Burningissues chose this dual-CPU motherboard for review because we wanted to see if a cheap dual-CPU board could be the basis of an optimum
super-stable burning platform. We assume most potential users of this motherboard will know what it looks like & its advertised specification: anyone wanting MSI's picture & specification please go here.
Burningissues has received an unexpected amount of feedback & information after our report on the MSI 694D-ProA dual-CPU motherboard: The result's building into a growing series of linked tweaks & mods to get the best out of this impressive product. At present we cover: [hardcore]
vcore stability tweak; Northbridge cooler mod; BIOS-flashing guide; RAID-mod Stage 1 - converts the models A/AI to "RAID -lite" AR/AIR spec & RAID-mod Stage 2 -goes beyond anything store-bought & unlocks the full functionality & performance of the onboard Promise PDC20265
controller to give RAID 0/1/0+1, full 4-HD support, & variable stripe-size. We also illustrate the only secure yet reversible "RAID-mod" resistor-method so far - it's just like putting a plug in [er - well - that's exactly what it is].
Please accept our apologies right now: this is at present an "interim report" - based on 6 week's testing & use with our
board, plus reports from a number of other users. Why no nice tables & graphs? - well; firstly the more we looked into it, the more the usual benchmarks looked irrelevant to CD-burning-enthusiasts - [we're working on a tolerably meaningful suite for burningissues tests]; & secondly, we
don't feel this board is ready for a full review to burningissues' standards.
The market for this board - the first-released dual using the VIA Apollo Pro 133a chipset - is wide enough that you can see why MSI were eager to rush the thing to market. Potential buyers include: those
"upgrading" from an Abit BP6 [the 694D-Pro is unique in accepting Celeron1 PPGA's as well as FC-PGA PIII's]; those looking to try a relatively cheap duallie for the first time; those working at a graphics workstation; & gamers & overclockers looking to have some fun. There's
also the odd burning-enthusiast out there looking for a platform that could, in theory, never be blamed for the latest strata in the coaster-stack.
The mixed nature of this market - & the inbuilt qualities & limits of this VIA chipset - have made for an unusual set of design compromises by MSI. The board [confusingly] comes in 5 flavours, 3 of which
are at present available, though not in all markets - the US doesn't get Firewire boards:
694D-Pro: 686A Southbridge: ATA66 x 2, onboard VIA sound & dual USB.
694D-Pro A: as above + Promise PDC20265 ATA100 controller with 2 more IDE sockets.
694D-Pro AI: as model A + onboard two-chip firewire [impedes at least one PCI-slot]
694D-Pro AR: as model A but with a 20267 controller giving IDE RAID [not released]
694D-Pro AIR: as model AI with RAID [not released]
As you see, most models have quite a few fair-quality onboard goodies - the Promise 20265 [Ultra100] & 20267 [FasTrak100] can be, according
to iXBT, the best ATA100 controllers available; however, all these goodies make bids - with the advantage of insider trading - for scarce IRQ's, which would always make things tricky come setup time even if the BIOS was helpful - which [up to v.1.62beta] - it ain't.
Jumpers, the BIOS, & FSB's:
MSI have provided this board with a couple of blocks of jumpers for setting the basic CPU-type [Celeron PPGA or FC-PGA PIII] & for setting the actual FSB ranges: 66-79; 100-114; 120-200. The advantage of
this is that any CPU may be forced to try to run in a particular range; the disadvantages - when you get to the soft-BIOS - are that very, very few FSB-steps are available to users of 66 & 100 FSB processors - unless they can first POST at 133MHz.
Fortunately, due to the outstanding quality of Intel's cB0-stepping FC-PGA's, most "E" models of 7 or less multiplier can at least be made to POST at a 33% overbus.
Particularly annoying is the lack of the advertised 83/90/95 FSB speeds from either the 66 or 100 jumpered bus - this makes the board a plain bad choice for those with 366, 400, & 433 MHz Celeron PPGA's.
Unless you are completely certain your PPGA's will run dual at 100 FSB, or are content with 75 or 79, this board is in no way an "upgrade" from an Abit BP6: that board not only has small-step FSB selection; but also allows alteration of the +3.3v setting - an important feature for
The soft-BIOS has the now-usual series of screens for setting basic then more advanced properties. The two screens for "PnP/PCI Configuration" & "Frequency/Voltage Control" are at first
sight temptingly complete: experience with them in action is frustrating.
