To pink 0, there are certian ways to get around slow boot times of Stereo's or PVR's. Micron once demonstrated a flash based DIMM (I.E. Solid state speed, plugged into a normal DDR slot) that could get windows 98 to boot in 6 seconds from power on. The real future is backing up commonly used, small files in Flash (like DLL's, drivers and the like) in a sort of flash DIMM (Micron has demonstrated that you can get flash to go *fast*). As for audio workloads, I understand that in 20 channel 24/96 recording it can get up to it can get up to several hundred I/Os per second. Even if each channel only takes up 288 KBytes per second (I've done this on a calculator. 96KHZ times 3 bytes (24 bits) equals 288KBs.) the I/O load and latency sensitivity is amazing.
There *are* sollutions for that. For instance, 4 Cheetah 15K.3's in RAID0 on a 64bit/33MHZ PCI slot with 128MB/s of cache. Stuff like that is *made* for your profession. For people with purely I/O intensive loads, there is SCSI RAID. There are 10,000RPM HDD's that go cheap, and decently priced RAID controllers. For people like you, it's an acceptable sollution, if you're willing to pay the price. That's the way you get around the I/O bottle neck. You can't make IDE spin *any* faster because the noise would drive grandma mad and the heat would require active cooling for the first time in consumer storage. It's not a sollution. SCSI RAID is the sollution. And for people like you ,there are SCSI RAID arrays.
For people like grandma, there will be Flash DIMMs. No waiting, Ma! As lythography gets better, so will the capacity of such Flash DIMMs, which micron is working on as we speak. You could even get them to work with normal motherboards. And if you think that Flash's bandwidth stinks, may I remind everyone that bootup uses alot of small files, and thus I/Os per second (Which flash schools HD's over)
And may I remind everyone, that as lithography gets more and more powerful, we will eventually be able to cache a 20 channel 24/96 recording project into RAM, and it's oversampling temp file? Remember how not *that* long ago we barley had 2 megs of RAM in machines, and now we have 2000 in the most powerful workstations? Soon a gigabyte of RAM will become standard as the newest microsoft OS's soak up all the RAM. Soon, very soon, like maybe in the next 10 years, computers will be able to cache entire symphonies of multiple channel 32 bit 192KHZ audio in their memory. And remix them down to 24/96KHZ with time and memory to spare. After all, computer memory progresses a hell of alot faster than Audio Demands do. As computers march on, some tasks become quaint, even inconsequential. People used to make coffee over the recalculation of excel spreadsheets. Now they run in the blink of an eye.
Intel with it's 6GHZ CPU is helping us to go farther into the realm of AI, voice recognition, etc... making new things possible. how would you like your computer to greet you in the morning when you wake up? Cheer ya up when you're feeling down, with it's webcam eye peering at you, trying to figure out what's wrong and how to make you feel better? That's gonna be possible with a few billion transistors and 10+ GHZ.
Storage isn't going to be the bottle neck of the future. Memory bandwidth/latency is. Have you seen the incredible rate at which Caches have outpaced even the fastest of RAMs? PC2700 isn't even sufficent to stasify the pentium4 anymore (It eats up 1MB/s per MHZ.). The gap between memory and CPU has been *increasing*, not decreasing. RAMBUS may have helped to solve that, but they were ahead of their time, had bad ethics, and screwed up a good technology . Not only that, but RAM also has to be an I/O cache. Eventually it might get to the point where we have to stick multiple hundreds of megabytes of cache on the northbridge or make a dedicated path for cache on the motherbard, like the Pentium days. And that, my friends, is where the bottleneck for CPU growth lies. I imagine soon, some time in the future, Intel might stick a 128MB 64 bit octopumped yellowstone 10GB/s RAMBUS Cache on the mobo. Maybe 128 bit pathway, Octo pumped, etc.. just to keep the CPU from starving for data. It's gonna be awfully tough pushing all the 3d data around for face emoution recognition type tasks, and there's no way RAM could handle the load. Those types of applications require huge amounts of streaming bandwidth. Bandwidth that no RAM save maybe advanced forms of RDRAM and multiplie channel DDR-II will handle. And that'll be expensive. There has to be a way to stop the huge gap between CPU and RAM. And I believe the answer lies on offboard cache like the Apple's use.
Edit:288 KB/s, not 288Kb/s.