In the PC world, using your computer for 10+ years is quite common, especially if you haven't much hardware needs like Alex. I've got a PC I bought in 2017 with 32 GB of RAM. The only problem would be Windows 11 incompatibility (and that's easily solved with mainstream Linux distributions these days). On the hardware side, of course I can swap any failing part.
Daily driving a i7-2600K+24GB RAM (a broken slot) I bought around 2013. And I have a compute-heavy job (but I do have access to a computing cluster for large workloads though). My phone is a samsung S7. It brings me a lot of joy to limit my participation in consumerism.
I enjoy the discussions on this site, but sometimes I feel like I am on a different planet. 10 years for a computer, especially a high-end one, seems… extremely standard? Unless you need 4K gaming AAA games I guess.
Even for gaming this is becoming pretty normal. Looks like my 1950X and 1080 Ti with 64 GB of ram were also bought in 2017. I need to upgrade for a few games at this point but it's still great.
Proprietary software is the single greatest threat to a long service life for any hardware. Running Linux and other free open source software for my needs has allowed me to use computers for as long as I want and buy decent used machines with confidence. Linux and other open source software meant my dad who was dreading having to buy a brand new computer for Windows 11 could instead by a very capable machine for $250 that runs Fedora instead.
My current computers are all fairly old, with my primary laptops being a model of EliteBook release in late 2019/early 2020. I have two of them and they both have 32 GB RAM. The one cost me $200 and the other cost $150 a year later. I haven't yet run into anything they couldn't do well enough for my needs. They have 1080p screens and are decently solid. I won't be the least bit surprised if they make it past ten years old, though I may upgrade sooner to some other three-year-old $250 wonder. I will likely be able to put a clean install of Fedora on them and give them to people who will use them for another 5-10 years.
I recently upgraded to a fancy modern machine with a Ryzen 7 8840 and I barely perceive a difference in performance from my 7(?) year old Precision 5540. I get ~10 hours on Linux (with power save settings) from both, benchmarks are ~25% faster on the new one and things are slightly snappier but my life isn't better. Moore's law stopped for CPUs a while ago and e.g. L1 caches won't grow. We have had 10 year computers for a decade.
Transistor counts stopped growing exponentially ages ago, that's why multicore gained prominence. I'm not brave enough to say Moore's law died in 2006 since the Intel Duo etc. but many actively argue it died in 2014. MCM lets 2 pieces replace 1 but it's 2x the transistors at 2x the cost. Actual processors stalled and they're just redefining things. For example, RAM originally meant what's today called L1 cache. Slower things have been progressively added until optane failed on the market.
nothing CPU-bound anymore
Is a different but more important topic in my mind. Jevons paradox has led to countless wasted flops for no benefit. Increasingly many believe today's online experience is worse than in 1995. It's easy to critique such complaints, but it's clear more compute has not improved our lives either. Using electron instead of Lazarus doesn't even save devs time besides being slower even on this hardware and being produced by user-abusing systems. And even in terms of pure flops, my benchmark results are barely different after a decade:
~25% faster ... slightly snappier but my life isn't better
Tangential counterpoint: Lisp seems to require ~50-100MB of ram to really work (although Janet manages ~10MB) so overcoming that hurdle was concrete; for me it only happened around 2006.
This is a weird way to measure performance/interpret Moore's, it would imply that chiplets/horizontal scaling are not increasing transistor count/density.
but many actively argue it died in 2014. MCM lets 2 pieces replace 1 but it's 2x the transistors at 2x the cost.
Not sure how I would interpret the cost ratio with respect to the original prediction. Similar cost per die was somewhat implicit, but was never an explicit calculation like on that graph.
Even if we take your original comparison (Precision 5540 — Intel Core i7-9850H vs Ryzen 7 8840U), we're looking at 3 billion (Intel didn't publish transistor counts for the mobile chips IIRC, but similar silicon to the desktop nodes) vs 25 billion. A lot of that is admittedly NPU/iGPU/AVX stuff, but that arguably has a significant impact in a lot of real world workloads.
Actual processors stalled and they're just redefining things. For example, RAM originally meant what's today called L1 cache.
