I got this email, and I’m sure quite a few more are thinking along the lines of what this person said. A lot of this
was certainly along the
lines of what I had been thinking before looking into this more.
Just to make this easy to follow, the email comments are in bold italics; my comments are
in plain print.
Cenatek (and probably others too) is also selling software for running a
ramdrive from system memory. In my opinion, that’s a much better solution then
the PCI-card. Here is why:
1. It’s bootable. It can be configured to copy your boot-partition and boot
from it. Boot-time will get longer, but everything else will rock.
Unfortunately, this is the killer problem with a big ramdrive. He is correct, you can
configure it that way, and people have gotten it to work.
The problem is before the thing can
even start booting, you have to copy everything in the partition from the hard drive into the RAMdrive before the “real” boot,
and that will take much longer than just booting from the hard drive to begin with. This is especially true for Windows XP
Imagine making a Ghost image of a small partition of 2-4 Gb, and you’ve got a good idea of how long this would take each time you reboot. It will add a few minutes to the boot time.
Now if you leave your machine on 24/7 and rarely reboot, this might not be too much of a problem, but
for the average person, this is going to seem like rushing backwards.
2. You get 3.2GByte/s (with DDR400) instead of PCI’s 133MByte/s
That doesn’t seem to be the case at the moment; if you look at this forum thread, a person who did this and ran the ATTO benchmark on the “drive” got “only” 300MB/sec from it.
That’s much faster than a hard drive, but hardly as fast as raw RAM speed would lead you to expect.
3. You get real nanosecond seeks instead of adding the PCI and controller
latency of closer to microseconds.
You still have to go through the memory controller, and as the ATTO benchmark indicates, there’s a lot of overhead someplace.
4. Cost of the PCI card is astronomical. Just maxing out your motherboard’s
memory capability is much cheaper.
Well, when 4Gb of DDR costs over a thousand dollars, it’s pretty hard to call $800 for the card “astronomical.” The cost of this whole project is already in orbit.
5. The software can compress the “partition” to get even higher real
(uncompressed) bandwith and more important: use the memory cpasity more
Compression takes up CPU cycles, it remains to be seen whether that would product a net benefit
or loss in this situation. Just don’t know right now.
6. The software saves the ramdisk to the harddrive before shutdown. That way no
data is lost, and the OS is updated.
This become another long haul; overwriting a multi-gigabyte disk image each time you shut down.
In case of a crash, the OS will start
whare it started last time. Data could also be saved to harddisk in intervals
(IE. every hour)
Please note, start where it started last time, not where you left off. A crash means losing all unsaved (to the hard drive, NOT the ramdrive)
Right now, Windows 98 with some software would go fine on a 1.5GB ramdrive.
The problem is WindowsXP is now currently being used by half our audience (who presumably would be more the type to try this than the average computer user), and Windows 98 usage is below 20% (and shrinking fast). It’s hard to imagine someone willing to spring for a ton of RAM and not an OS.
Realistically, somebody willing to do this is going to use XP and is going to need 2-4 Gb of memory for it.
(The price of 1Gb modules will go down, and dual-channel mobos will tend to increase
the number of available RAM slots}
So ramdrives are absolutely the future, maybe sooner rather then later.
Well, trial of concept with this approach is probably sooner rather than later.
If I Were To Do This . . .
I think those who are intrigued by this idea expect absolute miracles from this. Why not? I did.
What this looks to be instead is rather a hell of a lot of an improvement for a hell of a lot of money, with some very big impracticalities for the average or even not-so-average computer user.
All that being said, just because I’m pretty much telling you not to try it doesn’t mean I won’t.
Right now, there are two places which offer this type software: Cenatek and Superspeed. Centatek doesn’t have an XP-compatible program yet (it’s in beta); Superspeed does, but is far too light on documentation.
Hopefully, that will clear itself up one way or the other.
Hardware presents a greater difficulty. Since this will obviously have to go into a modern system, that means DDR. Since I would kind of like this to not go obsolete in a month, we’re looking at DDR400. To realistically be able to operate the vast majority of the time purely in memory for what I use, we’re talking 4Gb of total system memory. Being realistic, the most RAM slots I can expect to get in any high-performance mobo is four, and the biggest modules I can reasonably afford to buy are 1Gb modules.
So I need 1Gb PC3200 sticks, and they don’t exist yet. 1Gb PC2700 sticks exist, though, and they cost about $300 a pop.
It’s probably reasonable to expect to see 1Gb PC3200 sticks available by the summer. What may not be reasonable is expecting the likely mobo, a Canterwood, to support 4Gb in dual-channel mode at 200MHz (nForce2s won’t support more than two sticks in dual-channel, and only a couple support four memory slots).
What’s probably the most unreasonable is to expect I persuading me to actually lay out $1200 first, and find out the benefits later.
Maybe we can arrange a try-before-you-buy review, and if it works out pretty well, then I can decide whether to get the rust out or not.