PC Client shipments in free fall Q1.

[nerp]

Visit any small mid business and you will see 99% are on PC with 2-3 screen, a lot of them are still on C2D or 1st/2nd gen core I. They're still good enough for them with the legacy program they depends on.

Same apply to home user, they just don't have a NEED to upgrade YET. But they will soon.

Mobile devices just isn't enough for productivity, let said you planning for the up coming road trip, you need to take notes, plan route, POI, book hotel and rental a car. Would you(or anyone) rather to do it on your phone/tab/laptop or on a PC with 2 big screen/kb/ms? I am sick of taking note on phone, I use evernote on PC and let it sync to my phone.

Cable installer - most home use wifi, a $10 USB stick is much cheap than cable installer
Plumber?
Electrician?

Plumbers and electricans enter peoples' homes. They see if people have desktop computers. They do not see them anymore. That is my point. I agree that desktop computers are great but they are not selling. That is the whole premise of this thread. Shipments are tanking. Yet people here are saying that they're not? Or that they love their desktop computers? Or here are 10000 reasons why we SHOULD be buying computers? OK. That's great. Does not explain why sales continue to tank.

 

[nerp]

Visit any small mid business and you will see 99% are on PC with 2-3 screen, a lot of them are still on C2D or 1st/2nd gen core I. They're still good enough for them with the legacy program they depends on.

Same apply to home user, they just don't have a NEED to upgrade YET. But they will soon.

Mobile devices just isn't enough for productivity, let said you planning for the up coming road trip, you need to take notes, plan route, POI, book hotel and rental a car. Would you(or anyone) rather to do it on your phone/tab/laptop or on a PC with 2 big screen/kb/ms? I am sick of taking note on phone, I use evernote on PC and let it sync to my phone.

Cable installer - most home use wifi, a $10 USB stick is much cheap than cable installer
Plumber?
Electrician?

No they won't. They will have a tablet or device that broadcasts to a larger screen if they have it. They are not going to buy a desktop computer. Again, gamers are a small fraction of the overall computer sales market. Intel is selling i7s, yes, but where do you think these chips are going? Not in people's houses. Again, why do people have trouble accepting this? They are the plain Jane facts. Consumers aren't buying desktops anymore. You or I or anyone else hanging out on this forum do not represent the overall market. As a result, us gobbling up hardware is not enough to stem the tide of PC sales declines.

You are correct that businesses have no need to buy tons of new desktops for mainstream corporate use. The few employees that need powerful workstations get them, but they're not representative of the overall corporate PC market.

 

[intangir]

Linus Torvalds is talking nonsense, but then, he is good at that; he does not have the experience to see that there are many and diverse solutions to many engineering problems, not just one solution to each.

... what an ignorant thing to say. I don't know of many people at all who could possibly have more experience in examining different engineering solutions in the computing world than the gatekeeper of code contributions to the Linux kernel. I think it's fair to say that if it hasn't passed through Linus, it's not been done.

He is also not considering the cost of solutions; it is true that you could buy an expensive dual-core CPU from Intel, to execute an algorithm, which could otherwise be run on a 'wimpy' and cheap ARM based multi-core SoC, but most of us are cost-constrained. As a case in point, I am developing a system which runs in Mathematica, and uses all four processors in a RaspberryPi 2, at times; it would run fasterstill with eight wimpy cores, so I look forward to a future Pi offering this; the Pi is an order of magnitude cheaper than Intel's Core CPUs; it is good engineering to use the cheapest solution to this; as I leave my Pi on 24 hours a day, the wimpy solution is cheaper on electricity too.

I don't think you truly have experience with many-core systems. This all just sounds like pie-in-the-sky ramblings "Oh, it works great on four cores, so it should run great on eight!" Unfortunately, there is such a thing as dark silicon that places an upper bound on the number of transistors you can power at once, independent of process, and this inescapably limits the benefits of increasing the number of cores you have. Even if you double your number of cores, you'd only be able to run them at a fraction of the speed without blowing your power budget, which would most likely neutralize or even regress your total performance. And there is of course a die-size cost for that which translates to a higher cost to you for all those extra transistors you're buying that aren't doing you a lick of good.