That’s silly. Caches have nothing to do with the end of Moore’s law or Dennard scaling. There have been computers with CPU caches since the 1960s, with similar semantics to current caches. The difference is that RAM is addressible but caches are not.
The chassis is from 2010 and the aftermarket replacement motherboard is from 2016. It's the best machine I've ever owned by a large margin, and I hope to continue using it for many more years.
The biggest constraint for doing this with a laptop is obviously the battery. My previous Thinkpad that I used until 2023 was an X301 which had two battery bays, but they were both fairly thin. It was impossible to order a working aftermarket battery for that device. (I tried twice and both times I got a battery that ballooned up and threatened to crack the chassis after 2 months.) But with the X210 the battery protrudes out of the bottom and is quite a bit thicker, which means that aftermarket batteries can be constructed out of high-quality, readily available lithium-ion 18650 cells. This makes a big difference!
I love my x270. It feels like a good, snappier alternative to the x201 and x230t that I used in college, with the same core DNA. I've not used one of these motherboard swap early x2xx machines, but been really interested in them for the last few years.
Swappable batteries were such a godsend, kind of unfortunate the industry moved away for the sake of thinness and aluminium unibody chassis.
I'd been eyeing the motherboard for the latest gen Intel chips (x210AI but with DDR5 RAM prices as they are feels like an unreasonable vanity purchase.
My parents still have an HP Pavilion dv7 that they bought in 2008 and don't want to let go. I put linux on it and an ssd, but other than that it's been running just fine for the occasional email check and spreadsheet update for 18 years. The new MacBook Pro from 2024 gets less use and it's hilarious.
I ordered a new M5 Max 14" MBP to replace my nearly 5-year-old M1 Pro 14" MBP, and my plan for that is to push it to a decade. The only reason I'm not pushing my M1 Pro to a decade is that I made compromises with the spec - 16GB RAM is a hair too little, and 512GB storage is absolutely too small. My new laptop will have 64GB RAM and 4TB storage (probably could've gone with 2TB but I didn't want to regret my choice...), and I expect that will be able to carry me to 2036. At least for a laptop... my gaming PC will probably get upgraded between now and then simply because I want to be able to play new games as they come out at reasonably high settings.
It's kinda crazy that the only reason I'm replacing a laptop nearly 5 years in is that I have nearly the base spec. If I had ordered that thing from the start with 64 GB RAM and enough storage, chances are I wouldn't have been even slightly tempted to order one of the new ones. I gotta admit, the stagnation in computer performance is pretty nice for curbing impulse spending :)
I got 13 years out of my Quad G5 as a daily driver, and even then only because I moved to the Raptor. It still works and does the specific tasks I require of it (mostly media and creative). And if we're not talking daily drivers, well ...
This is absolutely possible. My only computer is still a 2015 mbp and it gives me no problems. I think the key is not updating software (especially the os) and periodically pruning files/processes for disk space and ram. Also being ok with what you have is key. Sure, i occasionally come across some software that would require updating everything but 99 times out of 100, my life is perfectly fine not using that software and making do with the tools already at my disposal. I will say that I’ve been forced to mostly give up ios development and xcode because of this but honestly, good riddance.
It should be doable. I bought an M1 Max with the same ten year idea. The five year mark is coming up and I've felt no impulse to upgrade. If the OS support dries up, there's Asahi.
Unless they switch CPU architectures again. A friend of mine is replacing his Mac laptop and we thought of taking it off his hands for my wife (who prefers Apple). But it's an x86, which will only be supported for another 3 years.
My daily driver was manufactured in 2012, IIRC I bought it in 2018, and am still almost always happy with it today. Of course, I did replace the HDD with an SSD and max out the RAM at 16 GB.
jmiven | a day ago
In the PC world, using your computer for 10+ years is quite common, especially if you haven't much hardware needs like Alex. I've got a PC I bought in 2017 with 32 GB of RAM. The only problem would be Windows 11 incompatibility (and that's easily solved with mainstream Linux distributions these days). On the hardware side, of course I can swap any failing part.
nicoco | 12 hours ago
Daily driving a i7-2600K+24GB RAM (a broken slot) I bought around 2013. And I have a compute-heavy job (but I do have access to a computing cluster for large workloads though). My phone is a samsung S7. It brings me a lot of joy to limit my participation in consumerism.
