Why Moore's Law, not mobility, is killing the PC
networkworld.com
networkworld.com
But they're wrong in general. Memory capacity has continued to climb (from 2 GB with Vista, to 4 GB with Win7, to 8 GB today). I wouldn't build a PC today without 16 GB of RAM.
Then we get to talk about the MASSIVE amount of improvement in IO in the last few years. The 7200 RPM HDD is dead. SSDs are now affordable. You can get a 256 GB SSD for £100.
When consumers start to experience an SSD system compared to their old "slow" HDD system, they will be at the stores in droves wanting to upgrade. SSD's speed improvement is very noticeable indeed, like night and day.
Where I work, we're stuffing users into 2GB/1core VDI instances. They work fine. The only new client purchases are replacements of dead stuff, devices that improve mobility, and some dev/kiosk machines.
If I'm understanding your situation correctly, how do your thin clients work out in practice, and, how do people deal with scenarios in which they would normally use laptops? (Business Trips, Airplane Flights, etc...)
Now you can deliver a decent desktop through a VMWare VM-centric solution or a Citrix/Microsoft Server computer solution.
The use cases work out like this: Stationary workers get PCs reimaged with a locked down OS...PCs get replaced by thin clients over time; Mobile workers either get laptops with remote access clients or tablets. The mobile folks love it, because they aren't tied down to some horrific work-issued laptop.
The big remaining problem is Microsoft licensing. Unless you have an EA, the licensing is both obtuse and expensive. Lots of big enterprises are rolling out thousands of VDI desktops -- 2011/2012 was the point where the financial case started to make sense.
My girlfriend has some cheap laptop which is also a couple of years old but its more than enough for her web surfing and office needs.
Good thing I have a second laptop that I could continue to use for normal tasks at the same time.
It was worth the SSD upgrade for Outlook alone. Import a 6GB PST file in 20 minutes instead of 6 hours.
Memory has improved with DDR technology. (What are we up to now... DDR4?) Disk is improving with SSDs, which offer a huge speed improvement. If you've never used a box with a SSD in it, it must be seen to be believed.
But the real bottleneck I think is the Internet. Imagine what we could do with gigabit peer-to-peer IPv6 networks in areas like games, collaborative development tools, teleconferencing, etc. Things like virtual reality "metaverses" could really arrive, etc. These in turn would require more CPU power and parallelism. Super-fast gigabit+ WANs would deliver the killer apps that we need to keep pushing the speed barrier.
Once again, it is the Internet carriers that are the barrier to all progress. Breaking the telco oligopolies is the thing that must happen to unlock the future. Everything-- maybe even the continuation of Moore's Law-- depends on that. The telcos, cable companies, and wireless carriers are the trolls guarding the bridge to the future.
Now imagine your data bill after just a couple hours of live streaming HD video. Something's gotta change. The bandwidth keeps increasing as technology improves, but the price per bit is actually getting higher.
I really think Google Fiber nationwide would lead to killer apps that would revitalize the PC and the high-end chip market. I'm thinking of rich constant collaboration, virtual reality, next generations of MMORPGs where you can really richly interact with the world's physics (e.g. shaping materials, in-world chemistry, etc.), tele-presence, digital telepathy with brain-computer interfaces, distributed supercomputing, ... lots of stuff. All those things are things that are CPU-hungry. Many are embarrassingly parallel, so they can utilize many-core chips efficiently.
If Intel, AMD, Dell, Apple's PC/laptop division, IBM, etc. knew what was good for them they'd get together and invest in gigabit+ broadband nationwide. It's also got to be network-neutral, since all the things I mentioned above are at least somewhat P2P. The PC industry is spiraling down because of a lack of killer apps for high-end boxes. Most of these are waiting on fast networks.
And if we look at local desktop just as a dumb terminal and using the cloud as the PC , offering compressed display doesn't take that much bandwidth.
And i hadn't seen any really interesting apps in 50/100 mbps networks available around the world.
One example that keeps coming to mind is "real" virtual reality (including next-generation MMORPGs and tele-presence stuff). Consider the bandwidth requirements of a world whose physics are malleable: in-world chemistry, in-world shaping of novel materials, in-world machines that actually work in-world or that function by means of realtime-modifiable embedded scripting, etc. All that state would have to be synchronized across millions of clients in near-realtime, requiring things like vastly distributed P2P "rumor mill" replication protocols.
You can do that with Google Fiber grade Internet access. You can't do that with what we have now.
And the economics will be quite nice for the game developer since they pay per use for onlive servers. Said economics would also be nice (at least in the beginning) , for users who wouldn't need to purchase an expensive machine.
If you have a street full of families with everyone all streaming 1080p near constantly you are going to need to greater capacity.