Dark Silicon and the End of Multicore Scaling

First, we kept the power budget constant (our default budget is 125 W) and varied the level of parallelism in the PARSEC applications from 0.75 to 0.99, assuming that programmer effort can realize this improvement. Performance improved slowly as the parallelism level increased, with most benchmarks reaching a speedup of about only 15 at 99 percent parallelism. Provided that the power budget is the only limiting factor, typical upper-bound ITRS-scaling speedups will still be limited to 15x. With conservative scaling, this best-case speedup is limited to 6.3x.

For the second experiment, we kept each application’s parallelism at its real level and varied the power budget from 50 W to 500 W. Eight of 12 benchmarks showed no more than 10x speedup even with a practically unlimited power budget. In other words, increasing core counts beyond a certain point did not improve performance because of the limited parallelism in the applications and Amdahl’s law. Only four benchmarks have sufficient parallelism to even hypothetically sustain speedup levels that matches the exponential transistor count growth, Moore’s law.

And you know what core count is required to get there? 442 cores at 8nm! The cherry on top? 35 cores gets you 90% of the performance of 442 cores! Brilliant! You can spend 12 times the transistors to get 1.1 times the performance! Why is no company jumping at the chance to do this??

Keep in mind, that's absolute best-case assuming the software can achieve 99% parallelism. Real-world performance gains will be even less. These calculations don't take into account memory latency or bandwidth limits, they don't take into account thread synchronization or operating system serialization, they don't take into account system memory sizes. All of these are shared resources in multicore systems, that would unavoidably cause smaller-threaded workloads to have more resources per thread than when you're running the system with the max thread count, and further depress any possible gains to be had from increasing the number of threads.

 

[moonbogg]

Long live Sandy! Hell yeah! Go Sandy-E GO BABY GO!!!

 

[2is]

No they won't. They will have a tablet or device that broadcasts to a larger screen if they have it. They are not going to buy a desktop computer.

It's clear now you haven't actually stepped into a business environment and are making things up as you go.

 

[Systems analyst]

When one actually reads and comprehends what Linus said, he says that "end users" are fine with on the order of four cores. He wrote this in 2014. This wasn't a statement of " this is true now and forever", either. Unless you are trying to purposely misrepresent what he wrote in order to set up a strawman, clearly he is only talking about the typical PC user having workloads that work better with fewer, beefier cores; the magic number 4 was an order of magnitude statement, not a specific hard value. He is saying that the people who evangelize extreme parallelism as a solution to everything are wrong, not that nobody anywhere, ever would ever want more than 4 cores. I realize one might not like his sometimes brash writing style, but he makes good points if you actually have good reading comprehension.

Linus said:
End users are fine with roughly on the order of four cores, and you can't fit any more anyway without using too much energy to be practical in that space. And nobody sane would make the cores smaller and weaker in order to fit more of them - the only reason to make them smaller and weaker is because you want to go even further down in power use, so you'd still not have lots of those weak cores.

So the whole argument that people should parallelise their code is fundamentally flawed. It rests on incorrect assumptions. It's a fad that has been going on too long.

Note:
'nobody sane would make the cores smaller and weaker in order to fit more of them - the only reason to make them smaller and weaker is because you want to go even further down in power use, so you'd still not have lots of those weak cores.' - this is contradicted by the advances in process nodes; these advances allow the provision of more cores, without weakening them; I could usefully use 10 cores now, with no significant programming effort required. A tool like Mathematica makes it easy to use multiple cores; does he know anything about Mathematica, or its like - I wonder.

'the whole argument that people should parallelise their code is fundamentally flawed. It rests on incorrect assumptions. It's a fad that has been going on too long.' - this is saying that more cores has no future for us, which is just not true. The provision of more cores, even just ARM Coretex A53s, is sufficient to drive algorithm growth.

There is research going on to unwind loops, and map the task onto multiple processors, using certain topolgy.

Linus is a hero of the computing revolution, but he is not all-knowing; I just disagree with him on this issue of parallel processing on small computer systems.

 

[Systems analyst]

... what an ignorant thing to say. I don't know of many people at all who could possibly have more experience in examining different engineering solutions in the computing world than the gatekeeper of code contributions to the Linux kernel. I think it's fair to say that if it hasn't passed through Linus, it's not been done.

You note that I said 'solutions' not computer systems; I have worked on systems where there are many radically different technologies available; you choose the one appropriate to your present application; this experience has informed me that it is very difficult to make broad statements in engineering; you never know when someone will solve a particular problem, and you cannot rule out future developments.

