Continuing along current trends we can expect our algorithms to run many times faster on Desktop Computers in 10 years (let's pick a number) 1024x faster.
Do you know of any algorithms that are within 1024 times of running on average Desktop Systems and that would dramatically change the kinds of software we can write there?
I have my thoughts, but I'm more interested in the community's ideas.
I don't know for sure but I am pretty sure salesmen wouldnt have trouble traveling anymore :)
Raytracing for games is certainly within three orders of magnitude of feasibility.
First one quibble 1024 orders of magnitude would be 10^1024x faster, not 1024x.
Clearly certain of graphics processing (games!) would benefit. Also video and audio processing software would do well.
Factoring - and consequently hacking public/private keys would be quite a bit faster and the typical key size would be moving up commensurately.
A lot of applications/systems won't be bound by CPU (and aren't now) - disk and network IO will continue to be the throttle.
We'll just eat up those advances with more layers of indirection, as we've done in the past.
Software that was a whole lot less efficient, yet still faster.
(Or is that just me?)
Something that would use a BOINC-like project even more effectively, i.e. calculate and fight AIDS, cancer or solve other scientific problems.
According to past experience: 1024 slower and larger versions of what we do now.
Ten years ago, my desktop machine did almost exactly what my current machine does. The big difference is that it now connects to way more computers (network applications) and more smaller devices (mostly iPod and cameras). Also, games have more polygons. I don't see why that will change anytime soon.
This guy http://en.wikipedia.org/wiki/Ray_Kurzweil has something to say about that in his book http://en.wikipedia.org/wiki/The_Age_of_Intelligent_Machines.
Freely available here along with pretty much everything else the dude's written. http://www.kurzweilai.net/meme/frame.html?m=10
Software will grow to consume its newfound resources. I, on the other hand will still be trying to resist feature creep and write the smallest, lightest application I can.
I'm pretty sure we'll see a lot more poorly written software by amateurs with the latest point and click RAD tool; simply because computers are so fast that the difference between efficient well written and inefficient software is undetectable. At the moment the difference is a matter of a second or two in response time for most things.
The notion of Garbage Collection will be extended to non-memory resources, such as open files and database connections.
As programmers we'll use software that will analyze our code for bugs in ways we only do as humans today. Think what lint / compiler warnings / PreFAST do, but 1024x as much analysis.
As you type your code, you will get immediate feedback about not just compile errors, but also test results: you'll know immediately which unit tests you've broken.
We'll be able to get that much closer to modeling every neuron in an entire human brain. See you soon, SkyNet!
In writing real-time video automation software for the broadcast industry, the speed of the platform is a huge consideration. In fact we deliberately develop on under-spec'd machines so that our software will over-perform when deployed to end-consumers.
There is only so much that you can reasonably do in 1/25th of a second in the PAL world (or 1/30th for NTSC, or 1/50th for HD, etc). So this imposes a considerable limitation on implementing our creative ideas and moving the industry ahead.
The move from hardware to software-based CODECs has been the most significant advance for us in recent years and I can see this as a major factor as platforms get faster. Real-time rendering offers almost limitless possibilities for broadcasters, especially in the HD domain. But it is quite a way off being a reality.
The question is directed at 'Desktop Computers' and you may think that I am referring to Server technology, however it is much of the same when it comes to development.
To answer the question: Mashing large amounts of meta-data with media streams in real-time. That’s the holy grail for us and x1024 would certainly help. Just can’t wait 10 years for it.
Read "The Singularity is Near" by Ray Kurtzweil if you want a really in-depth answer to the question.
Better OCR would mean that office software could become gigantic self-organizing searchable indexes of scanned documents. Imagine scanning something that looked like a speadsheet and then having it behave like one.
UML diagrams that get generated by examining a digital photograph of a whiteboard.
How 'bout ... the fabled "WinFS" that does away with a hierarchy of directories in favor of a self-organizing cloud of all of your documents/entities.
Actually, I was asked a few years back to help [large semiconductor company] answer this exact question - what are the ('consumer') uses for what they termed 'terabit/terahertz computing'. Sadly, I don't have my notes with me right now, but if I recall correctly, some of the conclusions were:
There are some hard tasks in the area of machine learning and artificial intelligence that will benefit from faster computers, for example:
To illustrate: Speech recognizers have a language model built in. For all words that the acoustic model hypothesizes it decides how likely they would actually be, given a number of previous words that have already been hypothesized. This search space of previous words becomes very big, and usually the language model will just look at the hypotheses for the previous two words, so it deals with trigrams. Otherwise search would take too long. But if speed (and hopefully, memory) won't be an issue it can take much more information into account (i.e. use 6-grams, for example, or models that are better than simple Markov Chains altogether), and make more accurate predictions about what was spoken into the microphone.
Encryption algorithms! 1000 times faster computers would ease brute forcing and other exploitation methods for sure. Security specialists will sure be under more pressure than the specialists of other software areas.
Maybe the past can help tell us what the future will bring...
I did some back of the envelope math with numbers I found on the net (not sure how reliable the sources are, but I think it will support my argument)....
My first PC was a 486 33sx. On the wikipedia I found that a 486 33sx performed 27 mips.