I enjoy the discussions on this site, but sometimes I feel like I am on a different planet. 10 years for a computer, especially a high-end one, seems… extremely standard? Unless you need 4K gaming AAA games I guess.
slondr | 18 hours ago
I'm typing this on a laptop released in 2018 which now has 80 GB of RAM. I forsee using it for quite a few more years.
rileylabrecque | 18 hours ago
Even for gaming this is becoming pretty normal. Looks like my 1950X and 1080 Ti with 64 GB of ram were also bought in 2017. I need to upgrade for a few games at this point but it's still great.
gnafuthegreat | 19 hours ago
Proprietary software is the single greatest threat to a long service life for any hardware. Running Linux and other free open source software for my needs has allowed me to use computers for as long as I want and buy decent used machines with confidence. Linux and other open source software meant my dad who was dreading having to buy a brand new computer for Windows 11 could instead by a very capable machine for $250 that runs Fedora instead.
My current computers are all fairly old, with my primary laptops being a model of EliteBook release in late 2019/early 2020. I have two of them and they both have 32 GB RAM. The one cost me $200 and the other cost $150 a year later. I haven't yet run into anything they couldn't do well enough for my needs. They have 1080p screens and are decently solid. I won't be the least bit surprised if they make it past ten years old, though I may upgrade sooner to some other three-year-old $250 wonder. I will likely be able to put a clean install of Fedora on them and give them to people who will use them for another 5-10 years.
veqq | a day ago
I recently upgraded to a fancy modern machine with a Ryzen 7 8840 and I barely perceive a difference in performance from my 7(?) year old Precision 5540. I get ~10 hours on Linux (with power save settings) from both, benchmarks are ~25% faster on the new one and things are slightly snappier but my life isn't better. Moore's law stopped for CPUs a while ago and e.g. L1 caches won't grow. We have had 10 year computers for a decade.
vpr | 20 hours ago
Is the claim that transistor counts are no longer exponential? Or rather is nothing CPU-bound anymore?
veqq | 14 hours ago
Transistor counts stopped growing exponentially ages ago, that's why multicore gained prominence. I'm not brave enough to say Moore's law died in 2006 since the Intel Duo etc. but many actively argue it died in 2014. MCM lets 2 pieces replace 1 but it's 2x the transistors at 2x the cost. Actual processors stalled and they're just redefining things. For example, RAM originally meant what's today called L1 cache. Slower things have been progressively added until optane failed on the market.
Is a different but more important topic in my mind. Jevons paradox has led to countless wasted flops for no benefit. Increasingly many believe today's online experience is worse than in 1995. It's easy to critique such complaints, but it's clear more compute has not improved our lives either. Using electron instead of Lazarus doesn't even save devs time besides being slower even on this hardware and being produced by user-abusing systems. And even in terms of pure flops, my benchmark results are barely different after a decade:
Tangential counterpoint: Lisp seems to require ~50-100MB of ram to really work (although Janet manages ~10MB) so overcoming that hurdle was concrete; for me it only happened around 2006.
vpr | 2 hours ago
This is a weird way to measure performance/interpret Moore's, it would imply that chiplets/horizontal scaling are not increasing transistor count/density.
Rome was at 40 bililon, Genoa's at 90 billion, and Turin 150 billion (Milan was same dies as Rome so flat there).
Not sure how I would interpret the cost ratio with respect to the original prediction. Similar cost per die was somewhat implicit, but was never an explicit calculation like on that graph.
Even if we take your original comparison (Precision 5540 — Intel Core i7-9850H vs Ryzen 7 8840U), we're looking at 3 billion (Intel didn't publish transistor counts for the mobile chips IIRC, but similar silicon to the desktop nodes) vs 25 billion. A lot of that is admittedly NPU/iGPU/AVX stuff, but that arguably has a significant impact in a lot of real world workloads.
fanf | 5 hours ago
That’s silly. Caches have nothing to do with the end of Moore’s law or Dennard scaling. There have been computers with CPU caches since the 1960s, with similar semantics to current caches. The difference is that RAM is addressible but caches are not.
technomancy | a day ago
My current machine is an nb51 X210: https://blog.mattgauger.com/2022/08/01/the-51nb-x210/
The chassis is from 2010 and the aftermarket replacement motherboard is from 2016. It's the best machine I've ever owned by a large margin, and I hope to continue using it for many more years.