 

[Systems analyst]

I don't think you truly have experience with many-core systems. This all just sounds like pie-in-the-sky ramblings "Oh, it works great on four cores, so it should run great on eight!" Unfortunately, there is such a thing as dark silicon that places an upper bound on the number of transistors you can power at once, independent of process, and this inescapably limits the benefits of increasing the number of cores you have. Even if you double your number of cores, you'd only be able to run them at a fraction of the speed without blowing your power budget, which would most likely neutralize or even regress your total performance. And there is of course a die-size cost for that which translates to a higher cost to you for all those extra transistors you're buying that aren't doing you a lick of good.

Dark Silicon and the End of Multicore Scaling


And you know what core count is required to get there? 442 cores at 8nm! The cherry on top? 35 cores gets you 90% of the performance of 442 cores! Brilliant! You can spend 12 times the transistors to get 1.1 times the performance! Why is no company jumping at the chance to do this??

Keep in mind, that's absolute best-case assuming the software can achieve 99% parallelism. Real-world performance gains will be even less. These calculations don't take into account memory latency or bandwidth limits, they don't take into account thread synchronization or operating system serialization, they don't take into account system memory sizes. All of these are shared resources in multicore systems, that would unavoidably cause smaller-threaded workloads to have more resources per thread than when you're running the system with the max thread count, and further depress any possible gains to be had from increasing the number of threads.

I fully understand that there are two schools of thought on this subject, as represented in the various threads in this forum; but you must realise, as per my first point, that there is research going on, on tackling parallelism; it is very difficult in general to say, in engineering, that something cannot be done; that requires a state of knowledge beyond the wit of one man, or even one team. For example, if you are analysing the share prices of the roughly 100 companies in the FTSE100, that lends itself to parallelisation; multiple cores are very useful; you may say that, say, 100 cores could not be fitted in a single SoC; but SoCs are being developed now with 100 and even 250D: cores; this seems to me to contradict your belief that hundreds of cores are not useful. I will try to dig out a pertinent reference that I have seen, on unwinding loops in code.

 

[iCyborg]

If you had an i7 2600K OC or 3930K OC, I'd totally understand the smack talking but i7 920 is obsolete. If you don't find it obsolete, it simply means you don't use your PC for anything intensive like gaming or rendering or video work. If you did, no way would you be saying that there is no viable upgrade path from an i7 920.

At the end of the day, there is absolutely nothing wrong with using any outdated PC if it fits the users needs. However, let's not try an downplay the gigantic level of improvement that Intel has brought from an i7 920 to an i7 6700K/i7 5820K and soon to launch $389 6-core BW-E.

In Dec 2008 I bought i7 920, that I still use, to upgrade from a single core P4 2.4Ghz bought in Aug 2003, so 5.5 years between them. It's 5-10x more performance, huge difference, P4 is barely enough to run Vista alone. This is what I consider a gigantic leap. i7 920 to 6700K doesn't come anywhere close to this kind of difference in experience. And it's been 7.5 years.
And P4 itself was an upgrade for a Duron 700 MHz after some 4-5 years, yet another far bigger improvement than 920->6700K would be. So maybe 920->Sandy is a 'huge improvement' for you, but to me it's not even worth the 1-2 hr hassle to build machine + reinstall OS, I literally wouldn't take 2600K+MB for free... Yes, the first part of the quote is true for me, I'm no longer a power user - I've lost interest in gaming, and for the some video encoding I do, it's always run in background, 5 mins or 8 mins isn't much difference, I'm not tapping my fingers on the table waiting for something to finish.
Heck, my i7-920 desktop PC is still more powerful than ~80-90% new laptops (to be fair, I upgraded Radeon 4870 to 7950 + SSD, so it's not the exact same system as bought in 2008).

And I may be in minority, but I live in a condo and electricity is included in my rent. Whether I use only my smartphone or run 4x AMD 9590 systems 100% 24/7, my electricity bill would be the same.
Even if I paid for it directly, I really can't imagine that I would be doing all my computing on my 13" i5-5200U laptop, and only fire up the desktop when I really need it so as to save on the electricity bill, I practically never use my laptop at home. And it would be a bigger difference in kwh than 920->6700K...

