This review was not on our schedule, its idea sprang up spontaneously. And the review itself is rather out-of-track: we usually compare different hard drives of equal capacities, but today we will look at the drives of the same model (or nearly the same, as they differ not only by capacities). Well, let me explain.

When Maxtor D540X-4G hard disk drives appeared in stores, we took our time to study a new sample. But there was one peculiarity in this HDD that prevented us from putting the results up on our site. The point is that the minimal capacity of D540X-4G is 120GB and there was nothing to compare this drive with (WD1200BB and IBM 120GXP belong to another weight category). Meanwhile, as we looked at other models from Maxtor, we wondered more and more where this corporation was going to. And now, to dot our i's and cross our t's, we will compare D540X-4G with its junior mates, D540X-4D.

Both models feature 40GB platters, but D540X-4G was somewhat re-designed and has 4 platters. The main difference between the models is the interface: D540X-4G supports Big Drives and Fast Drives technologies. We have already told you about them in our Maxtor D740X-6L HDD Review, but it's time to expand our knowledge on the subject.

New Technologies

During the development of the ATA (AT Attachment) interface they had to deal with the limitations imposed over the hard disk drive capacity more than once. For example, the 528MB limit was caused by incoherent maximum number of tracks of the BIOS functions and the commands of the HDD (BIOS allows addressing 64 times fewer cylinders, but 16 times more heads). The solution was simple: they started substituting addresses in BIOS. It was then that another addressing system was introduced. It indicated the linear number of the HDD sector starting from the beginning of the drive (Logical Block Addressing). So, this is good evidence proving that already then they could have got rid of all traditions of the older cylinder-head-sector (CHS) addressing system, but it was much simpler to add address translation rather than change the whole thing. But anyway, when HDD capacities exceeded BIOS capabilities, LBA became the only choice. To overcome 8.4GB limit, BIOS acquired new functions, the so-called "Extensions of the 13th Interruption" (Int13Ext), which use LBA and not the cylinder, head and sector numbers. The next crisis happened in our time: LBA didn't allow addressing over 137GB. The matter is that the standard ATA-device registers store address in 28bits. Big Drives technology helps to solve the problem in a rather ingenious way: the address is written into the registers twice. High 24 bits come first, and then come the low 24 ones. So, the address length is 48bit and this allows addressing up to 144PetaBytes (144 million gigabytes). The register that stores the number of required sectors was also re-designed accordingly and now one command can process 32MB instead of 128KB! The importance of these innovations was clear to all ATA committee members (ANSI X3T13) and it was included into the ATA/ATAPI-6 standard. Things were different with another technology promoted by Maxtor…

Fast Drives means further development of UltraATA and introduces an additional data-transfer mode, UDMA-6. The bandwidth of the ATA interface thus reaches 133MB/sec, which equals the PCI bus bandwidth and PCI now is the industry standard for communication of all devices in the computer. Other HDD makers didn't support Maxtor (probably, displaying their independence), that's why the new protocol is still under consideration by X3T13 committee and is only going to appear in the future ATA/ATAPI-7 standard. Anyway, it doesn't prevent Maxtor from making ATA/133 drives and chipset manufacturers and ATA-controller makers - from producing ATA/133 compatible chipsets and controllers.

To close up the topic, let us prophesy a little. As the entire world seems to fancy SerialATA (which we don't share: it's a perspective technology, but until there is a faster local bus, it's just convenient, not more), the classical ATA interface won't develop any further. And do we need it? The data capacity reserve is simply colossal, and the bandwidth is twice as high as what could be used up now…

Review Objectives

The main question we wanted to answer in this review is how useful the ATA/133 interface is. It's no secret that drive capacities are growing much faster than the progress in read/write speed goes. Doubling of the data density hardly ever results in an even 20% speed growth. Our today's "guinea pig" couldn't even reach 37MB linear speed, so does it really need higher bandwidth? Of course we shouldn't forget that the specified bandwidth is the peak theoretical value and the real speed isn't even 90% of it, but the bandwidth reserve is still considerable.

The second aim of the review is to uncover similarities and differences in the new (ATA/133) and old (ATA/100) electronics from Maxtor.

And the last aim is to check the dependence of the tests results on the drive capacity, on the cluster size or on the HDD features.

