Science What will computers be like?

[Urwumpe]

PS where's the Cray desktop everyone was promised, that could be a practical leap...

A Cray II had 1.9 GFLOPS - The best available CPU has about 120 GFLOPS today, the best available GPU 515 GFLOPS in double precision (in single precision, these GPUs are much better)

How many Crays do you need on your desktop? With some money, you can have over 600 Cray II computers on your desktop.

 

[Capt_hensley]

Only one, the fastest, so I can make realy cool realtime space flights with every view possible at over 60 FPS, using 8 or more 1080p flat panels, just so I can say I did it.

I read an article about having a Cray mini main frame, with multiple terminals...
http://www.cray.de/Products/CX/Systems.aspx

The really cool thing about Cray was they did everything in active, liquid cooled, memory, were still stuck with HDDs or SDDs, so slow, so dull. The Cray looked like a lava lamp, trey shiek!
http://www.slowlane.net/travel/gallery/Museums/1976_Cray_1A_Super_computer_at_Science_Museum
How sad, the 1A collecting dust.

But seriously, we were supposed to have an ultra fast Cray SC on our desktop by now wern't we? Along with flying cars, and mini nuclear power generators. I still long for those days.

 

[Jarvitä]

As you've just been told, you (probably) have the equivalent of several hundreds of 80s supercomputers on your desktop, and several dozens more in your pocket if you own a modern smartphone. The exponential growth of computing power per money can not be denied.

 

[T.Neo]

The exponential growth of computing power per money can not be denied.

And the presence of airliners that can fly me to another hemisphere in just a few hours also cannot be denied. Doesn't mean airliners will suddenly sprout SSMEs and fly to the Moon.

I'm still wondering what recent developments in several key areas of computing, such as memory storage, would look like if plotted on a graph...

 

[Jarvitä]

I'm still wondering what recent developments in several key areas of computing, such as memory storage, would look like if plotted on a graph...

Unsurprisingly enough, just as exponential.

Doesn't mean airliners will suddenly sprout SSMEs and fly to the Moon.

Because there is no economic imperative that would drive companies to invent the technologies and provide the economy of scale to bring us lunar spaceliners. In computing technologies, however, the incentive is very much there, inventions and breakthroughs are being made on schedule, and even with the exponential increase and accelerating returns, we are almost a full century away from hitting the limits of the physical universe as we currently understand them.

 

[T.Neo]

Unsurprisingly enough, just as exponential.

I never said anything about cost, I was talking about capability.

It is an interesting point though. Just as airliners have 'maxed out' in terms of overall capability, their efficiency has increased over the last few decades. We may start to get to a point where capability hits a 'sweet spot' of demand and feasibility, and then gets improved constantly over time.

Because there is no economic imperative that would drive companies to invent the technologies and provide the economy of scale to bring us lunar spaceliners. In computing technologies, however, the incentive is very much there, inventions and breakthroughs are being made on schedule, and even with the exponential increase and accelerating returns, we are almost a full century away from hitting the limits of the physical universe as we currently understand them.

And yet if you spoke to someone from the 1950s, they might have believed spaceliners to the Moon. Just because an economic demand is theorised about, doesn't mean it will actually exist. This is also a facet of the 'sweet spot', and although it's a lot like the "who will ever need more than so many kilobytes of RAM" argument, it can't be ignored wholesale.

We will likely hit the limits of our (current) technology long before the 100 year mark...

 

[Jarvitä]

We will likely hit the limits of our (current) technology long before the 100 year mark...

Yes, just like we hit the limits of mechanical computers, and the limits of vacuum tube technology. Paradigm shift is nothing new to the computing industry.

 

[T.Neo]

We hit the limits of piston engines and canvas stretched over wood. We went on to jet engines and riveted aluminum structures.

We could go on to scramjets and hafnium diboride, but we don't... or at least, we haven't.

Paradigm shifts come at a cost. And often they come at a cost that just isn't worth it, especially when there isn't demand- the sweet spot again.

 

[Keatah]

I believe the next paradigm shift in computing power will come from the continued evolution of parallelism AND the upcoming intel/nvidia merger. I believe the hardware will come first and continue to establish itself through basic marketing and pervasiveness. You can observe this as intel is doing it with onboard graphics in the i5/i7 series. Intel is putting the hardware out slowly and steadily. But please be aware, that unless a new computing medium is developed, the best you can hope for is "evolution" of current designs. Expect no major changes hardware wise.

