Bryan Crow -> RE: SV-1A (11/10/2007 1:58:38 PM)
THE REST OF THE STORY, FWIF:
I found this thread while looking for ABIT information and recognized it as my own. In case someone else happens onto it, here's how things turned out. (You may want to skip most of this, especially if you aren’t interested in speech recognition. I’ve laid it out with the main stuff first and paragraphed so you can pick and choose.)
I bought the ABIT SV-1A in March 2005, Rumor is, only five were produced. In August 2005, my honest genius dealer finally got through to a U.S. ABIT tech by phone, and we managed to reconfigure the BIOS for the Adaptec controller.
It has performed faultlessly ever since, and I rarely turn the machine off.
I use it as an auxiliary now, because early the next year, I put together a much more capable machine for Dragon NaturallySpeaking. It's a whiz for AutoCAD and Photoshop too—and other applications that require quick I/O performance. For me, fast read-write times are generally more apparent to the user than faster processor speeds. Everybody who’s tried my rig shakes his head in awe. Even clicking on files wows them. It’s like switching a lamp. Pow! Most things pop up right now.
I based this contraption on a 2006 ASUS SLI mobo and, eventually, a water-cooled FX-62. What makes it much faster than any Core2 Duo(gotta have one) machine I've seen is its storage. Its Raptors are the slowpokes, used only for bulky long-term storage. The 16GB PQI flash memory, hottest thing available at the time, is for frequently-used stuff. But the crowning glory—and the key to its blazing performance—is the boot drive, a Gigabyte iRAM with four 1GB sticks of DDR.
Yes, only 4GB, not much more than needed by 64-bit Windows XP Profesional—with its paging files on one of the Raptors—but for all practical purposes, it’s really a 12GB boot iRAM.
How's that? I use Windows Disk Management to mount volumes from two other iRAMs to Program and Windows folders on the boot drive. The boot drive acts as if the other eight gigs are contained within it. It surprised me to find how many Windows files work just fine tucked inside mounted volumes. Of course you have to rename them by the same names as the Windows files pasted inside.
Why not just create a RAID 0 array? Gigabyte’s RAID software for iRAMs supports 32-bit operating systems only. At least, it did the last time I checked. And for a long time, there were no RAID controller cards that would do the job. There’s at least one now, but it would require a fourth PCI slot.*
A Tyan Tiger, for one, would provide four on a Socket 939 motherboard. Besides, even the fastest controllers slow down the iRAMs—with their response times measured in billionths rather than thousandths of a second. Mounted volumes entail communication between iRAMs. But it’s direct, not split into RAID packets and reassembled from them, processes that take extra time.
I used to mount separate volumes for 50 folders, but you can place lots of programs in a single folder for 32-bit programs and do the same for your 64-bit programs and user files under Documents and Settings. DNS is still a 32-bit program, but it works visibly faster on 64-bit Windows.
I haven’t tried Vista yet. One of its attractions is its native speech-recognition utility, available in the 64-bit version as well as 32. Early takes were pessimistic. Not up to DNS, I keep hearing and reading. But a buddy who was using DNS on a machine patterned on my iRAM rig has switched, Here’s what he says:
“For me, the Vista speech program is easier to use and at least as accurate.”
I’ll believe it when I see it, but I have no brand loyalties. I’ll give it lengthy and earnest try.
You always want more performance, and this rig will soon be a record two years old without so much as a successor on my drawing board. It does what I need it to do, and despite the addictive lure of building computers for their own sakes, I prefer to spend my time working with computers not on them.
So why am I searching the Web for technical information this morning? Just a spot of trouble getting the older rig to recognize the boot RAID 0 array again after adding a Raptor.
Back to the iRAM rig I call “Dragon” (for DNS): Its only drawback is that when I let my office-study get too warm, the iRAMs get too hot and drop out—or maybe it's the motherboard. Usually, they go back to work once they cool, but not always. So I'm also putting some extra touches on its Koolance water-cooling system.
By the way, I finally found a way to back up the boot iRAM and clone the other two. Acronis Enterprise Server for Windows is the only backup system I’ve found that’ll not only support a 64-bit OS but work with dynamic volumes such as those mounted onto the boot drive Windows installation.
Still, you don’t want to restore them every day. The Koolance system for the CPU had enough capacity for six more cooling blocks—two HHD blocks, two system RAM blocks, and two motherboard blocks. The hard drive blocks and one of the motherboard blocks actually cool the three iRAMs. RAM doesn’t generate a lot of heat? Sixteen gigs of it do. They generate a hell of a lot more than the Raptors—or the SCSI drives in the FX55 machine, for that matter.
Or the video card. I’m not a gamer, so I don’t take advantage of the SLI slots.
Heat from 16 DIMMs spreads to the motherboard. Some shutdowns may arise from running it too hot.
Now, the whole package runs cool with only two tiny fans plus those for the separate Koolance EXOS system—which I keep on the other side of my desk—yielding a stable computer with no significant flat spots.
But why iRAMs? Why not non-dynamic memory? Why not flash memory like my PQI? You can now buy faster and much larger flash-memory HDDs than my early-model PQI, but none of them touch the performance of dynamic memory (RAM). The fastest are hardly faster than the fastest Raptor and not a bit faster than my Fujitsu SCSI drives in RAID 0 array. The iRAMs are at least ten times faster, and the difference is readily apparent when you dictate to DNS. When I tried installing Windows on the PQI solid-state drive, it behaved about like the SCSI machine. The lag between my dictated utterances and the display of recognitions (text) made editing on the fly forbiddingly tedious.
With the iRAMs, your words appear less than a half-second after the last millisecond of each utterance. Some strings of words turn into text even faster. But the greater contribution to productivity is how quickly you can make corrections and teach DNS new words and names. Its lexicon comes with far more words than, say, a Webster’s unabridged, including names, brand names, technical terms, and slang. But until the first time I pronounced his name and spelled it aloud, it didn’t know how to spell Pojanowski, a friend’s last name. It has ever since.
*Funny thing about PCI slots: The SV-1A was intended for a market that didn’t pan out for Gigabyte, a market for small home servers for fast processors. This is the only Socket 939 motherboard my friends and I have found with true PCI-X slots. I’m not talking about slots for graphics. I’m talking about true 64-bit PCI. And the reason I had to have at least one is the Adaptec RAID controller for the SCSI drives. The 2230SLP is a 64-bit controller with dual channels to take advantage of SCSI drives in RAID arrays—plus RAID-on-chip to avoid robbing cycles from the CPU. Maybe I’ll spring for a monster server mobo that’ll accommodate the SCSI controller as well as a RAID controller. Plus a fourth iRAM. Not to mention an even faster processor. I always have to have the reigning champ—Intel or AMD. Meanwhile, I just discovered that the 2230SLP may well work in an ordinary 32-bit PCI slot. If this proves true, it may not as fast as in a 64-bit slot, but I wonder if you could see the difference.