Testbed and Methods

Our testbed was configured as follows:

• ASUS P3B-F mainboard;
• Intel Pentium III (Coppermine) 600MHz CPU;
• 2 x 128MB PC100 ECC SDRAM by Hyundai;
• IBM DPTA 372050 HDD;
• Matrox Millennium 4MB graphics card;
• Promise Ultra100 TX2 and Promise Ultra133 TX2 controllers;
• Windows 2000 Professional SP2.

We used the following software:

• HDTach 2.61;
• WinBench99 2.0;
• IOMeter 1999.10.20;
• FC-Test v0.3.

Before the tests the AAM register of all HDDs was set to OFF position (FAST mode) with the help of Maxtor AM Set Utility. For WinBench tests all the drives were formatted in FAT32 and NTFS as one logical drive with the default cluster. The tests were run four times each, the maximum result was taken for the diagrams. The drives didn't cool down between the tests. The tests in Intel IOMeter were run in SequentialRead, SequentialWrite, DataBase, WorkStation, FileServer and WebServer patterns. If you are looking for the detailed description of these patterns, please see our previous articles.

Testing Participants

The main hero of our today's review is Maxtor D540X-4G hard disk drive (code-named Neptune) with 120GB storage capacity (4G120J6 model number). As usual, the last digit in the model number is the number of read/write heads.

   

It will compete with its mates of smaller storage capacities. Below are the specifications different by all three drives.

The specs suggest the following: the more read/write heads has the HDD, the higher is the moment of inertia and thus the higher is the average access time. One- and two-platter models are a perfect example, but the three-platter one seems to be an exception. Let's check it all out!

Performance

HDTach 2.61

HDTach opens the show, as usual.

First thing to catch one's eye in this diagram is the considerable advantage of D540X-4G in read from the buffer speed even in the UltraDMA/100 mode. But the write speed here degrades quite tangibly and we can't write it off to HDTach measurement errors. We will see this gap in the next tests, too. The strange speed values in HDTach reveal a certain order.

The average access time by D540X-4G turned to be lower than even that of the 40GB D540X-4D model. Well done! By subtracting the average sector wait time (the specs say it's 5.5ms) from the average access time, we get the following benchmarked average seek time:

• Maxtor D540X-4G (4G120J6) - 9.3ms;
• Maxtor D540X-4D (4D080H4) - 11.1ms;
• Maxtor D540X-4D (4D040H2) - 9.8ms.

Low-level measurements showed that D540X-4G features improved average access time, higher interface bandwidth, but lower write speed in case of UltraDMA/100 protocol.

Winbench99 2.0

Here are some linear read graphs for your reference:

• Maxtor D540X-4D (40GB) (Graph);
• Maxtor D540X-4D (80GB) (Graph);
• Maxtor D540X-4G (4G120J6) with UDMA100 (Graph);
• Maxtor D540X-4G (4G120J6) with UDMA133 (Graph).

Let's start out with NTFS as it always has 4KB cluster size, whatever the drive capacity is. Anyway, the bigger is the storage capacity of the drive, the shorter is the moving distance for the heads during sector seeking (in case all the other conditions are the same), which should definitely be to the advantage of big HDDs.

Business WinMark results don't actually depend on the drive storage capacity, but D540X-4G is ahead of others even in the UltraDMA/100 mode. Unlike High-End WinMark, where ATA/133 D540X-4G is quite noticeably ahead of the others. As we see, higher drive storage capacity ensures good performance growth in this test, except the case when we have ATA/133 electronics working in the UltraDMA/100 mode.

But the total rating is no good matter for making judgments. Let's take a look at each parameter.

ATA/133 D540X is best on any protocol in three tests out of seven. In all others UltraDMA/133 helps it to keep the banner up.

In FAT32 the higher disk capacity necessitates larger cluster size, which should result in slower work with small files (with the size of a few sectors) and faster processing of large files.

As we see, an unreasonably large cluster (in a 80GB partition it could be 64KB?) negatively impacts applications speed. The 80GB model lost all its advantage over the 40GB one, which it had in NTFS. Anyway, the UltraDMA/133 protocol helps the 120GB model to keep its first place.

WinBench results suggest that:

• ATA/133 electronics algorithms differ a lot from those used in ATA/100;
• The new drive performs pretty badly with UltraDMA/100.

A few words on the low-level results. WinBench noticed no evident difference in access time between the 40GB and 80GB models, while the 120GB one is progressing. Who's right, HDTach or WinBench? Let's invite a third party, Intel IOMeter to figure out the answer to this question.

