As soon as we posted our article called HDD Roundup: 7 Hard Disk Drives with 20GB Platters, we got the whole bunch of e-mails from our readers demanding the tests of IBM HDDs. Well, why not? This review is devoted solely to the hard disk drives by IBM and their performance against the background of the already considered testing participants.

We will be taking a closer look at Deskstar 40GV and 75GXP, which were manufactured a while ago already. However, they were so much ahead of their time when launched, that even now they remain very up-to-date and quite competitive. High data density plus fast mechanics - this was a brilliant combination, which helped these models win the customers' hearts. Even if you are already familiar with IBM DTLA hard disk drives, this review is worth your attention, because we applied a number of new tests and carried them out in different operation systems.

In our roundup we will compare the performance of IBM drives with that of the latest models discussed in the previous HDD roundup: Quantum FB Plus AS, Seagate Barracuda ATA III, Seagate U5, Western Digital Protege (WD200EB), Western Digital Caviar (WD200BB), and finally Fujitsu MPG3204AT. We will divide all our tests into two parts. There will be two groups of HDDs: with 7200rpm and 5400rpm rotation speeds.

Besides, we will also apply a new IBM HDD utility aka IBM Feature Tool, which allows changing the noise-to-performance ratio for IBM DTLA drives (Deskstar 40GV and Deskstar 75GXP).

Well, and the last thing, before we start our story. Actually, it looks as if only 50% of all readers manage to read the reviews to the end. That is why we decided to briefly mention the basic conclusions in the very beginning to excite you a bit :-)

Our tests showed that IBM DTLA HDDs performed just brilliantly not yielding anywhere to their competitors. IBM DTLA 307020 proved the fastest almost in every single test although other hard disk drives boasted higher data density per platter. Among the drives with 5400rpm rotation speed, DTLA 305020 won the second prize having let WD Protege forward. To tell the truth, the difference between them is so insignificant that we believe WD Protege owes its victory only to a four times larger cache-buffer.

So, even today IBM hard disk drives remain the performance leaders. The only one, which may take the laurels from them, is the new IBM 60GXP model.

Well, are you curious to find out the details now? Then go ahead and read more!

Testing Participants

We will not dwell on the peculiarities of all the HDDs tested, since we have already discussed them in our HDD Roundup: 7 Hard Disk Drives with 20GB Platters. We would like to say a few words about the main heroes of the today's story - IBM hard disk drives.

The first model is IBM Deskstar 75GXP (DTLA 307020). It features 7200rpm rotation speed and a 2MB buffer. Unlike all other testing participants, this hard disk has two platters with 15GB per platter data density and three heads instead of two. Another peculiarity of this baby is the fact that if you use three sides (7.5GB each) completely, then the total space used should be equal to 22.5GB. However, for some reason - most probably for some marketing purposes - IBM decided to limit this value to a solid number: 20GB. Judging by the linear read graph, IBM simply reduced the number of cylinders used in this HDD model, i.e. the platters are used not to the full extent. So, these cylinders "finish" in the interval with higher data density.

Here is the linear reading for IBM DTLA 307015 (1 platter):

You can easily count the number of intervals with similar data density: there are 15 of them. And now take a look at IBM DTLA 307020 (2 platters):

There are only 12 intervals (by the way, the fewer intervals there are, the better average seek time your drive will show). It doesn't matter that much for the tests, because almost all of them are run in the first interval with the highest data density. However, if the tests were run in the end of the disk drive then IBM DTLA 307020 would be definitely much faster than DTLA 307015. For our review we selected a HDD with 2 platters on purpose, because we were guided not by the principle "one hard disk drive - one platter", but wanted all our participants to be of the same storage capacity.

The second model is IBM Deskstar 40GV (DTLA 305020). This one complies with all our requirements: 1 20GB platter, 512KB cache buffer and no "outstanding surprises".

Well, here are the technical details of the hard disk drives tested. First come those with 7200rpm rotation speed:

To tell the truth, they look like twin brothers, don't you think so? The major differences can be noticed only in the last row. This parameter directly depends on the number of cylinders and fastness of the heads block.

And now here are the drives with 5400rpm spindle rotation speed:

The differences are again quite insignificant. Only Western Digital Protege WD200EB stands out a bit due to its 2MB cache buffer and very long average seek (full stroke) time.

Testbed and Methods

Our test system was configured as follows:

• Intel Coppermine 600MHz CPU;
• ASUS CUBX-E bios 1007A;
• 2 x 128MB PC133SDRAM by Hyundai;
• Matrox Millennium 4MB graphics card;
• Windows 98/Windows 2000 Pro.

Now we would like to say a few words about our testing methodology.

In order to check the HDDs performance in different UDMA modes, we had to use the following controllers:

• UDMA33: the controller integrated into i440BX chipset;
• UDMA66: Promise Ultra66 controller;
• UDMA100: integrated Promise Ultra100 controller.