The allocation of resources within the BIOS is exasperatingly rigid: after much experiment it appears, for instance, completely impossible to have two bootable PCI disk-controllers without at least one sharing
an IRQ with the AGP-slot [or the onboard ATA100, or - aaargh - each other]. The BIOS appears to be wholly reluctant to assign IRQ's 5 or 9 to any system-device at all. This sort of needless hassle can make setting-up your machine optimally a very, very long process.
To make it worse, the settings are unusually "sticky" & it is necessary to frequently jumper the CMOS clear before you can start afresh - of course then having to reset all variables down to the
system-time. In a fairly wide experience of modern motherboards this reviewer has never come across a soft-BIOS offering such poor user-control over the properties it is intended to adjust.
The "Frequency/Voltage Control" screen appears at first sight to offer individual control over each CPU's multiplier [what's the point of this with all Intel socket370 CPU's being multiplier-locked?]
& vcore. The FSB is set for both together.
The vcore range goes up to an astonishing +0.4v - this would take a cC0 stepping FC-PGA [1.70 vcore default] up to 2.1v, which is a very risky level strongly discouraged by Intel. It would be much more helpful
to have the + settings in 0.05v steps over a smaller overall range; since getting two CPU's to run together, especially when overbussed, is a delicate process.
FSB-speeds available within the ranges are few & widely spaced for those with the bus jumpered to 66 or 100: for the former, only 75 & 79; for the latter, only 110 & 114. For those with "EB"
FC-PGA's, or "E's" able to POST at 133, the offered range is much greater: 120 through on up in 12 wide-ish steps to a notional 200MHz.
Very interestingly, the board appears to offer a 1:5 FSB:PCI ratio at around & above 166MHz; however no known production board listed above has yet been reported as POSTing single or dual over 147FSB to see
whether this in fact works - & it looks like folk have tried hard on this motherboard with low-multiplier or unlocked FC-PGA's together with SDRAM known to run stably together at 166MHz FSB in i815 chipset motherboards.
The "independent" vcore adjustment isn't: even with the latest BIOS' an adjustment to the primary CPU - #0 - will often change the setting of CPU #1; in addition many users have reported that a warm
reboot will randomly reset one or both CPU's vcore to as high as 2.0v - that's very near the level at which you might risk permanent damage to a FC-PGA. As a final annoyance it is very difficult - if not impossible - to get both CPU's vcore-setting similar. All in all, these problems are so
severe - & have persisted through so many revisions of the BIOS - that they appear to have their root in quality, choice of components, or poor basic design.
Board layout & heatsinks [what heatsinks?]:
It's a big board, & will get worryingly near a deep CD-device even in large "midi" towers like the GlobalWin 802 we use; but with few PCI-slots  & no ISA's. There is a CNR [modem-only] riser
- if you want or can find a device to fit this.
Plus points - to this reviewer - are the tolerable-quality onboard sound, which can accept input from two CD-devices, & the possibility of attaching a third-party adaptor for two extra USB ports to a header
provided. There are four SDRAM slots, & the latest BIOS will apparently allow 4 x 256Mb sticks to be fitted - though some confusion persists as to whether these may be PC133 or PC100 only.
Negatives include the very cramped space round the socket370's: you can only fit a limited range of fansinks, & even the compact GOrb is tough to fit - especially to socket
#1, where the ATX connector interferes with getting this fansink seated firm & flat on the tiny bearing-surface of a FC-PGA. For those needing high-performance cooling, Alpha fansinks fit; for those requiring ultra-performance fansinks, we shall be testing to see if any fit & can work
without interference from each other.
Each socket 370 is thoughtfully fitted with a heat-sensor on a thin arm which may be carefully bent to press the sensor up against the base of the CPU: sadly, these are laughably inaccurate - anything up to 25%
out [cool, as usual with these falsely-comforting things] measured against our temperature-monitoring setup.
No heatsink was fitted to the Northbridge of the VIA chipset: I measured this running at 19.5°C [36°F] over case-ambient with just one 100FSB CPU under no load - ie considerably hotter than the CPU: a reliable
source has reported temperatures in the mid-50's °C under load with dual FC-PGA's @ 933. These levels are far, far too hot & the review-board was hurriedly fitted with an old 486 fansink. Current production boards are fitted with a small heatsink - we strongly recommend changing this for a
proper fansink/chipset-cooler. The Northbridge of a dual motherboard works much harder than on a single, & not providing a 10c heatsink on a board theoretically overbussable to 200MHz was a shoddy piece of "accountant's design"
Memory - not as in: "thanks for the . . .":
Like many VIA 133a-based boards, the default memory-performance of the 694D-Pro is poor compared with that of a BX-board. This motherboard has also, it appears, an unusual sensitivity to memory specification
[not necessarily the same as quality], with many problems from installation onwards having their probable cause either in generic memory unable to cope with the rather abrupt timings of this chipset, or in conflicts between the settings derived from the SDRAM's SPD [an EPROM on each stick with
information about the memory's specification - crucially its timings] & attempted user-settings.
