Atomic I/O letters column #82Originally published in Atomic: Maximum Power Computing Last modified 16-Jan-2015.
I purchased a Skymaster Serial ATA 150 Sil3112 Controller Card, model N791, and need help from the manufacturer of the card. Would you possibly have a contact e-mail address for this manufacturer? Any help would be fantastic, as I cannot find them.
I am running a computer using an Asus motherboard running the onboard Nvidia RAID 0, and when I install this card as an add-on card with a single drive attached, at POST I get the following error:
Not enough space to copy PCI Option ROM [00:08:00]
and my original onboard RAID disappears and the computer tries to boot from the add-on card. In the BIOS setup program, the onboard Nvidia boot option actually disappears!
On the card there is a place where a jumper could go to change the add-on card to "IDE Mode" (as opposed to RAID mode), but there are no pins to put a jumper on there. I wanted to ask Skymaster if I could modify the card by closing this particular jumper myself.
It may sound strange that I would want to modify the card, but unfortunately computer components are not readily available here in Zimbabwe. I think the error means both RAID Controllers can't occupy the same boot ROM space.
First up: You certainly can connect the jumper pads on the board by soldering any old little bit of wire on there, as long as all it touches are the jumper pads. That may perhaps solve your problem; if it doesn't, it should be easy enough to remove the jumper wire again.
I doubt this'll actually change the option ROM configuration, though, so you probably are going to have to get a new controller card.
Fortunately, the going rate for SATA cards on eBay is now little more than $AU10 including delivery for something based on a Promise or Highpoint chip. So there must be some way for you to lay your hands on one, even in Zimbabwe.
Forget trying to contact "Skymaster", though. Skymaster is just the name that some Australian distributor of cheap PC gear decided to slap on their products. There is no "Skymaster factory", and if you can find someone who supports Skymaster gear in particular then you're a better man than I.
I don't even know exactly what device you've got there. The Silicon Image 3112 is a well enough known early SATA controller chip (it dates back to 2002), but the only "Skymaster N791" I can find is a DSL router.
Fortunately, this doesn't actually matter. Because Skymaster is not a real company, there are no mysterious problems that're particular to Skymaster gear. If you buy hardware from some obscure little manufacturer who make their own weird firmware and drivers, you may have problems that nobody else has. That is not the case here, though; the problem you're having is fairly well-known.
When a PC starts up, it loads the "option ROMs" for its basic hardware into memory. Those ROMs allow the system BIOS to deal with the basic hardware that's needed to start the computer - things like elementary video, keyboard input, and hard disk controllers.
When this rather prehistoric process happens, each necessary ROM is loaded into its own little spot in the first 640 kilobytes of memory. I've written about this before in this article, which is mainly about the much larger Memory-Mapped I/O "holes" for modern graphics cards.
There's some flexibility to the ROM-loading process, so different pieces of hardware don't often step on each other. Each vehicle trying to find a place for itself in the low-memory parking lot has a range of parking spaces it can accept.
But it's still possible for two or more pieces of hardware to end up wanting the same parking space, especially if something in the system has dumb firmware that insists on reserving most of the parking lot for itself. Then, if the clashing pieces of hardware are hard disk controllers, you'll get the kind of problem you're having here.
"Not enough space to copy PCI Option ROM" is one such error message. "Insufficient Memory to Shadow PCI ROM" is what an Intel-chipset board will say in the same situation.
Unfortunately, I don't think there's actually any way to make these two disk controllers play nicely together. Perhaps bridging the jumper will make a difference, but I doubt it.
But there's nothing special about the old Silicon Image 3112 - it's just a basic two-port SATA controller with the usual RAID 0 and 1 options that're mainly handled by the driver software. Since you're not even using the RAID functions, any SATA controller card at all will solve your problem.
You could also attach the extra drive with a USB-to-SATA box or adapter cable, but cheap though those are, SATA cards are even cheaper.
I'm curious to know if there's a rule-of-thumb that I can use for determining when I should be changing the thermal grease (or something) on various heat sinks. The very nature of the stuff means that if I was to remove a heat sink to check whether it had dried out, I'd end up having to apply some more anyway. What's a reasonable interval?
If the goop was applied properly in the first place (a thin smear, not a big iced-doughnut blob), and if it hasn't degraded into uselessness, you don't need to replace it at all.
It's hard, as you say, to tell if thermal compound has degraded without taking the heat sink off. If nothing's actually overheating then you might as well let sleeping dogs lie, but if you simply have to look, and then all you find under the heat sink is a dusting of powder, you know that you need to put fresh goop on before you put the heat sink back on again.
But you don't know whether the dried-out compound wasn't still working perfectly well before you ruined everything by peeking.
