5 ms·

Around then, '98 I think, I remember benchmarking the first computer I bought for myself, a Pentium II @ 350 MHz, and being floored to learn that I had just bou

by retrac 5y ago

Around then, '98 I think, I remember benchmarking the first computer I bought for myself, a Pentium II @ 350 MHz, and being floored to learn that I had just bought a desktop supercomputer! It was in the same ballpark as a single processor Cray Y-MP. I'm pretty sure some Cray Y-MPs hadn't even been retired yet at the time. From about $2.5 million to $2500 in about 10 years, for comparable hardware.

wink 5y ago
To this day I cherish the memory when a friend and I got to one of the local computer shops and bought 2 identical PII-350s and, as high school students buying with a random mix of 10s, 20s and the occasional 50 DM bill and the shopkeeper lamenting the students and their small bills on the one hand and being happy to accept this sale on the other hand of course.
david38 5y ago
You might be right. It takes time to rewrite software. But damn, those machine cost a pretty penny to run. Still, I would love to have one.
geocrasher 5y ago
Indeed! The primary difference wasn't the processing power but rather the parallel capacity of the Cray to solve a completely different kind of problem.
tomnipotent 5y ago
Everything from 1997-2000 was really just magical, for lack of a better adjective. 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.
- reassembled 5y ago
  I still have my original Celeron 300A, made in Costa Rica, that I overclocked to 450MHz on an A-Bit BX6 2.0 mobo to play Quake 2. 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.
  - throw0101a 5y ago
    For some reason I have the box for the Asus CUSL2 motherboard still sitting on a shelf: * https://www.anandtech.com/show/578 https://www.anandtech.com/show/578 (Never bothered to throw it out.)
  - tomnipotent 5y ago
    > to play Quake 2. 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...)
    - bennysomething 5y ago
      Don't forget Goldeneye!
  - mpalfrey 5y ago
    Did that do much? I know a Voodoo 2 could be SLI'd but first I've heard about this working on a Voodoo 1. 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)!.
dredmorbius 5y ago
Supercomputers don't so much buy speed as time. 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.
- agumonkey 5y ago
  there's a thread on slashdot IIRC where a guy explains that a good old cray, due to its massive cache and bandwidth, can crunch number matrices at massive "speed". I forgot the details but basically each cycle the processor could recompute a whole 1MB or 8MB matrix, and steadily so. Whereas a pentium4/core cpu, clocked way higher and more "modern" can only process that amount of data for a small time before all caches are filled and performance plummets.
  - bombcar 5y ago
    People compare supercomputers to general purpose computers and forget things like this - the supercomputer and the program it ran were often designed in parallel to do a particular thing very very fast.
    - agumonkey 5y ago
      Yeah goals impacts architecture. But it's interesting to see that computing systems have parameters and structures. Cycle clock, memory, cache .. they don't paint the whole picture. The constant massive data crunching of a Cray is not visible on frequency. That said, I'm curious about what a 2020 iphone can do, even in HPC numerical contexts..
iso1210 5y ago
At work in 1997 they bought a rack full of IRIX graphics processing which costs 000,000s to do a specific task When I started in 2003 the same task was performed on a 2U machine costing in the region of 2k.
lordnacho 5y ago
I remember during the 90s, whenever I or anyone I knew got a new machine, it would be noticeably faster than previous machines. Clock speeds rocketed during the era, and you could actually tell just from waiting for Windows to start. 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.
- Joeri 5y ago
  It wasn’t just that things got faster quickly. Hardware would rapidly become too old to run modern software. My first pc was a 386 33 mhz with 8 mb ram around 1993, and by 1998 software had passed it by to the point where I literally had no use for it except hand it down as a legacy gaming computer to my younger sister. I remember being forced to use 5 year old 386 computer lab pc’s at uni in 1997 and them being too slow to do our programming assignments on. 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.
  - zozbot234 5y ago
    IBM-compatible PC's were expensive prior to the late-1990s, but there were plenty of lower-cost competitors in the home computer market.
- api 5y ago
  I call today's situation "more's law." Clock speeds and single core performance are largely stagnant but the horizontal growth is still amazing. 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.

by retrac 5y ago

wink 5y ago

To this day I cherish the memory when a friend and I got to one of the local computer shops and bought 2 identical PII-350s and, as high school students buying with a random mix of 10s, 20s and the occasional 50 DM bill and the shopkeeper lamenting the students and their small bills on the one hand and being happy to accept this sale on the other hand of course.

david38 5y ago

You might be right. It takes time to rewrite software. But damn, those machine cost a pretty penny to run. Still, I would love to have one.

geocrasher 5y ago

Indeed! The primary difference wasn't the processing power but rather the parallel capacity of the Cray to solve a completely different kind of problem.

tomnipotent 5y ago

Everything from 1997-2000 was really just magical, for lack of a better adjective. 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.

reassembled 5y ago
I still have my original Celeron 300A, made in Costa Rica, that I overclocked to 450MHz on an A-Bit BX6 2.0 mobo to play Quake 2. 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.
- throw0101a 5y ago
  For some reason I have the box for the Asus CUSL2 motherboard still sitting on a shelf: * https://www.anandtech.com/show/578 https://www.anandtech.com/show/578 (Never bothered to throw it out.)
- tomnipotent 5y ago
  > to play Quake 2. 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...)
  - bennysomething 5y ago
    Don't forget Goldeneye!
- mpalfrey 5y ago
  Did that do much? I know a Voodoo 2 could be SLI'd but first I've heard about this working on a Voodoo 1. 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)!.

dredmorbius 5y ago

Supercomputers don't so much buy speed as time. 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.

agumonkey 5y ago
there's a thread on slashdot IIRC where a guy explains that a good old cray, due to its massive cache and bandwidth, can crunch number matrices at massive "speed". I forgot the details but basically each cycle the processor could recompute a whole 1MB or 8MB matrix, and steadily so. Whereas a pentium4/core cpu, clocked way higher and more "modern" can only process that amount of data for a small time before all caches are filled and performance plummets.
- bombcar 5y ago
  People compare supercomputers to general purpose computers and forget things like this - the supercomputer and the program it ran were often designed in parallel to do a particular thing very very fast.
  - agumonkey 5y ago
    Yeah goals impacts architecture. But it's interesting to see that computing systems have parameters and structures. Cycle clock, memory, cache .. they don't paint the whole picture. The constant massive data crunching of a Cray is not visible on frequency. That said, I'm curious about what a 2020 iphone can do, even in HPC numerical contexts..

iso1210 5y ago

At work in 1997 they bought a rack full of IRIX graphics processing which costs 000,000s to do a specific task When I started in 2003 the same task was performed on a 2U machine costing in the region of 2k.

lordnacho 5y ago

I remember during the 90s, whenever I or anyone I knew got a new machine, it would be noticeably faster than previous machines. Clock speeds rocketed during the era, and you could actually tell just from waiting for Windows to start. 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.

Joeri 5y ago
It wasn’t just that things got faster quickly. Hardware would rapidly become too old to run modern software. My first pc was a 386 33 mhz with 8 mb ram around 1993, and by 1998 software had passed it by to the point where I literally had no use for it except hand it down as a legacy gaming computer to my younger sister. I remember being forced to use 5 year old 386 computer lab pc’s at uni in 1997 and them being too slow to do our programming assignments on. 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.
- zozbot234 5y ago
  IBM-compatible PC's were expensive prior to the late-1990s, but there were plenty of lower-cost competitors in the home computer market.
api 5y ago
I call today's situation "more's law." Clock speeds and single core performance are largely stagnant but the horizontal growth is still amazing. 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.