A feature of the VIA 133a chipset is its ability to enable the SDRAM clock to be set to +33 at 66FSB, + or -33 at 100, or -33 at 133. This theoretically allows old memory to be migrated to this board; however,
since many SPD's are inaccurate [or perhaps conflict between plural sticks] this can lead to an infuriating variety of faults traceable to memory which has worked perfectly well at similar speeds in other motherboards less pedantic BIOS'.
There are settings to enable or disable SPD detect, both overall & specifically at some FSB settings; however the overall disable appears not to work reliably or thoroughly in any BIOS-version & SPD is
not available for all FSB's. At present it is advisable to use "good" SDRAM - with "good" = having a strictly truthful SPD which will satisfy the BIOS' rigid needs.
This "cheap" dual-board thus realistically requires the best brand-name CAS 2-2-2 PC100 or PC133 SDRAM you can afford; it is also best to fit two or more identical sticks rather than one big one - this
will allow 4-way interleave, which can noticeably improve memory performance to near-BX levels.
The review here has some sensible suggestions for optimizing memory from inside W2K using H.Oda's clever WPCRSET utility, derived from the more
general series of pieces also at viahardware.com
Soyo managed to put hardwired SDRAM 4-way interleave into the BIOS of their good-quality 7VCA motherboard several months ago, soon followed by other manufacturers using the VIA 133a: MSI's inability to enable
this important feature by default through multiple BIOS revisions is a worrying sign of either technical incompetence or disinterest in fully supporting this motherboard.
Lack of this feature is on its own enough to make this motherboard a poor choice for gamers.
Setup & OS install:
BIOS' earlier than 1.4 are unusably flaky; we recommend v1.50, 1.53beta, or 1.62beta -
though bear in mind a beta BIOS is out there for testing. It is recommended to flash from a boot-floppy with only minimal system-files, plus the flash-file & Award-flash utility [& the specific: "biosup.bat", if provided]. It can also save anguish to fit CPU #0 only &/or to
run the CPU[s] & your SDRAM at default speed when doing this very delicate operation.
MSI are busily adapting their terrifying "Live-BIOS" utility for this board: this enables downloads & immediate flashes whilst in a "Windows" OS' inclusive of W9x - anyone who would
voluntarily do this without a hardware BIOS-backup likely wears a pastel-coloured badge reading: "I believe in fairies."
After getting the BIOS & resources to a tolerable compromise - we achieved no better with a lightly-loaded system with only one AGP-card, two PCI's, one onboard serial port, the LPT, USB, & onboard sound
enabled [both IDE channels + second serial disabled] - you'll want to install an OS capable of exploiting the advantages of dual SMP-enabled CPU's. We'll assume you intend to install Windows2000. It is recommended to enable ACPI [in the Power Management setup BIOS-screen] if you are installing
W2K doesn't come with drivers for the PDC20265/20267 ATA100 HD-controller - each are different - & the drivers on our install CD were wrong: go to
Promise to d/l a working driver-set & expand it to a floppy. If you have a 4x AGP card you might also want to have a later "4in1" VIA driver-set than the [stable] 4.23 on our CD; also d/l the latest stable version of "usb.inf" & the onboard sound drivers if you
intend to use either. The very latest VIA drivers are in our experience more likely to be available here than at VIA or MSI.
You can choose either to install W2K with your boot HD connected to the Promise controller's primary channel, or with it temporarily connected to the primary IDE channel controlled by the VIA 686A Southbridge,
swapping the cable over once W2K is installed together with its correct Promise driver. The second process is quicker & simpler, avoiding the frantic stab at F6 in the few seconds W2K install allows for notification if you need to install a third-party HD-controller-driver from a floppy.