For the Promise controllers we used the drivers ver. 1.60 (build 33). In fact, these drivers were initially intended only for Promise Ultra100, however, after a slight modification of the inf-file, they could be also run for Promise Ultra66.

The disk drives were connected as Master-units to a separate IDE-channel. We used FAT32 and NTFS file systems to format each of them as one logical drive of the maximum size with the default cluster. DMA support in Windows was enabled. All the tests were run 4 times and then the average results were taken for the diagrams. The HDDs didn't rest for cooling down between the tests.

We used the following benchmarks in our review:

• Windows 98:
  • WinBench 99 1.2
  • Adaptec Threadmark 2.0
• Windows 2000
  • WinBench 99 1.2
  • HDTach 2.61
  • IOMeter 1999.10.20

All the hard disk drives were supplied with the UDMA100-protocol enabled as default. In order to get more precise results and to avoid conflicting situations we used special utilities to switch the modes to UDMA33 and UDMA66 correspondingly. For your convenience we can offer you these utilities for download here. Some of them, however, can't work properly with the hard disk drive connected to an external UDMA controller that's why we would advise you to connect your HDD to the integrated controller before switching the UDMA mode with the help of these utilities. The main thing you should remember about is to make sure that you returned everything to the initial position :-)

So, you can download these utilities here:

• Quantum FB Plus AS
• IBM
• Seagate
• Western Digital
• Fujitsu

Performance

Access Time

As you remember, the hard disk drive, which took the lead here, won the overall competition in our previous testing. Therefore, we would like you to take the average seek time parameter quite seriously.

IBM Deskstar 75GXP (DTLA 307020) managed to leave behind the winner of the previous roundup, Quantum Fireball Plus AS. Probably, it owes this success to its dual-platter construction.

The slowest one again appears Seagate Barracuda ATA III.

The situation mirrors the previous graph with that only difference that we have Fujitsu MPG3204AT drive instead of Quantum.

Well, IBM HDDs proved to own the fastest mechanics in both groups. Now let us see whether it will tell on the results in file tests.

Data Density

In order to measure the maximum and the minimum data density we resorted to Disk Inspection test from ZD WinBench package (the table below contains the results obtained for UDMA-100 mode):

The champions here turn out both Seagate hard disk drives and the IBM models are just a little bit behind them. Note that IBM Deskstar 75GXP (DTLA 307020) appeared better here than Quantum Fireball Plus AS, which features platters with higher data density. Take a look at the diagrams to get a better idea of the results:

As we have already pointed out in the previous article, WD200BB boasts very strange data density geometry. The density in the beginning of the disk is not that high and it falls down very slowly. So, by the end of the disk the tracks feature pretty high data density, so that this hard disk drive appears the second after Seagate Barracuda ATA III. IBM Deskstar 75GXP (DTLA 307020) is just a few points behind.

All the three HDDs prove very close to each other, and only WD200EB falls behind them for some mysterious reason.

Here Seagate U5 managed to retain its leadership while IBM Deskstar 40GV (DTLA 305020) turned out only the fourth.

Due to higher data density in the beginning IBM hard disk drives can compete with those HDDs featuring 20GB platters on equal terms. The closer is the end of the drive, the faster drops the data density by IBMs, which undoubtedly tells negatively on the read speed.

WinBench99 for Windows98

First, see the results for HDDs with 7200rpm:

In Business Disk WinMark IBM Deskstar 75GXP (DTLA 307020) yielded to its competitors, however, in High-End it took revenge. So, we can't say that the data density per platter predetermines the results in these tests.

Well, what about the second group?

We have always said that a cache can never be too large for a hard disk drive and these diagrams seem to be proving our words. WD Protege (WD200EB) with a 2MB cache won the race in all modes in Business Disk WinMark. The second position is firmly held by IBM Deskstar 40GV (DTLA 305020). If it had a bit larger cache then it would have undoubtedly beaten WD Protege boy, but...

In High-End Disk WinMark IBM Deskstar 40GV (DTLA 305020) is ahead of the competitors in UDMA100 and UDMA66 modes, but in UDMA33 something unexpected happens: the leader yields to WD Protege (WD200EB).

All in all, IBM Deskstar 40GV (DTLA 305020) looks not bad at all and deserves to share the first prize with WD Protege (WD200EB) in the 5400rpm category.

Adaptec Threadmark 2.0

Although in UDMA100 mode IBM Deskstar 75GXP (DTLA 307020) performed not in the best way, it still proved formally better than all other drives tested, because it surpassed everybody in two modes out of the three. Therefore, we announce it a winner.

Among the drives with 5400rpm rotation speed, an indisputable leader is IBM Deskstar 40GV (DTLA 305020).

