Serial ATA with the Abit NF7-S

Serial ATA at last!

I've been excited about Serial ATA for a while now. Not, I should point out, because of the speed. As you are probably aware, every time the ATA drive spec ups the speed (ATA/33, ATA/66, ATA/100...) it's at least six months before drive manufacturers produce disks that can stress the new, faster bus. So the fact that Serial ATA has a top burst data rate of 150Mb/s is neither here nor there.

But there are much more significant reasons to lust after SATA. Hot swapping, for a start. Because I'm a geek, and I live with a few other geeks, about every other week there's some dead computer for us to fix and we spend a lot of time yanking hard drives out of machines, dropping them into one of our own computers, and backing them prior to our seventh Windows reinstall of the month. Well, if SATA wants to let me accomplish that without having to mess about shutting down my PC when I put the disk in, that is just fine by me.

But, in my opinion, SATA has a killer feature. And this is it:

What you have there is an eighty-wire ATA/133 cable next to the sleek form of a seven-wire SATA one. I used to run a PC with three optical drives (one SCSI, two IDE) and three IDE hard disks. That involved four big, fat, flat ribbon cables looped and coiled through a very small amount of physical space. A very small amount of space, I should add, that also contains a half-dozen power wires. And then I rather optimisitically pointed a fan at it and tried to cool my system through this tangled rat's nest of wires.

Not a great idea, I think you'll agree.

So when I bought a Abit NF7-S motherboard a week ago, I was excited. The board has two SATA headers on it and it Abit have thoughtfully thrown in a cute little adaptor to convert an existing parallel ATA drive into a serial one. They've even bundled a SATA cable, the cunning blighters. This is what it looks like connected to a drive:

Try to ignore the awfulness of the "Seriellel" brandname, though. It'll colour your perspective unfairly.

But I couldn't help wondering: this clearly isn't a real SATA solution, what with the adaptor and all. What if whatever translation the adaptor is doing slows the transfer rate down? I've just laid down a bundle of cash on one of those 8Mb cache Western Digital drives and the last thing I want to do is inadvertently lose a precious 5% of performance. Ahem.

The test

The pertinent bits of the machine I'm testing all this on are as follows:

• AMD Athlon XP1700+ @ 1834Mhz (2210+, fact fans)
• Abit NF7-S nForce 2 motherboard with integrated Scilicon Image SiL3112 SATA controller
• MSI GeForce 4 Ti4200
• Western Digital WD800JB hard drive

The nForce driver bundle version 2.00 was installed, as was the SATA controller's driver from the motherboard CD (as a quick websearch reveled no more recent version). The drivers being used were reported by Windows as follows:

The test consisted of multiple runs of Ziff-Davis WinBench 99's disk test subsection. Now, a disclaimer. I'm not storagereview.com, however much I wish I was, and I don't really know a lot about hard disk benchmarking. I do know enough about benchmarking to know that synthetic tests are not really all that trusthworthy, but that real-world tests are insanely hard to obtain decent repeatable results from. Hence, I have erred on the side of caution in this test and used a synthetic benchmark, reducing the chance that I can somehow break it.

On a clean install of Win2k, I installed the drivers listed above, then defragged the hard disk. I ran WinBench's disk tests three times for the SATA controller, rebooting between tests, then three times with the disk attached to the onboard parallel ATA controller.

Results

For the impatient:

But that's not all the story. There are odd things afoot here. Note that although the SATA controller apparantly whomps the parallel one at the Business Disk mark, it is neck-and-neck in the High End test.

Here, for the thorough amongst you, are the individual results, taken directly from WinMark99:

You'll notice the results are nicely repeatable, so there's no random spurious stuff here. They really do perform the same in one benchmark but differently in another.