 

[RussianSensation]

In Dec 2008 I bought i7 920, that I still use, to upgrade from a single core P4 2.4Ghz bought in Aug 2003, so 5.5 years between them. It's 5-10x more performance, huge difference, P4 is barely enough to run Vista alone. This is what I consider a gigantic leap. i7 920 to 6700K doesn't come anywhere close to this kind of difference in experience. And it's been 7.5 years.
And P4 itself was an upgrade for a Duron 700 MHz after some 4-5 years, yet another far bigger improvement than 920->6700K would be. So maybe 920->Sandy is a 'huge improvement' for you, but to me it's not even worth the 1-2 hr hassle to build machine + reinstall OS, I literally wouldn't take 2600K+MB for free...

You know this was already addressed by me in post #55 in this very thread?

"I went from a 1998 Pentium II 233mhz MMX (never overclocked) to a 2001 Athlon XP1600+ overclocked to 1800+ speeds, or a 7.73X increase in performance in only 3 years. Then in 2003, just 2 years after my Athlon XP1600+, I got a Pentium 4 2.6Ghz "C" that overclocked to 3.2Ghz. That means from 1998 to 2003, or in 5 years, my CPU performance increased 13.73X. By 2006, I had E6400 @ 3.4Ghz (so more than 2X Pentium 4 "C" 3.2Ghz) and by August 2007, I upgraded that to Q6600 @ 3.4Ghz (double the cores). If I apply straight up linear logic, that means 13.73X * 2X for C2D and another 2X for 2 more cores = 54.92X from Fall 1998 to August 2007.

If I were to apply the same high standards, even for a small time-span of 1998 to 2003 where CPU speed for me went up almost 14X, I might as well never upgrade again for 30 more years then. See how flawed this logic is for CPUs? "

Yes, the first part of the quote is true for me, I'm no longer a power user - I've lost interest in gaming, and for the some video encoding I do, it's always run in background, 5 mins or 8 mins isn't much difference, I'm not tapping my fingers on the table waiting for something to finish.

Ya, so you are not the target market for modern PCs, it's as simple as that. I already addressed that point too. If the PC does everything a user needs, he won't upgrade. For 100 old consumers, Intel doesn't need 100 new buyers. They just need more buyers buying higher end i5/i7 CPUs that offset the loss of buyers who don't wants new PCs; and that's exactly what's happening == record i7 sales.

Heck, my i7-920 desktop PC is still more powerful than ~80-90% new laptops (to be fair, I upgraded Radeon 4870 to 7950 + SSD, so it's not the exact same system as bought in 2008).

Again, if your PC satisfies you, that's fine. Now for every 1 of you, there is me. I have an i7 IVB laptop that beats your i7 920, I also have an i7 6700K and soon I will get an i7 6800K. In 3-4 years, I will sell all of these and upgrade all of these systems with new i7s or AMD's 8+ core CPUs. Since you bought your i7 920, I built 25+ Intel systems, the worst of which had an i5 2500K. Intel loses sales to customers such as yourself since you are no longer interested, but it more than makes up for it by targeting people who love and still want desktops. Everyone who asks me for advice on what system to build gets recommended an i5/i7 and that means instead of buying junky Celerons, dual cores and i3s, every single system I built for friends/family had a minimum of an i5 2500K since 2011. In 2015/2016, I only built i7s. That included 2 x i7 4790Ks, 2 x i7 5820Ks, 1 x i7 6700K.

And I may be in minority, but I live in a condo and electricity is included in my rent. Whether I use only my smartphone or run 4x AMD 9590 systems 100% 24/7, my electricity bill would be the same.
Even if I paid for it directly, I really can't imagine that I would be doing all my computing on my 13" i5-5200U laptop, and only fire up the desktop when I really need it so as to save on the electricity bill, I practically never use my laptop at home. And it would be a bigger difference in kwh than 920->6700K...

I already explained in this thread that even if not accounting once for electricity costs, it made no sense to hold on to an i7 920/i7 860 anyway if you were smart about timing your upgrades. It's very simple: Total Cost of Ownership or TCO.

If I spent $700-800 on an i7 920 platform in 2008-2010, by now at best it's worth $200. That means it would have been a real loss of value between $500-600. Instead, I went i7 860 -> i5 2500K -> i7 6700K and after reselling all those parts and reinvesting the resale value, it cost me no more than $500-600 anyway. The difference is I had a very fast system in 2009-2010, 2011-2014 and now in 2016.

Just like right now I'll keep the i7 6700K + Z170 mobo temporarily, but then I'll just sell them in 3-4 years and may even reuse DDR4 I have in the future (more $ saved). Assuming I sell the CPU+Mobo for $225 in 3-4 years, it's exactly half of what I paid for them new in 2016. That means my TCO for that platform over 3-4 years is a mere $225. I will then take $225 from the resale value and buy an i7 9700K or i7 10700K in 2018-2019 and keep going.