Now, the shift will actually happen when programmers (with their sidekick compiler companions) learn how to write software to take advantage of multicore architecture.

Aside from specialty supercomputers, right now, these multi-core chips and data center things and render farms are pretty much non-functional and still wriggling around in the mud. The faint glimmer of sophistication that shows itself on the spec sheets is immediately covered up by bloated software. Layers and layers of API's that appeal to programmers that are too lazy to learn the real hardware. Far too much abstraction. Kick it down a few notches I say.

When that happens, and economic criteria set aside for a while, then, and only then will the promise of multicore happen.

So, in summary, the next shift will happen in software; as hardware continues to evolve on a smooth line. To have a real genuine shift in hardware, someone will have to re-invent a basic computing element called the transistor. Ever since back-in-the-day, the transistor has evolved, in chemistry and shape, but it is still a switch. Until something else that does NOT behave like a switch is built - hardware will evolve and not shift.

So for now, the programmers have the ball in their court!

 

[jedidia]

Aside from specialty supercomputers, right now, these multi-core chips and data center things and render farms are pretty much non-functional and still wriggling around in the mud. The faint glimmer of sophistication that shows itself on the spec sheets is immediately covered up by bloated software.

They will do so automatically when they have to. I.E. as soon as they know that they can't just wait for a better CPU to come out which will make their stuff run better, they will take the trouble to optimise for parallelism. It has taken the games industry several years to recognise the true potential of Graphics accelerators, but once they did there was no stopping them.

Similiarly, optimisation is having a golden age on smartphones. Look at how carefully IPhone apps get optimised (they're fighting for every single CPU-cycle!). Once the serial potential of PC-CPUs maxes out, people will be all over parallel optimisation as if their lifes depended on it, don't you worry.

 

[Keatah]

Well, in some areas of computing there is an insatiable need for more FLOPS and GLOPS and whatever.. But in others we are certainly overpowered and there is absolutely no need for further advancement.

I have used Office 2003 for many years AND I see absolutely no need to upgrade it to 2011. OTOH, there are some applications I upgrade monthly. And some hardware I'd **like** to upgrade daily!

It all depends on the need. Keep in mind, marketing is slick and will convince you, without you knowing it, that you need far more than you really do.

Once you think critically about your current hardware/software needs and current setup, you'd be surprised how mis-matched they are. How did that happen?

 

[Jarvitä]

So, in summary, the next shift will happen in software; as hardware continues to evolve on a smooth line.

The hardware only "evolves on a smooth line" on a logarithmic plot. The only way software is evolving is adding more bloat to properly abuse the exponentially improving hardware. It'd take a fundamental change in how computer science is taught to teach new programmers to properly use massively parallel systems. That'll probably happen in 10-20 years, when the current generation of tinkering children become professional engineers and professors.

 

[T.Neo]

The only way software is evolving is adding more bloat to properly abuse the exponentially improving hardware. It'd take a fundamental change in how computer science is taught to teach new programmers to properly use massively parallel systems.

Maybe, you have thought about the fact that certain things only require a certain level of capability?

As Keatah said, there is some software that there isn't really an imperative to update. And often when there is, it isn't because such software is 'exponentially better', but slightly more capable with a few new added capabilities, bug fixes and so on. In other words, it's reached the point where it is improving, rather than advancing- it has reached the 'sweet spot'.

The main drivers I can see for greater computing power are graphics (gaming, etc) and simulations, primarily in scientific and engineering fields.

A word processor for example only needs so much capability. There comes a point where getting it to use the extra capability will not only not be advantageous, but be dis-advantageous. And at that point making the software better within what it does, rather than making it bigger, will be the major factor in development.

That'll probably happen in 10-20 years, when the current generation of tinkering children become professional engineers and professors.

Some of the more cynical people may have you believe that in 10-20 years will come the global apocalypse, as the current generation of children grow up, and merge all languages of the world into a single, mutually unintelligible txtspeak...

 

[Capt_hensley]

I'd be happy that if what I click now, did something now.