And that's only at 1080p, what about high resolution 4K displays etc? What about when as well as keyboard and mouse input you want to send output for high precision accelerometers , biometric data etc etc.
What I'm ultimately talking about is a "640k is enough for anyone" mind-lock mentality around bandwidth. There are all kinds of possibilities when you have hundreds of megabits or more to work with.
The set of things for which you need a superfast machine is really very small (compared to the whole,) especially if you are happy working within the limitations of the tools you have to hand, which covers a huge quantity of creative people, and most of the really good ones.
Following specs came from Google[1] the other year when ICS was released:
-6GB of download.
-25GB disk space to do a single build.
-80GB disk space to build all AOSP configs at the same time.
-16GB RAM recommended, more preferred, anything less will measurably benefit from using an SSD.
-5+ hours of CPU time for a single build, 25+ minutes of wall time, as measured on my workstation (dual-E5620 i.e. 2x quad-core 2.4GHz HT, with 24GB of RAM, no SSD).
Lots of RAM allows you to switch the compile output over to a ramdrive (or if not, move it to a separate SSD if you have two). Moving the entire source to a ram drive or fast ssd also makes editing much faster (especially the initial load into an IDE like Intellij or Eclipse) or grepping through the codebase. Doing it on anything less becomes somewhat tedious from my experience.
[1] https://groups.google.com/forum/?hl=en&fromgroups=#!topi...
I've just never felt the need to lean on IT to upgrade/replace my current Desktop though. Runs just fine.
Also, 4GB is plenty for Windows XP. Perhaps you have not used it for so long now that you've forgotten, but it only ever required 64MB and ran perfectly fine with 128MB-256MB even with high performance apps. All that has happened since is a factor increase in memory requirement for a limited set of applications.
With Linux guest instances, it is easy to cap and keep very low the maximum memory usage. 6x VM with 256MB each is still only 1.5GB.
Without a change in software requirements, I doubt many consumers would feel the need to upgrade anything. CPUs/PCs really have been fast enough for many years now.
Also, yes, virtualization wasn't always needed, but without it anything slightly CPU intensive will be awfully slow. I can see it working for CLI-only linux/bsd systems, but not much else. Do you ever compile something bigger on your guest systems as an example?
I do all my interactive work on a (Circa 2010) Macbook Air - the Windows XP system is for Visio, Dynamips, Vmware Workstation, PowerPoint, Visio, Microsoft Word, and Outlook+Lookout (I have 10 years of Exchange PST's cached on that box) - I don't think a faster system would give me much more than 10% improvement in performance - because I'm never really CPU bound on that box. And, honestly, I spend maybe an hour a day on it - the other seven I spend on my MacBook Air. I just don't see any justification to upgrade it. It gets the job done, and it's fast enough.
With all that said - the upgrade from my 2010 MacBook Pro to my SSD based 2010 MacBook Air was an awesome performance jump. Totally worth it. CPUs may not make much of a difference but WOW, do SSDs in a MacBook rock your world!
Multicore has been the best "next step" they've come up with, but the required difficult software rewrites prevented consumers from seeing speedups.
Total on-chip exceution resources have essentially kept on the Moore's law path as CPU vendors punted to multicore designs. 99% of PC software is serial since parallel software is too hard to make.
Runing the same code faster was the reason that software used to become more hungry each CPU generation and forced people to upgrade their machines. New CPUs used to make the programmer more productive, not less.
My 7 years old laptop has higher display resolutions than anything on the market (except two new laptops). It is also impossible to buy laptop with 4:3 screen ratio.
My 2 years old desktop CPU overclocked at 5GHz is faster than anything on the market, last two CPU generations can not reach such performance.
And some newest SSD generation actually have worse performance characteristics than previous generation.
So true, and it applies to Macs too!
Long ago, I was looking forward to a 17" MBP but Apple discontinued it.
My 2008 15" MBP has an Intel SSD plus a 2nd large spinning drive (optibay), mate display and a removable battery.
Sometimes I wish I could just keep buying spare parts for it forever. Too bad you can't upgrade the CPU too.
That's a fairly extreme overclock! I imagine you must be able to get some impressive results from a modern i7 with the right cooling though?
I remember when overclocking was a thing that anybody into PC gaming did for economical reasons, these days it hardly seems worth the effort of fiddling around in the bios menu.
Intel does not bother soldering heatsink into chip anymore (saves 50 cents). So newer CPU can not be overclocked without disassembling and replacing some parts.
Results are not bad, but in reality it is just 30% speedup.
Docking stations have been tried before, and they didn't work particularly well, but the time's right now for a great tablet and a great docking station. The dock needs to have keyboard & mouse inputs, network connectivity, and be able to output to a nice monitor. (Allowing user to have two monitors?).
Someone needs to sell that as a kit, with some kind of Office suite installed.