The biggest constraint for doing this with a laptop is obviously the battery. My previous Thinkpad that I used until 2023 was an X301 which had two battery bays, but they were both fairly thin. It was impossible to order a working aftermarket battery for that device. (I tried twice and both times I got a battery that ballooned up and threatened to crack the chassis after 2 months.) But with the X210 the battery protrudes out of the bottom and is quite a bit thicker, which means that aftermarket batteries can be constructed out of high-quality, readily available lithium-ion 18650 cells. This makes a big difference!
eyesinthefire | 22 hours ago
I love my x270. It feels like a good, snappier alternative to the x201 and x230t that I used in college, with the same core DNA. I've not used one of these motherboard swap early x2xx machines, but been really interested in them for the last few years.
vpr | 20 hours ago
Swappable batteries were such a godsend, kind of unfortunate the industry moved away for the sake of thinness and aluminium unibody chassis.
I'd been eyeing the motherboard for the latest gen Intel chips (x210AI but with DDR5 RAM prices as they are feels like an unreasonable vanity purchase.
jm | 19 hours ago
Mac OS update limitations seem to screw all my Mac hardware.
I do better on the windows machines, running the OS on a vm so I can start over once while, and run multiple machines when I need to
ag | 15 hours ago
My parents still have an HP Pavilion dv7 that they bought in 2008 and don't want to let go. I put linux on it and an ssd, but other than that it's been running just fine for the occasional email check and spreadsheet update for 18 years. The new MacBook Pro from 2024 gets less use and it's hilarious.
pmc | 22 hours ago
I ordered a new M5 Max 14" MBP to replace my nearly 5-year-old M1 Pro 14" MBP, and my plan for that is to push it to a decade. The only reason I'm not pushing my M1 Pro to a decade is that I made compromises with the spec - 16GB RAM is a hair too little, and 512GB storage is absolutely too small. My new laptop will have 64GB RAM and 4TB storage (probably could've gone with 2TB but I didn't want to regret my choice...), and I expect that will be able to carry me to 2036. At least for a laptop... my gaming PC will probably get upgraded between now and then simply because I want to be able to play new games as they come out at reasonably high settings.
It's kinda crazy that the only reason I'm replacing a laptop nearly 5 years in is that I have nearly the base spec. If I had ordered that thing from the start with 64 GB RAM and enough storage, chances are I wouldn't have been even slightly tempted to order one of the new ones. I gotta admit, the stagnation in computer performance is pretty nice for curbing impulse spending :)
classichasclass | 20 hours ago
I got 13 years out of my Quad G5 as a daily driver, and even then only because I moved to the Raptor. It still works and does the specific tasks I require of it (mostly media and creative). And if we're not talking daily drivers, well ...
nolanvoid | an hour ago
This is absolutely possible. My only computer is still a 2015 mbp and it gives me no problems. I think the key is not updating software (especially the os) and periodically pruning files/processes for disk space and ram. Also being ok with what you have is key. Sure, i occasionally come across some software that would require updating everything but 99 times out of 100, my life is perfectly fine not using that software and making do with the tools already at my disposal. I will say that I’ve been forced to mostly give up ios development and xcode because of this but honestly, good riddance.
kevinc | 14 hours ago
It should be doable. I bought an M1 Max with the same ten year idea. The five year mark is coming up and I've felt no impulse to upgrade. If the OS support dries up, there's Asahi.
alper | 12 hours ago
I've had a Macbook Pro last me 9 years, so it's fairly common in Apple land (also with continued operating system upgrades).
sjamaan | 10 hours ago
Unless they switch CPU architectures again. A friend of mine is replacing his Mac laptop and we thought of taking it off his hands for my wife (who prefers Apple). But it's an x86, which will only be supported for another 3 years.
gcupc | 7 hours ago
My daily driver was manufactured in 2012, IIRC I bought it in 2018, and am still almost always happy with it today. Of course, I did replace the HDD with an SSD and max out the RAM at 16 GB.