What makes a hard drive fast? An interesting question, no doubt, and a topic beyond the boundaries of one simple article. Generally speaking, however, elements that affect hard disk performance, in order, are spindle speed, seek time, sustained transfer rate, and buffer/firmware algorithms. This last factor, unfortunately, is somewhat difficult to quantify. Cache size is the only uniform specification easily reported. Algorithms and programs are terribly abstract, however- they don't read well off of a spec sheet.

There is, however, no doubt that the nebulous drive electronics can significantly contribute to overall disk performance. Witness, for example, the Quantum Atlas 10k vs the Fujitsu MAG3182LP. Though the Fujitsu boasts faster access times, speedier transfer rates, and the same buffer size, it's handily trounced by the Atlas 10k.

What about buffer size on its own, however? Recently two companies, Western Digital and IBM, released drives that trumped the competition by raising buffers from 512k to a massive 2048k, unheard of in the ATA landscape. I've read many newsgroup articles as well as posts in our own Discussion forum citing the 2 megabyte buffer as one of the main reasons a user purchased an IBM/WD.

Until recently, it's been difficult to ascertain the difference that a roomier buffer makes. "A/V optimized" drives often included larger buffers, but they were but one of several improvements made in such models. Maxtor's 4320 series included an interesting dichotomy: 512k of cache on larger models, 256k on smaller units. But differences in platter count can also obscure test results.

The release of the DiamondMax Plus 5120, however, finally gives us a chance to test the effect of buffer size in a controlled environment. Though the original OEM model was to feature 512k, retail-boxed models feature 1024k. Why? Maxtor itself has been quite subtle in pointing out the distinction between the two units. A likely hypothesis would be the retail boxes of the WD Expert that would be sitting next to those red DiamondMax Plus packages, trumpeting a massive 2 meg buffer. Marketing-wise, perhaps the 5120 does much better in sales with the narrowed gap in buffer size.

Savvy StorageReview.com readers, however, want to know the bottom line: Does the 1 meg 5120 perform better than the 512k version? And what ramifications do the results present when it comes to buffer size in general? Let's take a look at the numbers!

Overall, the comparison yields a dead-heat with razor-thin margins. The 1 meg 5120 slides past the 512k version by a negligible 1% in the Business Disk WinMark under Windows 95. The High-End Disk WinMark displays a 0% difference. Under NT, the Business Disk WinMark shows no difference between the two drives, while the High-End tests yield a surprising but still negligible result: The 1 meg drive actually lags behind the 512k version by 3%.

A closer look at component High-End WinMark scores is warranted. Here, three results leap out with margins higher than the rest. Under Windows 95, the 1 meg Maxtor outperforms the 512k drive by 10% on the MicroStation SE test. The situation reverses itself when it comes to Adobe Premiere tests- here the 512k version pulls ahead by 10%. It seems these two results balance each other out, netting identical performance in the average score. In Windows NT, however, the 1 meg drive's advantage in MicroStation is erased, yet the 512k drive manages to stay ahead of the 1 meg disk by 9% in Premiere. Thus, the drive with the smaller cache actually pulls ahead slightly in the overall score.

ThreadMark presents virtually identical performance for the two drives. In both Windows 95 and Windows NT, the 1 meg 5120 edges by its 512k brother by margins of less than 1%.

What should we conclude from these differences? First and foremost, these results show that your opinion of the DiamondMax Plus 5120 should not change one bit. If you thought it was a great drive, the 1 meg unit is still a great drive. If, however, you feel the 5120 falls short in areas that count, the extra cache doesn't help one iota.

Looking at the bigger picture, however, the results displayed here show us something that shocks "common wisdom": A larger buffer in and of itself is no asset and makes little difference in performance. One of our more prolific participants in the StorageReview.com Discussion Forum hypothesized such a theory a while back. Rather, it's the ineffable, undisclosed "firmware/drive electronics" that make more of a difference.

Take a look at the likely scenario here: buffer size is doubled with little change in firmware. Net result? Little change in speed. Take a look at the converse: the DiamondMax Plus 5120 and Fireball Plus KA, though possessing only 512k of buffer, nevertheless manage to significantly outperform the 2-meg-buffer IBM 22GXP/WD Expert in several tests. What can we ascribe the excelling results to other than superior firmware/programming?

An increased buffer size and corresponding changes/improvements in algorithms/firmware may very well yield improvements one would expect. Indeed, improvements in algorithms/firmware alone can propel a drive to the forefront (again, observe the Atlas 10k vs the Fujitsu MAG3182LP). As shown here, however, increasing buffer size alone is virtually meaningless. It's quite a pity, as buffer size is quantifiable and thus "spec sheet-able" while superior routines are more abstract and not an easy target on which a manufacturer may focus marketing muscle.