In just a few years we saw the Intel Pentium II and Celeron (insane overclock), the AMD K6/K6-2, along with discrete 2D/3D GPUs like 3dfx Voodoo3 and Nvidia GeForce.
Thanks to low-ping affiliation, I stumbled into an early Counter-Strike clan known as "The Space Offensive" (TSO) in the Seattle area that went on to be an early participant in the competitive gaming scene as "The Speakeasy Offensive" (after corporate sponsorship). I've been hooked on the cycle of gaming/hardware loop ever since.
I managed to snag my buddy’s old 4MB 3dfx Voodoo 1 to pair with the one I already had with the VGA pass through cable, an early form of multi-GPU.
In just a few years we had Quake 2 (1997) to Half-life (1998), Thief: The Dark Project and Quake 3 Arena/Unreal Tournament (1999). The year 2000 would see Dues Ex, The Sims, Diablo II, Perfect Dark, Final Fantasy IX, and Jet Set Radio... Not to mention Spyro, Red Alert 2, Paper Mario, Majora's Mask, Escape from Monkey Isand, Hitman 47...
So much amazing history in such a short period of time (let's not mention popular cinema...)
I miss those days. I went from a P133 with 72mb RAM (bought a load of RAM cheap in '98) to a P3 500 with a TNT2 Ultra and 128mb RAM in 1999. Technology progressed at an insane rate.
Edit, I had a K6-2 500MHz in one machine around that kind of time too. Wasn't as good as the P3 personally though but was cheap and reasonable. Started me off down the path of using AMD CPUs in my builds (Duron, Athlon, Sempron) until I started moving away to Mac laptops predominantly (which coincided with finishing university and writing software for a living)!.
IIRC the Voodoo 1 didn't have 2D capability so the pass through was to let your 2D handle that drawing.
Sounded like this was 2x Voodoo 1s stacked together. I probably need more coffee this morning!
This is always good policy.
* https://www.anandtech.com/show/578
(Never bothered to throw it out.)
Don't forget the race to 1Ghz. Intel and AMD spent those years doubling core clock speeds every year. If you bought a system in 1996 and replaced it in 2000 the CPU might be four or more times faster. For some things that was a real geometric increase in computing power.
That it was. I worked at a small computer animation studio around that time, and that was when a home built 'cheap' Intel machine with a GeForce cards started to first match and then surpass our high end SGI and Intergraph machines costing 5-10 as much in many workloads.
When I left almost all our animators where using 'home built' computers with Nvidia cards.
A little while later, Apple released the G4 xServe. We bought 2 loaded with the largest disks at the cost of several thousand dollars. But hey, OSX Server. It has PHP & MySQL and is rack mountable, etc. We also used them for file servers for all the design files.
Got all of our development sites migrated over and the performance was noticeable worse. The HP was at least 4 times faster serving up the sites. It was amazing.
Graphics cards did something similar towards the end of the decade, each time a friend got a new one it was a like a whole new world.
Since any one person would still be hanging on to their machine for a few years, you could really feel the slowdown when you went back home.
Pc’s also cost way more back then. That 386 was bottom spec and it was still 1700 euro in 2022 money. Being into computers was an expensive hobby.
Look at the original M1 in the MacBook Air vs. the M1 Ultra or whatever they call that 48-64 core beast. Single core is about the same. If your load is not parallelizable the M1 MacBook Air will do it about as fast as the new Mac Studio.
In the x64 world they continue to squeeze more instruction level parallelism out and push the clock rates slightly higher but those come at the cost of higher power consumption. It's pretty stagnant there too unless you are willing to invest in a refrigerated liquid cooling system and overclocking capable hardware.
But on multi-threaded loads the performance is still going up the way single core speeds were from 1995 until 2005. That's because a mixture of smaller process nodes, larger dies, better memory bandwidth, and better core interconnect fabrics are letting the equivalent of a whole data center be crammed into a single package.
I think that is going to keep going for a while. We will probably see 256 cores on a desktop machine within 10 years and rack mount servers with 1024 cores in a single 1U case.
A supercomputer is a decade or two ahead of consumer-grade hardware of the time.
And it's not just a matter of processor speed. Parallel disk I/O, total memory size, cache, and a number of other tricks factor in.
Today's cluster supercomputers are mostly just ("just") racks and racks and racks of fairly standard-grade CPUs, with high-speed switches.
GPUs have introduced their own curve into things. They work well for highly-parallelisable problems. Unfortunately, the ones that seem to have caught many people's attention are ultimately somewhat boring.
That said, I'm curious about what a 2020 iphone can do, even in HPC numerical contexts..
When I started in 2003 the same task was performed on a 2U machine costing in the region of 2k.
Still, I would love to have one.