Slashdot Mirror
What Applications Will Drive System Performance?
Foredecker asks: "Companies like AMD, Intel, NVIDIA, ATI and others are continuing to drive silicon performance to new levels. Of course, every day computing (basic web browsing, email, word processing, spreadsheets, personal finance, and the like) don't require a Intel 3.2Ghz P4 with Hyperthreading or a AMD Athlon 64 FX and their associated platforms. Of course, there are apps that will leverage today's high performance platforms. Games are an obvious category, as is video editing. I'm looking for apps that will be widely adopted and will drive volume hardware shipments. Things that come to mind are: effective, speaker independent voice recognition, accurate repeatable object recognition in digital photos and videos (or from live feeds such as web cams). What other application categories are there that will drive the need for bigger-faster-better hardware platforms?"
I mean, what the hell? This isn't the first time I've seen this kind of thing. Why are the guys who are hired to pinch pennies in corporate always the ones who aren't happy until their toilets flush with freshly imported springwater?
Similarly, the applications running in curtained memory are going to stack up at an alarming rate once Longhorn and other platforms start to see pervasive digital rights management. As every bit of data being generated or passed from application to application is being tested against dozens of different filters, CPU time is going to go up in smoke, and it will be illegal to stop these activities from taking place in most countries.
Games and pr0n.
And pr0n games.
I see you haven't had the opportunity to experience Gnome....
Anyone else notice how a system will pause when you put in a cdrom or format a floppy? This goes away when you have an SMP system. HT enabled p4's also remove this to some degree.
I'd like to see some typical performance on these types of activities, things that can "pause" a system for a couple seconds.
Loading websites with tons of thumbnails, searching hardrives with/without indexes (search pauses explorer). Programs that can Spike the CPU, use up all the buffer on a device, peg out virtual memory, freezes programs so you cant switch between them.
More multitasking benchmarks with responsiveness being goal. All benchmarks I see are geared around 1 app, how fast can you go, not how smooth can it go. This is why everyone is so interested in Linux kernel 2.4 MM patches or 2.6 low latency patches, to make the system smooth and responsive. People notice these "lags" in windows and linux.
Animated 3-dimensional paperclips, obviously.
Right now CGI is still expensive enough that most independents and hobbyists don't include it in their films. Affordable, rapid CGI could be a possible killer-ap for high-performance hardware. Currently, professional moviemakers must agonize over the creation of any CGI effect. It's a tedious process that involves using wire-frame animation, rendering and so on. If this process could be speeded up and simplifed, it might encourage more widespread adoption of CGI effects among hobbyists, giving them the ability to make movies they never could have before with their limited budgets.
Imagine being able to 'direct' a VR character almost as easily as you would direct a real life actor. When the technology gets to that level, we could see an explosion of new movies by people outside the Hollywood cookie cutter. Filmmakers with radicially new ideas who are too young to have developed a 'rep' in Hollywood could be creating some very professional looking films. Think of it this way: right now there are lots of people who write fanfics of their favorite movies or TV shows. But actually creating an episode of Star Trek, for example, is just not possible right now. With improved technology, perhaps these creative individuals might very well be able to make their own episodes, largely using CGI. Imagine taking a sci-fi movie that you like for the most part but hated the ending of. You load your CGI software with images of the main characters and battleships from the DVD and create CGI models of them. Now you can create a new ending that's more to your liking. Better yet, you can burn the new version of the movie with your ending (forget the "Director's Cut", this is "Mike's Cut") onto DVD and trade with your friends.
Right now we are all still pretty much at the mercy of Hollywood to make films that we like. Very soon, the balance of power will shift and creative individuals who have lots of ideas but budgets nowhere near those of studios will be able to create some very impressive looking films. And then Hollywood will have to get their ass in gear and show us something that we couldn't do ourselves in our own living room.
GMD
P.S.: Several people have mentioned that pornography has historically been a big driver of technology. Can you imagine that boom that the adult market will get when people can make their own adult films using CGI characters? Think plots of porn flicks are stupid? Wish for something better? Hell, just load your CGI software with images of Jenna Jameson and make your own film with her as the star.
watch this
What applications will drive the market? The same ones we have now, but written in python using bubblesort.
Well, maybe not bubblesort. But non-optimal algorithms, slow languages, and copying memory instead of passing a reference/pointer will increase.
That's not even a totally awful thing, because it means that the performance of already efficient applicatoins/languages/coders goes up as well. It also means that it's easier to start writing code without mastering TAOCP first.
So encourage your friends to write game engines in perl.
Processing power becomes very important at the two extremes of the Hertz chart :
.. ditto. Decrypting a message saying that Pearl Harbot is going to get bombed in 12 hours doesn't do us much good if it takes 14 hours to decrypt.