Any time you intend to use the Promise controller for your boot-HD, you must set the BIOS' HD-boot order to SCSI first, then "ATA/SCSI" to ATA first if you should have a bootable drive attached to a
If you take the trouble to set up your settings & resources adequately on a recent BIOS the install should go smoothly. Some folk have had nightmare problems & there is much talk in the relevant forums
of returning many boards as defective. There's no special reason to believe MSI's quality-control to be worse than normal; but this motherboard appears to be abnormally sensitive to memory-quality, BIOS-setting-hassles, & voltage-control problems - all eager to first show themselves during an
install. A few folk have also traced random install problems to one of the fansinks [usually #1, due to the space-problems described above] being poorly seated.
We have seen above that the board just plain ain't a sensible "upgrade" for those using PPGA Celeron CPU's in an Abit BP6, & a mediocre choice for gamers or overclockers due to its poor default
memory performance, lack of FSB options, & wretched vcore instability.
Of the two remaining possible markets we mentioned at the beginning of this report - those trying a duallie for the first time & graphics professionals looking for a cheap workstation - it is the latter,
oddly enough, who are best served.
Many rich folk have stuck in a pair of 700e processors & two or more sticks of expensive SDRAM & found the board will work stably enough at 933 MHz. Often, professionals have a fairly straightforward
setup, running their OS off a 10k LVD SCSI HD, & keeping their work/data on nice big cheap ATA100 dustbins. When a working tool, it is worth the investment of time & hassle to get this motherboard set up correctly & this sort of configuration represents good value-for-money in this
For a first-time duallie . . . . hmmm: depends on how much skill & patience you have - indeed, on how particular you are about a machine working properly. You will also be much more likely to have a
hassle-free experience with this motherboard if using brand-new top-quality components - especially memory. In any event, the resource-sharing hassles [common to many modern motherboards] are so extreme here & with such an unco-operative BIOS that this is a very tricky board to set up
properly when loaded with the normal array of resource-hungry cards in an all-purpose home machine.
Several reviews have called this board stable; perhaps it is more exact to say that problems may be disguised by the inherently greater stability of a dual-CPU machine running a SMP-aware OS. Comparisons with
similar dual motherboards using the same VIA chipset will allow us to see how stable the MSI 694D-ProA really is: both Epox & Abit [to mention two well-respected makers] have announced similar products. We intend to compare each to the 694D-ProA.
"Optimum super-stable burning platform"? - a personal view:
Pursuing stability at all costs; I have been satisfied for some time with all-SCSI HD & CD-device setups; running off two bootable single-channel PCI hosts [UW & LVD]. The inability of this board to
support two such bootable hosts [or my inability to find an answer over 6 weeks of testing] without IRQ sharing makes the ideal of multi-tasking-while-burning just a little less than carefree.
The sheer complexity of this motherboard is perhaps one reason why common solid utilities like the v-com Boot/Partitioning-Commanders & Spinrite5 will not run at all - though doing so happily on BX-boards
with the identical all-SCSI setup.
All in all I think this board was rushed to market, then inadequately supported; above all in MSI's failure to provide a trustworthy BIOS. It is an interesting challenge to the confirmed tweaker, a novel toy for
the fairly rich, & possibly a cost-effective tool for the graphics professional used to the truly extortionate all-up cost of many previous dual systems. Otherwise . . . well, I guess it's up to MSI to show us they can offer wholehearted support for a product best described as half-baked.
At present this motherboard cannot be recommended unless you're able to put in all-new best-quality components & don't demand the thing works perfectly. If it were s/w you'd call it beta; but it has been an
interesting exercise seeing whether a very large manufacturer can make & support a reasonable-quality affordable motherboard of this kind - with the courteous exception of tech-support at MSI's Australian subsidiary, I have received no reply to any Email to various parts of the MSI
corporation - this seems a common experience.
In the abstract, dual-CPU FC-PGA machines are highly desirable; & in W2K & Linux have two mature OS' to exploit their qualities. In practice I would recommend waiting a while till a reliable reviewer
pronounces the trick's been done with this or another similar motherboard - the tactful review at 2cpu.com, read carefully & between the lines, could have saved me six weeks' hassle.
We will complete this review by providing burning-oriented performance-figures when satisfied the board is, or can be, properly set up to run two CPU's stably at similar vcores & to use our normal HD &
CD-device-controllers without IRQ-sharing.
Burningissues has offered MSI the opportunity to respond to a draft of this report before publication: it appears not untypical of their support of this product that we have received no reply. The many owners of
this product requiring support contribute largely to the lively & helpful forums at 2cpu.com; & maintain a busy mailing-list.
> 694D-ProA Tweaks 1
> 694D-ProA Tweaks 2
>>> 694D-ProA Tweaks 3
694D-ProA Tweaks 4 >>>