I hate waiting for anything to think about what to do, then do it. Constant circling, hourglass turning, coffee making, eat dinner first, hurry up and wait, mentality of computing is getting old. I'm too old to wait, just do it now! I want to have fun now, NOW!!!!

That is to say, can't wait for technology to catch up to we humans. Currently the fastest know computer in the world. Hopefully not for long.

 

[Artlav]

I'd be happy that if what I click now, did something now.

I hate waiting for anything to think about what to do, then do it. Constant circling, hourglass turning, coffee making, eat dinner first, hurry up and wait, mentality of computing is getting old. I'm too old to wait, just do it now! I want to have fun now, NOW!!!!

Get a hardware one generation ahead of your software, or get the software one generation behind your hardware.

I have a Linux laptop with a 900Mhz chinese CPU, an SSD and a Gb of RAM, and it's about as responsive in anything that does not need good graphics (Orbiter) as electronically possible.

Another laptop i have got an 1.6Ghz Intel CPU, an SSD and 2 Gb of RAM, Windows 7 on board, and it's about as slow to respond as humanly possible in anything that does not run as a steady state (movies, Orbiter, etc).

And a big home computer with quad-core 3 Ghz Intel CPU, SSDs and 6 Gb of RAM also runs Windows 7, and it's as responsive as electrically possible in any application that does not insert artificial slowdowns.

So it's not so much a question of available technology, but a question of matching hardware and software.

 

[Capt_hensley]

When I purchased just over a year ago I had the best you could buy. After 6 weeks of using it, I was faster than the new computer. This has been the story of my computing life. It's a super computer or nothing for me.

Current config:
Zenith GA X-58 Extreme MB, I-7 920 Quad Core, 2.66 ghz(OC to 3.3ghz), 12 gig DDR3, 256gig SSD, ATI 9550, 500mg DDR3 Vid cards tri-SLI together(thats tree vid cards), win 7 sp1. This was the best available when I bought it.

I'm ready for the I-7 1000x 3.6ghz hexa-Core chip, my MoBo will take it no prob, but it's not available yet. I might upgrade the video to a single GTX 590.

What will it be like? It will be like a slow computer, just like all my other top of the line slow computers. That's what it will be like.

 

[Jarvitä]

You're clearly having trouble either selecting appropriate software for your computing hardware, or managing said software. This has nothing to do with the hardware being inadequate.

 

[staf7]

larger hard drive even though i think 1 terabyte is WAYYYYYYYYYYY more than what anyone would need

possibly an Exabyte of space depending on how much further you go into the future

 

[Capt_hensley]

You're clearly having trouble either selecting appropriate software for your computing hardware, or managing said software. This has nothing to do with the hardware being inadequate.

I like to think that I install the software, bare bones style, I only implement approved updates, and try not to overload the OS with BS packages that just take up space and cause congestion.

I keep TSRs down to less than 5, my AV, Video card, and Audio controller are the only thing I allow in the Task-bar. Windows, puts in the network and Volume for me.

I use Avast, and turn off active scanning when I'm working on graphics. I turn it back on when I surf. I make sure I do a full scan every month, and use SpyBot online as well.

I use everything I install, and I only install what I need to get the job done.

The biggest trouble I think is what I do. Graphics design. All the programs have to write arrays out, this is where the CPU and memory of the GPU get a real workout. Even with tri-SLI the cards have to take some time to report the render. On average about 3 seconds, on a bad day about 20 seconds.

Graphics software:
Autodesk 3DS Max 2012
Adobe CS 4 Design Suite
Corel Paintshop Pro Photo 3x
Mesh Editor 2

Orbiter is the only game/sim I have installed, I get about 25/30 FPS on a good add-on, but the tri-SLI should be able to handle a lot more.

Productivity:
MS Office 2010 Home and Student
IE 8.0
Expression Web 4
Visual Studio 2010 Pro
7Zip
Win Rar

If a future computer were to work for me it would have to handle all this and more, and report renders in less than 1 sec. It would also give Orbiter over 30 FPS without the occasional hiccup.

 

[Pyromaniac605]

larger hard drive even though i think 1 terabyte is WAYYYYYYYYYYY more than what anyone would need

:blink:
You think so? I'm sure that 1Tb is pretty average nowadays...
I have 1.5Tb and it's getting used up pretty fast.