Intel IOMeter

The access time numbers here are taken from the DataBase pattern results, that is, for 8KB random data blocks, which is not very correct. Nevertheless, it suits well to compare three similar hard drives and from now on we will use a special pattern to measure access time. By the way, an express test showed that the difference in access time to one sector or 8KB doesn't exceed 0.3ms.

We see the 0.2ms difference between the 40GB and 80GB models that WinBench has also shown in NTFS. Read access time of the 120GB drive is lower than that of the 40GB model, but only by 0.3ms. Write access time here doesn't correspond to the physical one (it should be about 1ms higher than read access time) as the lazy write is allowed. As we see, new electronics can't boast efficient lazy write.

The diagram shows lazy write efficiency, which is calculated as the ratio of average read access time to the average write access time.

And now you can look at the graphs for HDDs tested in the DataBase pattern:

D540X-4G is much faster than its junior mates in reading, but slower in writing. It also demonstrates indifference to write operations share increase, which indicates simplified lazy write algorithms. The 40GB model is a little faster than the 80GB one thanks to lower seek time.

Now we will check sequential read/write.

The new ATA/133 electronics quite expectedly improved read and write speeds when processing data blocks smaller than 4KB. By some inexplicable cause, the speed of the one-platter model is 300KB/sec faster than that of bigger drives.

Strangely enough, more heads in the HDD result in slower linear write speed! By the way, D540X-4G seems to have serious problems with sequential write on the UltraDMA/100 protocol.

The WorkStation pattern tries to emulate intensive user's work in various applications with NTFS5 file system.

As both: 40GB and 80GB models share the same firmware, the performance increase on transition to two simultaneous requests by Maxtor D540X-4D (4D040H2) should be regarded as an unsystematic error. It is interesting that D540X-4G working with an ATA/133 adapter is at even with its junior mates under small workload, but loses in all other cases. Access time is not everything for a hard drive!

WorkstationRating = IOps(1) + IOps(2) + IOps(4) + IOps(8)/1,5 + IOps(16)/2

The results in this pattern depend a lot on the write speed, so no wonder that the "smallest" model took the best rating.

The next pattern simulates fileserver workload.

The new electronics does a great job here. Note also that the 80GB model always loses to the 40GB.

FileServerRating = IOps(1) + IOps(4) + IOps(16) /1,5 + IOps(64)/2 + IOps(256)/3

D450X-4G wins here due to better optimization for high workloads.

The HDD in a web-server serves not only for reading data, but also writing them onto it.

D540X-4G feels at ease here. Seems like the high capacity, and not the electronics do the trick.

WebServerRating = IOps(1) + IOps(4) + IOps(16) + IOps(64)/1,5 + IOps(256)/2

Rather unexpectedly, higher capacity leads to much better results in this pattern.

FC-Test

All the drives showed about the same results. In order not to overload the review, we will choose only those diagrams where the difference between the drives is obvious.

We see that the ATA/133 model copies small files from one partition to another slower than the ATA/100 one, in all other cases the difference is minimal.

The diagram below shows the average time an application needs to start up. It's calculated basing on the time required to read the Program Files and Windows folders, as the value, inversely proportional to the average read speed of one file.

To sum it all up, all the drives showed the same performance in FAT32, while in NTFS the bigger ones were a little ahead.

Conclusion

It's clear that the implementation of Big Drives and Fast Drives technologies requires serious re-design of HDD electronics. The UltraDMA/133 protocol requires lower latency, which should result in higher performance, but sometimes we see performance worsening instead. We learned that the new electronics features simplified lazy write algorithms. On the other hand, this improves data security (we all remember FAT on ATA/100 drives crashing in Windows98), so let's not blame Maxtor. Moreover, the performance decreases not very much, by a few percents. The new electronics has serious problems with UltraDMA/100 protocol, where the HDD speed drops below all acceptable points. Maybe this problem has to do with the given firmware version and is no longer the case for the HDDs selling now.

We can also say that the performance drop observed by new 80GB-platter models (for example, Fireball 3) in some tests is not just bad luck. The poor results date back to the first ATA/133 drive (D740X doesn't count as it was developed by Quantum) and now it looks more like a trend. Although they did correct the bug with UltraDMA/100 protocol.

As for the hero of our today's review, it's got its rivals now. The 120GB capacity is nothing extraordinary today and many manufacturers, Maxtor also among them, rolled out new products with these storage capacities. Well, we have got something to compare them with now :).