That means with my strategy, I take all the hard work and stress and throw it out the window. Instead of spending $700-800+ on a new platform and sitting on it until it becomes nearly worthless, I just upgrade more frequently and reinvest the resale value. At the end of the day over 6-8 years, your TCO and mine are very similar but I have close to top-of-the-line rig every 3 years and yours is not.

So I don't even need to expalin that Skylake/Broadwell-E is a mediocre upgrade compared to going from Pentium II 233mhz MMX to Athlon XP1600+ or moving from Pentium 4 C 2.6Ghz @ 3.2Ghz to E6400 @ 3.4Ghz because holding old parts and letting them become worthless costs more or less than the same as does upgrading more often. Problem solved.

This TCO concept can be applied to cars, headphones, smartphones, videocards, etc.

For example, let's say someone bought a November 2010 GTX580 1.5GB for $499 and still has it today. That card is probably worth $50, or $450 loss of value/real $.

Here is me. When 580 was $500, I bought an HD6950 for $230, unlocked it to 6970 speeds. That got me 85-90% of the performance of a 580 in real world games at the time. Then I sold the 6950 for $160 and bought a 7970 for $400. Then I sold the 7970 for $160 and bought a 390 for $245.

My total cost is:

- $230
+ $160
- $400
+ $160
- $245
= - $555*

*But let's assume I can sell the 390 right now for $100, it would mean my TCO is - $455.

That means if Buyer 1 bought a $500 GTX580 and held on to it until now, they had good performance for 2 years and downhill from there. I had great performance every year until now and we both spent the same amount of $ in real terms. It actually cost me way less since I can easily sell the 390 for way more than $100 now.

Now take this exact same concept and apply it to CPU platform upgrades. :thumbsup:


Note: In reality I actually paid $0 for my GPU upgrades since my ATI/AMD cards made $ from 2008 but the TCO concept still applies and it's a highly effective method to understand the true cost to own something imho. That's why I wouldn't buy a pre-built rig since it's way too hard to resell those parts separately.

If you start thinking about the concept of TCO, you no longer think about how you set aside $1500-2000+ on a new PC. Otherwise, you'd be making the exact same mistake as you just made with your last rig. You'd potentially go out and drop $1500-2000 on new parts and again hold them for 6-8 years, while they depreciate close to $100-200 at most. Instead, learn the concept of TCO and resell and roll-over/reinvest the savings towards new parts. This way, you'll end up "wasting" the same $1500 over 8 years but instead you'd actually have a fast system every year over those 8 years!

 

[VirtualLarry]

Instead, learn the concept of TCO and resell and roll-over/reinvest the savings towards new parts. This way, you'll end up "wasting" the same $1500 over 8 years but instead you'd actually have a fast system every year over those 8 years!

Insightful, thanks RS. Also, a good reason to buy only top-end parts. Buying bottom-end parts like I have, means pretty-much effectively NO resale value a few years down the line. (Unless you consider a complete "$50 CL PC" to be a viable resale value.)

 

[AtenRa]

Dont compare CPUs to GPUs. From Sandy (2011) to Skylake (2016) the performance increase in games will be very low in contrast of GTX 580 (2010) to GTX 980Ti (2015).

 

[gorion]

I just upgrade more frequently and reinvest the resale value. At the end of the day over 6-8 years, your TCO and mine are very similar but I have close to top-of-the-line rig every 3 years and yours is not.


This TCO concept can be applied to cars, headphones, smartphones, videocards, etc.

You should also take into account how much time you invest in it.
Usually people are time-constrained (unless you are unemployed or a student) and the time you are investing in looking for the right part, handling the resale and upgrade the pc should be factored in.

 

[IndyColtsFan]

It's clear now you haven't actually stepped into a business environment and are making things up as you go.

I think he was primarily referring to home users.

 

[StrangerGuy]

Insightful, thanks RS. Also, a good reason to buy only top-end parts. Buying bottom-end parts like I have, means pretty-much effectively NO resale value a few years down the line. (Unless you consider a complete "$50 CL PC" to be a viable resale value.)

Actually, low-end mobos hold their value really well and should do even more so since Intel is releasing 4GHz+ stock i7s.

