The Marketing of Performance and Vice-Versa
ATA/100, sometimes called UDMA Mode 5, is the big selling ‘feature’ these days when it comes to IDE hard disks and mainboards. The question is: Does it really offer any ‘realworld’ performance gain over ATA/66 (or even ATA/33 for that matter)? I’ve stated it publicly many times – No it does not, plain and simple. I get questioned all the time as to why I state that, many times being told that I am wrong (and sometimes not too politely). I’ve based my opinion on many hours of testing, using a number of different mainboards, chipsets and hard disks, so I figured it was time to do some actual comparison testing and publish the results.
First, you need to understand how a hard disk works when it comes to accessing the data from the disk itself, and then and transferring it to the mainboard. Rather then re-invent the wheel (and plagiarize someone else’s hard work) I’ll refer you to a couple of sites that have already provided much of the background information:
The main thing to remember is that the ATA/100 (100MB/s) speed ONLY applies to buffer to host transfers. That is, when the operating system can’t find the data needed in its own cache, and it is not in the controller’s cache, but doesn’t need to be retrieved from the actual disk itself (the platters). A small amount of data is cached in the disk’s buffer, and that is the ONLY data that will be transferred at ATA/100 speed. Any data retrieved from the actual platters cannot be done so at that high rate of speed – period. The chances that the system will be able to find what it needs in the disk buffer (unless your request are very small and fits the cache scheme) are slim to none, mostly due to its small size. So, your limiting factor is the actual data transfer from the platters to the buffer – which is well under the ATA/100 speed, under ATA/66 speeds, and not much better than ATA/33.
What actually has the biggest influence on how fast your system gets data (other than the operating system cache) from the hard disk? The sustained data transfer rate, which is how fast it can continually transfer data, or how fast it can find and transfer data from the hard disk itself (not the buffer) for use by your program. ATA/100 means that the hard disk can transfer data to and from the disk data buffer at 100MB/s. The only product specification I can find that quotes sustained data transfer is the IBM GXP at 37MB/s, well under both ATA/100 and ATA/66, and just above ATA/33 (check the benchmarks below and you can verify this). The WD 200BB I tested lists only a maximum transfer from the media (platter) at 50MB/s and a minimum of 30.5MB/s, the Maxtor I also tested states 49.5MB/s from the media… still well under ATA/66
What does it take to get a higher sustained transfer? It requires faster rotational speed and/or greater areal density. This is the main reason why today’s hard disks have better performance: higher rotational speed and packing of the data tighter on fewer and smaller platters. Today’s 7200-RPM ATA/100 hard disk show a very marginal loss when set to ATA/33 and perform pretty much the same if set to ATA/66 mode due to the sustained data transfer rate being roughly the same, or at best, slightly higher. This is also why a 7200-RPM hard disk is quite a bit faster than a 5400-RPM unit regardless of its ATA mode. Denser packing of the data on the platters (greater areal density) is also why some newer 5400-RPM ATA/100 hard disk can perform close to that of an older 7200 RPM ATA/66 hard disk.
So, if you want more speed using an ATA/100 hard disk, you need to wait for a 10,000-RPM (or better) IDE hard disk to see any real gain over ATA/66. Smaller & fewer platters will also help. One other thing I’ve not touched on is the effect of the operating systems disk cache. Within reason, the more memory you have, the larger the disk cache, and therefore the better the disk performance. This is one reason you see better disk benchmarks with more system memory installed.
Be the first to discuss this article!