Over the next six months or so, you’re going to hear a lot about flash memory being used as a techy Hamburger Helper.
You’ll also see articles like this one saying essentially, “Performance? What performance?”
What to think?
The first fact to realize is that flash memory is very, very good at finding things quickly, and very, very bad at moving them back and forth quickly.
With current flash RAM, it can find a piece of data much, much faster than a hard drive, but it’s a lot slower than a hard drive in getting it to the CPU, or writing data from the CPU.
So flash giveth, and flash taketh away, and especially for fairly sizable chunks of data, flash can taketh away more than it giveth.
The overall key factor to whether or not flash will be any good to you, or even any good at all, is how much faster flash can get and does get over the next few years.
There’s a hint in the Samsung article that a fairly quick-and-dirty way to get flash faster is to increase the page size of the memory cells. If so, expect flash cell sizes to get a lot bigger in the next few years. That will probably hurt the seek times a bit, but yield a strong net plus for overall operations.
However, none of this is going to provide anything like a “Wow” performance boost for general operations. For 2007, what you’re going to get for your money is a solid state hard drive that performs like a hard drive, not anything a lot better than a hard drive. That will improve, but anything that starts looking more than DRAM than a hard drive for speed will probably not come until one of the wannabe successors to flash makes an impact four or five years from now.
If you see an example or two of wowish technology (i.e., boot times), rest assured it will be due more to software tricks than raw firepower. It will probably also help a bit to have what will essentially be a big (if slowish) cache to free up the hard drive to look for other things, but again, nothing to write home about.
Something else to keep in mind is that CPUs really aren’t capable of processing billions and billions of bytes per second. As we saw with the iRAM, it’s not like the only thing stopping your computer from booting in a split second is a data-starved CPU. Yes, data transfer is a bottleneck, and a big one, but it’s not all bottleneck.
One of the ironies on this whole issue is that solid state storage will eventually force CPUs to change.
This is going to be one of those technologies that will creep up on us. It will get a foothold for nonperformance reasons, and eventually will speed up.
One thing for sure, though, if you have a limited, or even not-so-limited budget, this is going to be one of the lower-priority items for the next few years: not a whole lot of bang for a lot of buck for pretty much the rest of this decade.