WinBench99 for Windows2000 (FAT32)

When you change the operation system, the disk subsystem starts performing a bit differently, because it is not only the HDD fastness that tells but also the software environment: the operation system, implementation of caching algorithms, various buffer settings and the drivers, of course.

IBM Deskstar 75GXP (DTLA 307020) won in UDMA66 and UDMA100 modes having left the competitors quite far behind. Only in UDMA33 it yielded a bit to Quantum Fireball Plus AS. It was in the previous roundup already that we were surprised at the really small performance drop by Quantum Fireball Plus AS in UDMA33 mode, but this time we can say almost the same good words about IBM Deskstar 75GXP (DTLA 307020). Seagate Barracuda ATA III and WD200BB ran very close to Quantum Fireball Plus AS with UDMA66/100 controllers, however in UDMA33 they were totally destroyed. Anyway, the second winner is Seagate Barracuda ATA III.

Again IBM Deskstar 75GXP (DTLA 307020) is ahead of all, and the second prize in the two out of three tests belongs to WD200BB.

Now let's pass over to the second group of hard disk drives:

Well, WD Protege (WD200EB) is a formal leader. However, just take a look how close IBM Deskstar 40GV (DTLA 305020) is to the winner, even though it has a four times smaller cache! Frankly speaking, we can't perceive why on earth the HDD manufacturers stick to small cache buffers...

In general, the situation repeats. WD200EB is just a bit faster than IBM DTLA 305020, Seagate U5 is just a bit faster than Fujitsu MPG3204AT.

Actually, the results obtained in Win2000 are generally higher. IBM Deskstar 75GXP (DTLA 307020) has every right to be called the winner, while the slower model, IBM Deskstar 40GV (DTLA 305020), has just a little bit fallen behind WD200EB.

WinBench99 for Windows2000 (NTFS)

Here we changed not only the operation system, but also the file system: the disk drives were formatted in NTFS with the default cluster equal to 4KB. But did the leader remain the same? Let's find out...

IBM Deskstar 75GXP (DTLA 307020) wins in both tests. Then comes WD Caviar. The third and the fourth prizes are shared between Seagate Barracuda ATA III and Quantum Fireball Plus AS.

In Business Disk WinMark IBM Deskstar 40GV (DTLA 305020) performed just like WD200EB, however in High-End it couldn't stand the competitor's attack anymore.

Smaller cluster size (4KB) led to the general lowering of the results, but didn't tell on the situation. Among the drives with 7200rpm the winner was IBM Deskstar 75GXP (DTLA 307020) and among those with 5400rom - WD Protege (WD200EB).

HDTach 2.61

Seagate Barracuda ATA III shows the highest read speed of all, but seems to be just allergic to UDMA33 mode. It's interesting that IBM Deskstar 75GXP (DTLA 307020) demonstrates the same read average speed as Quantum Fireball Plus AS, although the latter features higher data density per platter. This explains the "unusual" construction of IBM fellow, which we have already described in the very beginning of our article.

And for the write average speed the construction peculiarities of IBM Deskstar 75GXP (DTLA 307020) do not work any more and it lags behind all the new generation HDDs with 20GB platters. Seagate Barracuda ATA III again leads in all the modes except UDMA33.

Here all the hard disk drives perform in equal conditions, because all of them feature 20GB platters. The drive with highest data density per track, Seagate U5, dashes forward. Then come Fujitsu, IBM and the last one is WD200EB. This is really strange since this HDD used to be the first in all file tests...

To tell the truth, writing is not a trump of IBM hard disk drives. IBM Deskstar 40GV (DTLA 305020) turned out slower than the competitors here. The highest results were obtained on Seagate U5, and the most stable performance belonged to WD Protege (WD200EB).

Intel IOMeter

Certainly, this is a pretty specific test. It was initially developed to test servers therefore some server peculiarities are now applied to single hard disk drives testing.

We suggest discussing briefly some features, which determine the server fastness. As you know, the server should spend the minimum of time on reaction to requests or should be capable to processing as many requests as possible during a certain period. For this purpose all "server" hard drives get a larger cache, lower average seek time and higher spindle rotation speed in order to reduce the rotational latency. Since the data on a disk drive like that is usually requested quite randomly, fast positioning appears the major factor influencing the performance. So, we may try to suppose that the hard disk drive with the lowest average seek time will be the one to win the first prize in this test. Well, we'll see right away if this is right.

Here we would like to veer a bit away from the current testing and to return to the previous roundup. Last time we noticed that our results obtained in IOMeter were somewhat (quite a lot, in fact) different from the results obtained by the guys over at StorageReview. Since we used almost the same test platforms, this difference should have been caused by different program settings or different drivers. As we found out, the drivers version 1.60 (build 33), which we utilized for Promise Ultra66 controller, do cache the HDD (even on a lower level) but only for a formatted drive! Therefore, these "super"-drivers distorted the results of the tests having added some software "optimization" because we did use only formatted drives that time. Of course, this caching is quite useful for real work but it prevents us from understanding the true dependence of the HDD performance on its parameters.

