As well as harddisk space. The feeling of just taking the entire old disk and putting it in an "old hd" folder, taking up a tiny corner of the new disk was awesome!
I recently upgraded a computer built in ~2013 to one with 2019-2020 components. Maybe it's because I went from lower-middle tier components to upper-middle tier, but I noticed a very significant performance boost: my NVMe drive boots in seconds (versus ~1 minute with my SATA SSD), and I can build large Rust projects nearly instantly without breaking a sweat (my old AMD FX CPU would turn into a radiator).
Magnetic hard drive to SSD gave the same type of boost...
Oh yeah, forgot about that one. Good call out.
On a 386, MS Word would make you sit around for 10-15 seconds looking at the splash screen. Then when you upgraded to a 486, you opened Word and it was like "bing!", you're up and running.

Now with magnitudes more speed and memory available, loading Word is...somewhere in between.

I run Word 97 under WINE on a decade-old Thinkpad running Ubuntu.

Opening it is a joy... what was a slow, lardy app when it was new is now lean, mean and fast.

In decades? The switch from mechanical drives to SSDs didn’t offer any noticeable improvement? I don’t upgrade every year but moving from an 8th gen proc to 11th and back again presents a pretty stark contrast.
This is part of why I'm going to rush to the Apple store the moment they announce an Apple silicon 16" MBP.
There was a point in time where the absolute best dollars/performance ratio was the AMD 386DX/40, which ran circles around the Intel 386DX/25 and 386DX/33, but was priced the same or less.

And was considerably less expensive than a very top end ($2500-3500 in 1992-1994 dollars) desktop built with something like a Pentium 60 or 66 MHz.

Inflation calculator tells me that a $2500 desktop PC in 1993 would be the same as about $4700 today. For 4700 you could build a real beast of a machine.

> There was a point in time where the absolute best dollars/performance ratio was the AMD 386DX/40, which ran circles around the Intel 386DX/25 and 386DX/33, but was priced the same or less.

I remember when the best price/performance was a 300 Mhz Celeron A that you could overclock to 450 Mhz on an inexpensive A-bit BH6 motherboard. Paired with a 3dfx Banshee, and I remember being able to build a respectable gaming rig for under $600.

These days, $700 will barely get you a GPU, even at MSRP.

If you are going to compare what you can purchase today vs then, you should be comparing what you got for $600 vs what you can get for about $1000 today.
In the around 1000 dollar point today, if you set the design constraint to 1080p gaming, you can do quite a lot with a $175 CPU in a $145 motherboard, add maybe another $150 of RAM, and a $100 NVME SSD. The problem is the video card availability and marketing pricing.
Fair enough, I suppose, but seeing as it's over 20 years ago, I'm having a hard time deciding where those parts would fit. I figured those were all mid-grade parts, and today, a mid-grade GPU would be an RTX 3070 which has an MSRP of $500-600 depending on the brand, and a midgrade CPU (like a current gen Ryzen 5) will be nearly $300, that doesn't give you much left for the motherboard, RAM, storage, etc...
I think they're suggesting you adjust dollars for inflation.
The AMD 5x86/133 was a similar kind of situation. Drop one of those into a decent 486 board w/ some L2 cache and you got better than 75Mhz Pentium performance at a ridiculously low price point as compared to a new motherboard, CPU, and RAM.
And again with the Athlon XP1800+ then Opteron.

AMD has had moments where it really stuck it to intel but always fell back to 2nd, I hope this time it sticks.

x86 vendors duking it out while Apple keeps both honest isn't a bad market for a buyer.

AMD wasn't even in the picture at that time really. Meaning their marketshare of desktop pc's were so low that no one had them. Back in those days it was Intel vs Cyrix.
AMD sold a ton of 386 and 486 processors and held much more market share than Cyrix. They were also very successful with their K6 and K6-2 Pentium competitor processors.
AMD absolutely was in the picture - they sold a ton of high-speed/low-cost 286 CPUs. In the mid and late 80s there was such a thing as a 286 12 MHz which sold for the same price as a much slower Intel part.

I'm referring to the whole time frame before the Cyrix 5x86 and similar were even a thing... There were plenty of AMD 286 and 386 CPUs sold in the early 1990s.

AMD sold lots of K6 and Athlons.
uh.. I actually worked at intel in these times... and I used to have a cube adjacent to Andy Grove (for some reason, we were on the same bathroom schedule, and peed quite a bunch next to eachother)

Anyway, my best friend and I ran the DRG game lab (developer relations group) - where we (intel) paid millions to gaming companies to optimize their games to the intel arch and things like SIMD instruction sets... we game tested (subjectively) games running on intel vs AMD machines... it was also the lab where we were able to prove that a subjectively performant PC could cost less than $1,000 === The Celeron Processor

I was even the person who first sent an email (1997) to engineering asking why we couldnt stack multiple processors on top of one another....

I later learned on a hike with a head of marketing that in the proc labs Intel had a 64-core test fab. (This was fucking 1998 when that was revealed to me under NDA etc...)

---

Nice. Were you responsible for 1999 Comdex Fall Rage Software Dispatched SSE demo used to manipulate benchmarks for a couple years by Anandtech (https://www.anandtech.com/show/260/9)? https://www.hardware.fr/articles/95-5/jeux-optimises.html

"looking closely at the demo, we can see - as you can see on the screenshots - that the SSE version is less detailed than the non-SSE version (see the floor). Intel would it try to roll journalists in the flour?"

Two other examples from that link are SSE "optimized" Rage Software Expendables running slower than original, and "Pentium III optimized" PowerSlide by GT Interactive with zero fps difference. At least those two never released, unlike "Designed for Intel MMX" POD from 1997 with one optional audio filter added and 1/6 of the box covered by Intel marketing.

One has to wonder how much was Intel bribing for such blatant lies and manipulation back then.

I was not, but I can tell you that Intel gave companies $1 million for "Optimized" games for marketing such..
I miss those days :-).

The original author missed that the 386SX had the memory addressing models that the 386 had in addition to the backward compatible 286 modes. So you could access your "Lotus eXtended Memory" much more quickly than you could on a 286 based DOS machine (aka the PC/AT). A lot of businesses that ran on large Lotus spreadsheets used the 'SX for just that reason.

And then in '98 the famous Celeron 300A[1] came out, which could comfortably be overclocked to 450 MHz and remain rock solid stable with a cache running at the same clock. It was an incredible time to be a home computing enthusiast.

[1] https://www.anandtech.com/show/174/3

Yeah, getting 3 years out of a machine in 1990 was about all that was possible. I had an AT clone (so 286) that I took to college as a freshman in 1988, and had the fastest machine in the dorm by a SIGNIFICANT margin.

Three years later, I bought a 386/33 because the 286 was, by comparison, dog slow. And it cost less than the AT had.

This was both awesome and frustrating. I remeber not being able to play networked games with friends at university because they had hardware that was a couple of years newer and the difference in performance and in what games we could run was immense.
91-96, 386sx16 to pentium 90, that's exactly what I did. The difference was huge.

I don't think we get the same kind of performance gap in desktop pcs anymore.

Cyrix was also in the 386 market.
Quake killed Cyrix - much of the grunt of quake was written in hand coded assembler optimised for the pentium.