Things measured in units per second (ie, frames per second, transactions per second, connections per second) will always benefit by faster performance on a faster machine.
Things measured in many 10's of minutes (ie, an hour or more to process one transaction) will also benefit from a faster box. This would be cryptography, video compression codecs, and physics models.
When the transaction time is more than 1 minute but less than 10 minutes you really do not gain anything by increasing the performance of the machine (unless you can increase it to the point it runs in less than a minute. If you compile code on a computer that takes 7 minutes to compile it, buying a new computer that is twice as fast still has you compiling for 4 minutes. No real difference between the two, really, from a user's perspective.
When the transactions are measured in per second, the difference between 15fps and 30fps is the difference between unusable, and usable - particularly when we are talking about first person shooters. The difference between processing 150 visitors a second and 300 visitors a second is the difference between getting slashdotted and not.
When the transactions are measured in hours, being able to double the performance makes the difference in whether or not a particular transaction is even possible. Nightly backups are not particularly effective if they take 28 hours to process. Nightly runs of an accounting system
As long as we have applications that take more than an hour to run, and as long as we are measuring applications in X per second (frames per second in the range of 1 to 100, transactions per second of more than 1,000 and less than 50,000) - we will benefit by having faster computers.
Glonoinha the MebiByte Slayer
Almost forgot.
Any difference is performance that requires a stopwatch or a special timing demo application to measure - isn't a difference.
183fps = 200fps in Quake.
pc3200 RAM = pc2700 RAM = pc3500 RAM
28fps in UT2003 = 30fps in UT2003
specINT 93158 = specINT 96452
Pentium4 3.06GHz = Pentium4 3.2GHz = Pentium4 2.8GHz.
Until you are talking about performance games of roughly 300%, or one machine being 3x as fast as another machine - it isn't worth replacing the machine. It would be silly to replace a 486DX2-50 with a 486DX2-66, even though you would get a 30% boost in processor speed. You wouldn't replace a PII/300 with a PII/366 even though you would get a 20% boost in processing speed - they are effectively the same speed and you probably wouldn't notice the difference. Replace that PII/300 with a PIII/900 though, go 3x as fast, and all of a sudden you can see big differences and a major improvement. Same thing with replacing the PIII/900 with a P4/2.8GHz.
Glonoinha the MebiByte Slayer
A good example would be Palm-type devices. As big-processor speed increases, there is also an increase in small-processor speed and efficiency (limited more by heat than anything else). This has given people a smaller, more powerful Palm-type device today than they could have bought five years ago. Another example is the DVR/PVR. The new two-tuner satellite HDTV receiver/recorders can handle the receipt and recording of two high-definition streams while decoding and playing back a third - my ancient Showstopper (ReplayTV), on the other hand, starts to chug when encoding/saving an NTSC transmission (at highest quality) while watching another (I paid &700 for my 20-GB Showstopper back "in the day" while the new 250-GB HDTV units will go for $1,000 and come down from there).
It will be interesting to see how long non-PC devices take to catch up to current top PC speeds and what applications (especially portable, non-notebook apps) spring from that.
Language translations.
I recall seeing an algorithm that partially ignored traditional dictionary-type translations and relied more on a relational database. For example, rather than work word by word through a given sentence, it attempted to relate that sentence to other sentences and solve in that matter. If it sounds confusing, it's mostly because I read about it quite a while back, and really can't recall most of the details. A sentence such as "Comment Allez-Vous?" would literally translate as "how are you going", or something to that effect (Allez is the second-person plural of 'to go' in French), but is obviously more colloquially translated as "How's it going?" Rather than concern itself with the meanings of the individual words, this algorithm would know the meaning of that phrase and use it as sort-of guidelines for how an unkown phrase would translate. And I'm sure doing that properly, in realtime, with no errors would require a ludicrous amount of processor power and be ridiculously useful. Go ahead an couple that with the above-mentioned truly accurate voice recognition and you've got the legitimate workings of a device most would consider to be science fiction.
To answer the question:"What Applications Will Drive System Performance?" Scientists and engineers need *raw* computing power. They need lots of RAM (and even more) and they need GHz, lots of them. They need fast GPUs for 0. Consider Computational Fluid Dynamics. Most problems today are calculated on off the shelf PC hardware. I know this for a fact. Today, you don't buy Indys or Sparcs anymore, they're much too expensive bang-for-bug wise. They're nothing but dead. People use dual Athlons with 3 Gigs or RAM to run their jobs, mostly one job per CPU, because parallel processing in most cases is not there yet (solving the Navier Stokes equations in parallel is still a major PITA). So in
short, those people will always buy the fastest PC stuff available, because for them it makes a huge difference whether a solution converges in two or in four days.