All thermal compound has to do is fill the voids between the heat sink and the chip package. Dry compound can do that as well as wet, if it's already filling the gaps and the heat sink isn't jiggling around. If a metal-loaded compound, for instance, has dried out completely, it may now work better than it did when wet.
Thermal compound datasheets may or may not quote service life for the compound. If they do, it may be for higher temperatures than the sub-boiling temperatures that're the worst the compound has to endure when you use it in a PC. Many thermal compound formulations will dry out after a few years even if the computer's never turned on, though.
So the numbers that're relevant to computer users are all pretty indistinct. One thing you can count on, though, is that letting dust into the junction area by taking the heat sink off is bad news. Dust - and, even worse, hairs - will hold the heat sink and chip apart, making a considerable difference to thermal transfer.
It's even more important to make sure you haven't shed a hair onto the top of the CPU than it is to make sure you use a small amount of goop.
I read your informative article comparing heat transfer compounds. Have you tried gold leaf, typically found in craft and art supply stores and used for gilding and book embossing? I'd be interested to know your results.
I suspect gold would be surprisingly ineffective.
Even the flattest of surfaces have many tiny air voids between them when they're pressed together. Thermal goop has to fill those voids with something that has better thermal conductivity than air, and also has to stay out of the way in the rest of the contact area.
Just about anything has thermal conductivity much better than air (glass beats air by a factor of 44...), so just about anything that'll stay in the gap without decaying or flowing out will get the job done.
Gold's thermal conductivity is excellent. Copper's conductivity is about 1.25 times that of gold and aluminium is not that much worse, at about 0.75 times, but gold won't have much (if any) oxide on it and is very malleable. So in practical situations, 24-karat gold slugs on the tops of processors and the bottoms of heat sinks would probably work quite well, if they were affordable.
I suspect that just putting a thin layer of gold in between an ordinary CPU and heat sink, though, would create two problems.
One: The gold wouldn't flow out of the contact points where, with no thermal interface material at all, there would otherwise be metal-to-metal contact.
Two: The gold wouldn't fill the voids very well. The gold in the metal-to-metal contact areas would hold the two surfaces apart, so the voids wouldn't end up any better filled than they were without any interface material.
If you squished the heat sink around a bit on top of the processor after latching it on then you might get the gold to move out of the contact patches and into the gaps, but I suspect there wouldn't actually be enough pressure to get this to work properly. You'd actually get the shredded ultra-fine gold leaf building up in lumps in some places and leaving voids in others, like a big sheet of tissue paper being mashed around between two mattresses.
The result would be an even worse contact than if you'd used no interface material at all.
Interestingly, at least one company sells a "liquid metal" thermal interface material, which is one of those "pseudo mercury" room-temperature-liquid alloys based on gallium and indium (plus, in this case, apparently also rhodium, silver, zinc and tin - though I bet there's actually very little of those in there).
It doesn't seem that even this liquid metal works any better than conventional thermal goop, though. This could have something to do with the fact that the thermal conductivities of gallium and indium are pretty lousy; aluminium has a conductivity of 237 W/(m*K); gallium is 40.6 and indium is 81.8.
And, just in case you were wondering, mercury's conductivity is terrible, at only 8.3W/(m*K). It's possible for liquid thermal compounds to do nifty thin-film and/or convection tricks to cool things better, but if a gallium/indium alloy doesn't manage it, I'll bet mercury wouldn't either.
[A reader's now reminded me that mercury will also eat aluminium, among other metals. I think the outer package for a lot of CPUs is aluminium.]
What is your expert opinion on "SATA cables should not be bent 90 degrees"?
This is a problem I have faced with my PC and a lot of my tech support customers.
It's ridiculous to say that any kind of electrical cable must not ever turn by 90 degrees. That'd preclude your hard drives from ever being anywhere other than standing out perpendicular at the end of the cable. I presume you'd have to cable-tie your drives to garden stakes or something.
Brand new cables come in packets in which they're looped back and forth several times, for heaven's sake.
All sorts of cables do, however, have bend restrictions. If you kink a cable too hard you can fracture conductors (or optical fibres).
The bend restrictions always specify a radius, though. The minimum bend radius for Category 5/5E network cable, for instance, is usually specified as at least eight times the outside diameter of the cable, and possibly rather more. If you violate that by just yanking the wire around the edge of a ceiling joist when you're running cable through your roof, you may end up with a flaky connection.
I think the minimum bend radius for normal SATA cables (obviously, only in the direction in which these flat cables want to bend...) is 25mm, but I wouldn't bet my life on that. It's quite easy to kink a SATA cable worse than that if you're threading it between chassis members in a PC case.