My current PC is an i7 940. The best source I could find on i7 was for the 965... 76,383 mips
76,383/27 = 2829, so well over 2x the figure stated in the question (1024).
On my 486 I: Used Windows, played games, tinkered with development (I sucked back then, but I was young), and used AOL (this was the early 90s).
On my i7 I: Use Windows, play games, tinker with development (I suck a little less now), and browse the web.
I predict that in 10 years I will: Use Windows, play games, tinker with development (but be pretty good this time), and browse the web (from my flying car, which will be invented by 2015 according to Back to the Future II)
That depends a lot on your definition of "faster"
We've reached somewhat of a limit on cpu speed due to things such as the speed of light
We will start to see more many way processors which are executing many independent pipelines of execution simultaneously. Only programs designed to benefit from this new way of thinking about architecture will be able to fully benefit from the "faster" machines of the future.
Just my two cents..
The SETI project will eventually find an extra-terrestrial intelligence.
I'm sorry to disagree, but this is infeasible any time soon.
However, we will probably have 1024x more cores and we should really think of ways to make good threaded programs that don't context switch too much and/or fight over data in memory.
Software is getting slower more rapidly than hardware becomes faster.
Probably we can also throw in some Murphy's laws... and the future does not look bright anymore ;)
It will be much easier to write poor software with acceptable performance, but writing something that could really exploit the speed increase, multiple cores and other not-yet-known technologies will be kind of a rocket science.
With current trends in programming, "optimize costs versus code quantity cost", we would have 1024 times slower code but 1024 times more functionality!
Much slower, much higher level languages. Development would be very fast and the results would be very CPU intensive.
Look at the original Doom source code, the one designed to run fast on 386 DOS machines with 16k of RAM. If we still wrote software like they did back then, the computer "experience" may well be 1024 times faster, but instead we've traded raw speed for dev speed.
Good deal? Maybe.
Wouldn't make any difference: my computer is not the bottleneck. It's waiting on I/O. Until/unless systems programmers start doing everything asynchronously, more speed isn't going to help me.
Real-time video effects (this assumes that video resolution will not be 1024x larger, since we won't need that kind of resolution for everyday viewing)
real time ray tracing. I don't think we can do real time global illumination at 1024x...
simpler designs. we could remove a lot of old optimizations as certain technologies get faster (hint hint ssd)
more parallelism, since we are reaching physical limitations on single core processors
Vastly better compilers.
I've been trying out code contracts for .net, and the main problem is that it takes forever to compile! Compile times went from seconds to minutes (on my old laptop). If computers were 1024 times faster, that issue would become irrelevant.
Also, instead of proving a method correct, a compiler could take a well-specified contract and generate a function to meet it. That would be an expensive operation by today's standards, but extremely useful. Sortof a compile-time prolog.
Virtualization will become absolutely commonplace for various purposes:
Weather forecasting will include trillions of more data-points, but it will still rain on what was supposed to be a sunny day.
Faster processing power is just that.
Give us more processing power, we'll write slower software :) An example : When I'm programming in Visual Studio and I type a dot ('.'), sometimes it takes up to 2 seconds for my computer to respond because it's looking for an auto-complete item in a big map. 2 Seconds on a current PC would mean 5 months on a 1978 computer. In that 2 seconds used for autocompletion someone else calculated an entire trajectory for the Apollo mission about 40 years ago ! :-D
Our bottleneck is (by now, taking a dual- or quad-core as 'normal' desktop) most of the times somewhere else. For example our processing power gives us the opportunity to calculate lots of stuff, but most of the time we need a lot of input/output data - so we're limited by ram/rom.
If I look at my (pretty average) desktop right now for example, it's a Q6600 at 4x 2.4Ghz. Pretty damn fast for someone who started on a ZX Spectrum :) But the best way to improve my systems speed right now is/was getting faster IO. I could buy a new CPU that would give me a 4 (cores) x 800Mhz speed bump, but I probably would hardly notice it with the current software.
Installing a striped raid hd-combo on the other hand improved the (perceived) speed of my machine a lot. If I'd be downloading at 8MB/sec, normally I'd notice my machine would be sluggish since it would spend a lot of time on IO on that single HD. Now it's divided by 2, which improves responsiveness a lot.
To cut my long story short, processing power is just one aspect - I'd rather see fast SSD's being introduced for the masses in the next year :)
Fast image indexing and searching, structure from motion and from static images in a fraction of the time it takes now, automatic essay-writing software, that would gather and combine information, inferring conclusions... quantum computer simulators with a small number of qubits (for illustrative purpose).
Evaluating the possibility of building a space elevator to link up the Space and the Earth. Here is my new idea.
To build the space elevator, a backbone is needed, I would like to suggest that hooking up the Earth and the Moon by a ~384,400km Carbon Nanotube cable.
With the super computer you mentioned, I would like to invite scientists, physicians & mathematician to simulate this crazy idea, to find out a way to work it out.
I think this is a key step for human being, and it is the very first step for connecting us to the Space.
For more details, see http://raptortalks-savetheworld.blogspot.com
p.s. I think 1024 times faster is still nothing. Human brain is still too slow to use computers. Our computer is 99% idle and waiting our commands.