 

[TeknoBug]

Smartphone craze is dying out, it is getting in same place PC market went 5+ years ago. One reason is because market is getting saturated, but it would not be nowhere this bad if majority of population finances would be better.

And VR will be in the same spot soon. Anything that gets heavily saturated (like those MMORPG's trying to copy World of Warcraft's success) wears out fast.

 

[sm625]

Not to mention 6GB (2GB x3GB0 of DDR3 in 2008-2009 cost an arm and a leg, I bet like $150-200 USD. Today 16GB DDR4 3000 is $80.

Actually DDR3 was cheap in 2009. There were all sorts of deals for $10 per GB or even less.

Here is one such deal thread:

http://forums.anandtech.com/showthread.php?t=277646

I remember buying a 4GB kit for $35.99 but I cant find the thread.

It is actually rather sad that memory price per GB have only fallen 8% per year the last 7 years. Other types of memory have fallen in price at 2-4 times that rate.

 

[nerp]

It's clear now you haven't actually stepped into a business environment and are making things up as you go.

You are skimming and not reading what I'm saying. I haven't declared anything about corporate PC shipments other than to say they constitute a much larger portion of the overall CPU shipments than what is sold to gamers on Newegg and Amazon. I have also outlined the reasons and proof (beyond the actual tanking sales numbers staring at us in the face on paper from the companies themselves) that HOME pc sales are tanking. I have not offered any puffery about what is happening in the corporate world in terms of desktop computer sales.

 

[2is]

You are skimming and not reading what I'm saying. I haven't declared anything about corporate PC shipments other than to say they constitute a much larger portion of the overall CPU shipments than what is sold to gamers on Newegg and Amazon. I have also outlined the reasons and proof (beyond the actual tanking sales numbers staring at us in the face on paper from the companies themselves) that HOME pc sales are tanking. I have not offered any puffery about what is happening in the corporate world in terms of desktop computer sales.

I read the whole thing. The guy you quoted said visit small business and you'll see desktops. You said no you won't, you'll see tablets projecting to a large screen. He's right, you're wrong.

 

[nerp]

I was referring to HOME users again. But whatever dude, believe the guy who misunderstood me instead of me.

 

[MrTeal]

I already explained in this thread that even if not accounting once for electricity costs, it made no sense to hold on to an i7 920/i7 860 anyway if you were smart about timing your upgrades. It's very simple: Total Cost of Ownership or TCO.

If I spent $700-800 on an i7 920 platform in 2008-2010, by now at best it's worth $200. That means it would have been a real loss of value between $500-600. Instead, I went i7 860 -> i5 2500K -> i7 6700K and after reselling all those parts and reinvesting the resale value, it cost me no more than $500-600 anyway. The difference is I had a very fast system in 2009-2010, 2011-2014 and now in 2016.

Just like right now I'll keep the i7 6700K + Z170 mobo temporarily, but then I'll just sell them in 3-4 years and may even reuse DDR4 I have in the future (more $ saved). Assuming I sell the CPU+Mobo for $225 in 3-4 years, it's exactly half of what I paid for them new in 2016. That means my TCO for that platform over 3-4 years is a mere $225. I will then take $225 from the resale value and buy an i7 9700K or i7 10700K in 2018-2019 and keep going.

That means with my strategy, I take all the hard work and stress and throw it out the window. Instead of spending $700-800+ on a new platform and sitting on it until it becomes nearly worthless, I just upgrade more frequently and reinvest the resale value. At the end of the day over 6-8 years, your TCO and mine are very similar but I have close to top-of-the-line rig every 3 years and yours is not.

So I don't even need to expalin that Skylake/Broadwell-E is a mediocre upgrade compared to going from Pentium II 233mhz MMX to Athlon XP1600+ or moving from Pentium 4 C 2.6Ghz @ 3.2Ghz to E6400 @ 3.4Ghz because holding old parts and letting them become worthless costs more or less than the same as does upgrading more often. Problem solved.

This TCO concept can be applied to cars, headphones, smartphones, videocards, etc.

Obviously it depends on the person. As he indicated by saying he wouldn't take a free 2600k+MB, he has no interest in building a system for fun. You're just looking at the cash cost of the upgrades you've done. How much time did you spend researching and picking out parts for your Sandy Bridge and Skylake build, looking for deals, and buying the parts? How much time was then spent dismantling the old build and packing away the parts? If you did a clean install, how much time was spend installing the new OS, applying updates and drivers, moving personal files, reinstalling applications. Even after that, how much time was spent taking pictures of the old parts, posting them for sale, dealing with buyers, and actually completing the sale?

