In order to have a look at the drive, we first pulled out the hard drive on the left (in a mindbendingly complex mess of screws and kapton space-tape that has to be seen to be appreciated. Apple’s official position on these machines is that doing this yourself voids your warranty, and I can appreciate their position. (We had an apple authorized tech center help us with these photos!) The top case must be removed, clocking in at ten small screws, the keyboard ribbon carefully disconnected (visible in the reassembly picture at the bottom), and then the hard drive carefully removed, the shock mounts (visible in the close-up of the OEM drive) swapped over, and the whole process reversed. It’s delicate, tricky work – don’t try it for yourself until your warranty’s already expired. Don’t pay more than $20 to have it done for you, either.
|Apple’s OEM Fujitsu hard drive…||…nestled into its little spot under the cable|
Once the harddrives were swapped, I fired things up and gave it a good day or two of usage to get the hang of how things felt. As expected, applications launched faster, load times in games were cut noticeably, and … well… it was basically exactly what I expected. But just how fast is fast? It certainly feels that way, but the placebo effect is strong, and well-documented. To find out just how much faster it was, I had first fired up my favorite Macintosh benchmark suite, XBench 1.3. I ran a comprehensive battery of “before” tests to establish a baseline, and then it was off to the labs to swap hardware.
|Apple OEM:||Seagate Momentus:|
First, let us examine sequential write events: with block sizes of both 4K and 256K, the Momentus offers double the write speed of the OEM disk. In read speed, the 4k blocks resulted in a modest improvement of about 50% (for very generous values of modest), but the larer blocks nearly tripled the read speed of the drive. In real world applications, games using large texture or other asset files will load faster – and for games with frequent load screens (Valve, I’m looking at you!) the pauses will go from immersion-breaking to … less immersion-breaking. Maybe just long enough to catch your breath and wipe the sweat from your palms, if you’re lucky. Random preformance offered more modest performance gains, respectively with 15%, 113%, 40%, and 37% gains. This more or less makes native command queueing and prefetch ineffective, as the drive controller will find it very difficult to guess what data will be needed next; the drive isn’t so much working smarter here as it is working harder. The relatively massive 113% improvement in uncached write speed can likely be attributed to writing data to the disk’s DRAM cache as fast as it could be done, (thus giving the command queueing software enough data to work out an efficient plan) and letting the write head move as little as possible – we’re back to working smarter, not (just) harder.
I know a lot of you out there live in Windowsland, so I also took a snapshot of the drive using HD Tach, which yielded higher burst rates and lower sustained rates than XBench, though it did not break the rates down by block size. To put it in perspective, I compared it to the Patriot Xporter XT, the fastest flash memory I own. The poor thing couldn’t even compete on sustained transfer rate (only 1/4) though like all things solid state, its random access time was miniscule. In contrast, the Momentus was not quite bus-limited on my SATA 1.5 system, but it was close. In comparison to SATA 1.5, the maximum theoretical transfer rate of USB2 is about 53 MB/sec, and in practice, any one device can only use about two-thirds of that… meaning that the Xporter was bus-limited too.
Why is this comparison to flash memory important? Unfortunately, real-world tests are hampered by the use of laptop drives in actual laptops – all of my other I/O is severely bus-limited, with even the biggest of tubes I have at my disposal (Firewire 800) resulting in bus-limited copy performance when attempting to push files around between external and internal drives. This no-faster-than-before result conflicts with both my impressions of the drive’s performance and benchmark results; I am therefore inclined to ignore it and move on.