MS could, if they put a lot of effort into trimming down MS Office, make a lot of money here. (As could a Lenovo / LibreOffice collaboration. Because although Excel is still much better than Libre's spreadsheet it's good enough for many people.)
It makes it my laptop essentially much more portable since I don't have to worry about disconnecting reconnecting a monitor, speakers keyboard, mouse external hdd, and power.
There is only a small cost saving since the expensive thing is going to be the peripherals and screen and you now have to deal with problems like employees taking the tablets home and losing/breaking them , getting malware on them or filling them up with personal stuff.
The cost of having a few computer components nailed to the desk is probably less than the loss of productivity when somebody can't work because their tablet is broken.
Are there many companies who are doing full on bring your own hardware apart from maybe startups etc who can't afford to provide extra computers to staff?
These days the devices that interest me are those that consume little power, have perpetual internet connectivity (WiFi or cell tower), and are instant on/off. It has to be portable and durable. Everything else is icing.
So the most common example would be games, with the new console refresh we might see a bump in technical requirements for PC games in order to keep up.
Perhaps we will see a hybrid model in which your phone gets additional CPU and RAM to work with when plugged in?
What about google glasses and other devices? The PC should find a way to interface with that too.
Yes, but indirectly over the network.
The hybrid model I've been using for about 20 years now is connect a wimpy machine to a big one using SSH/NFS and a couple other technologies. For a decade or so the end users at wife's employer have been using rdesktop/vnc for their hybrid solution.
Its actually "easier" for enduser devices to connect to a (private?) cloud than to try to do stuff themselves. So that's the more likely hybrid model.
I don't run exchange server at home, or even run exchange client at home. Exchange webmail, yes.
It depends on the latency tolerance that a use case might have. Why not have all of these above?
The only way to keep pushing performance would be for software writers to invent new and compelling things to do that genuinely require more power. It can be something as simple as a game-- I was a teenager when Quake came out, and I remember scraping together as much money as I could to get the absolute fastest machine I could (A P120) to play it. Before Quake came out, my old 486 DX2/66 was fast enough for the Linux hacking, adventure games, and BBS/Internet stuff I was doing at the time.
We need another Quake... or something equally cool. We need killer apps. Otherwise computers are going to go over like the aerospace industry did. We stopped pushing the envelope there in the 70s, again for economic reasons.
* What are the computational bottlenecks for driverless cars? Vector operations?
* If every car needs a high performance computer to drive how will this change the semi-conductor landscape?
Maybe the sandbox genre (Minecraft, Dwarf Fortress) will grow. Simulating huge worlds requires lots of processing power.
(For those who want to trot out the "GAs don't work well," "evolution is slow," etc. mantras, please delve into the literature first. EC has some interesting and unique results, the most important of which is that it's the only "AI" technique I'm aware of that can demonstrate true creativity. EC can invent novel structures ex-nihilo. Everything else can just optimize, find maxima, etc. Another way of saying it is that EC can find maxima involving unknown, novel parameters that it invents along the way, while everything else requires that you know all the parameters to be optimized ahead of time. So EC can work with unknown unknowns, not just unknown values of knowns.)
I'm nobody, but I have opined about this and the insipid taint of Good Enough several times [1]. Like "api" at the top of this comment thread, I feel there are bottlenecks that keep the desktop PC, and the laptop PC to a degree, constrained to the roles that we have presently envisioned for those devices.
My opinion is that desktop display technology is ultimately the root of desktop stagnation and, speaking positively, could be the root of a desktop renaissance [2]. I agree that bandwidth progress has slowed and storage capacity progress has slowed [3], but point out that only display technology has actually outright stopped and regressed. In 2001, the IBM T220/221 monitor achieved high-DPI on the desktop (3840x2400 at 22 inches). Since then, we banished high DPI from the desktop and obsessed about a marketing term named "HD" because we allowed the rules that govern our living rooms' displays to take over on our desktop. That we accept 24" desktop monitors with 1080 vertical lines in 2013 is heartbreaking.
I agree with api that bandwidth is to blame. But bandwidth hasn't stopped progressing and regressed (well, with one exception: mobile data caps). More importantly, I feel that behind the bandwidth complacency is display complacency. Put a 40" high-DPI desktop monitor on your desktop and I think the demand for ultra-high-definition content, high-clarity user interfaces, bandwidth, computation power, and storage capacity will all increase. Intel sells more processors, HP/Dell/Lenovo sell more PCs, and oh-boy would I be so happy to finally get a real monitor upgrade.
[1] http://tiamat.tsotech.com/technology-sucks
Lol ya this must be the reason. I bet that they have a hand in the self-driving cars too. Its all an excuse to require more computing power.
It might be a business but they never cared about money. The push these technologies to people out of pure love of new technology.