I get your position, because I love building systems too. It's a hobby though. If I actually look at the amount of hours I spend on doing a complete system update factoring in all the time above, it can easily be well over 20 hours. Really, if I didn't spend the time already browsing forums like this it could likely be more. How do you value the thousands of dollars worth of time that each upgrade costs?

 

[monstercameron]

I read the whole thing. The guy you quoted said visit small business and you'll see desktops. You said no you won't, you'll see tablets projecting to a large screen. He's right, you're wrong.

The macys that i used to work at switched most of the staff to lenovo thinkpad tablets and the jamaican restaurant uses an android tablet as apart of their pos system. It may be a trend.

 

[pj-]

And VR will be in the same spot soon. Anything that gets heavily saturated (like those MMORPG's trying to copy World of Warcraft's success) wears out fast.

Neither of those examples apply to VR and the examples aren't even similar to one another. Smartphone sales are decreasing not because people are tired of them, but because everybody has one. MMORPGs were a relatively long lasting fad genre of game, and people have gotten tired of them.

VR is not a fad (most people are indifferent or actively think it's stupid), and it will be decades before VR reaches the kind of saturation smartphones have. Assuming it ever does.

 

[Nothingness]

Smartphone sales are decreasing not because people are tired of them, but because everybody has one.

Smartphone sales are not decreasing in volume according to analysts. Forecasts are that volume will increase by less than 10% this year, but will increase nonetheless.

 

[iCyborg]

If I were to apply the same high standards, even for a small time-span of 1998 to 2003 where CPU speed for me went up almost 14X, might as well never upgrade again for 30 more years then. See how flawed this logic is for CPUs? "

Not really
Unless your premise (and my "flaw") is that people should upgrade (at least) every N years, no matter what the performance increase is and whether the current system satisfies their needs or not. N being a lot smaller than 30 obviously. I personally see absolutely no problem in not upgrading for 30 years if the system does what I want from it. (and doesn't break for 30 years...)

I mean, doesn't it seem logical that how fast the technology progresses would have an impact on how fast people find their systems outdated?

Again, if your PC satisfies you, that's fine. Now for every 1 of you, there is me.
Intel loses sales to customers such as yourself since you are no longer interested, but it more than makes up for it by targeting people who love and still want desktops.

At some point it was like that, and I myself was "one of you". But looking at my colleagues at work (who are all SW/HW egineers) and my friends/family, there are a lot more of me's for every one of you.
And the second sentence, I simply disagree. Are you saying that the slower pace of technology is causing enthusiasts to upgrade even faster and buy higher-end CPUs than before, thus compensating for those who hold on to their systems longer? This doesn't make sense to me, and news articles like in the OP also don't support this...

I already explained in this thread that even if not accounting once for electricity costs, it made no sense to hold on to an i7 920/i7 860 anyway if you were smart about timing your upgrades. It's very simple: Total Cost of Ownership or TCO.

1. gorion covered this above (and some others), and I also hinted at the same when I said I wouldn't take 2600K+MB for free due to the hassle. If you're someone who enjoys building systems and tinkering with HW all the time, then it makes some sense. I used to be like that in the late 90's and early 2000's. Not any more, now it's just a wasted time. I usually end up spending one whole weekend afternoon building everything, and then installing OS, and then configuring it the way I want, getting updates, installing all the applications I had before etc. I simply no longer enjoy doing this, I can think of dozens of ways I'd rather spend my time off.

2. Are you a student/MSDN subscriber who gets Windows licenses for free or something? Upgrading 3-4 times would mean $400-$500 in OS licenses which I'd need to include in my TCO calculation, and I imagine most people would.

3. About 90% of what I do nowadays is internet browsing and streaming video to TV. Having a faster system would make practically no difference.

If you start thinking about the concept of TCO, you no longer think about how you set aside $1500-2000+ on a new PC. Otherwise, you'd be making the exact same mistake as you just made with your last rig.

To summarize what I said above, there's more to TCO than just the resale values of components: time/hassle spent on upgrading/selling, OS licenses, the value of having a faster system. All-in-all, I'm not convinced that I'm making a mistake here. I'm not even complaining about the current situation - I actually *like* that I feel no need to upgrade frequently.