At first, we will show you what we obtained in case of unformatted drives and then try to illustrate the differences between the tests with and without "extra" caching.

First let's look at the dependence of Total I/O on the load:

In case of Linear load (1 outstanding I/O) all the HDDs performed very close to one another. However, the greater got the number of outstanding I/Os, the larger grew the gap between the drives tested. IBM Deskstar 75GXP (DTLA 307020) appeared the most sensitive to the load increase. This drive seems to feature not only the fastest mechanics but also the fastest electronics. Very interesting results showed another 7200rpm HDD, Quantum Fireball Plus AS: at first it surpassed WD200BB and then "let" the competitor catch up with it.

And now let's compare the "Sum Total" values of our participants:

Since we are mostly interested in the HDD performance in the most "loaded" modes, Sum Total was calculated as a sum of Total I/O in the last three modes: Light + Moderate + Heavy. And since the greater was the load, the farther ahead of its rivals got IBM Deskstar 75GXP (DTLA 307020), the overall difference between IBM drive and all the rest appeared just immense! Quantum Fireball Plus AS and WD200BB ran almost neck and neck, Quantum was just a tiny bit ahead. Seagate Barracuda ATA III fell considerably far behind, which wasn't surprising at all, because it had the highest average access time.

Well, it's high time we paid a bit of our precious attention to the drives with 5400rpm. :-)

The picture is very similar to what we saw in case of 7200rpm drives (Intel IOMeter: Workstation Access Pattern), don't' you think so? IBM Deskstar 40GV (DTLA 305020) starts a bit worse than some of its competitors, but then it makes up for the bad start and leaves them all far behind. Note that in this case the explanation about more platters, etc. doesn't work any more, since IBM Deskstar 40GV (DTLA 305020) has only one platter. The graphs for Fujitsu and WD drives interlace like two snakes and Seagate is again lagging behind, unfortunately...

The failure of Seagate U5 in Database Access Pattern is very illustrative. Unlike Workstation Access Pattern, which makes 80% read requests and 20% write requests, Database Access Pattern makes 67% reads and 33% writes.

We have just discussed the results obtained in the benchmarks carried out for the unformatted HDDs that's why they are very much different from the numbers you may have seen in the HDD Roundup: 7 Hard Disk Drives with 20GB Platters. As we have already told you, this happened because we veered slightly away from the standard testing methodology described on StorageReview having applied Promise version 1.60 (build 33) drivers with the integrated caching algorithm. Now let's find out how greatly this fact influenced the testing results.

The results of IOMeter test carried out for formatted hard disk drives turned out absolutely unpredictable.

Without integrated caching, Quantum Fireball Plus AS fell quite tangibly behind IBM DTLA, while with caching algorithms involved they performed almost equally. And in case of Western Digital, for instance, this caching didn't have any effect: neither positive nor negative.

So, you see that it really makes sense checking whether the HDD is formatted or not. That's why we will try to avoid this kind of confusion from now on. All the further tests will be carried out for unformatted hard disk drives.

We have every right to call IBM hard disks just brilliant in this test. Moreover, they proved "born server HDDs". We wish their reliability were as perfect as their speed…

IBM Feature Tool Utility

Some time ago when all HDD manufacturers were very anxious to introduce the most noiseless solution, IBM offered some utility on their web-site. This utility worked with IBM hard disk drives, which featured Automatic Acoustic Management. Now this option is supported only in IBM Deskstar 40GV (DTLA 305020) and IBM Deskstar 75GXP (DTLA 307020), however, IBM claims that all the upcoming HDD models, including 60GXP, will be also provided with it (download IBM Feature Tool here).

This utility allows "slowing down" the heads positioning, which reduces the noise level of the working HDD. When you run ibmftool.exe, the file creates a system (PC DOS) disk with IBM Feature Tool on it. It looks as if you'd better switch the AAM (Automatic Acoustic Management) modes from DOS rather than from Windows.

When you boot up your PC from this disk, the utility starts automatically and scans all the available HDD controllers. In our case Promise Ultra66 was approved, however, Promise Ultra100 wasn't. In situations like that we prefer to avoid the risks and connect the HDDs to standard IDE controller integrated into the chipset. All the hard drives detected during scanning are added to the list of available HDDs. The AAM parameters are set in a very simple way: you either agree with the suggested default value (128), or set it manually with a special slider. We didn't want to invent anything and were quite happy with the default settings. However, you may select any value lying between 128 and 255.

The first thing that struck us dumb was the total absence of any noise during heads positioning. For the system harddisk we selected IBM DTLA 307015 and switched it to silent mode as well. In the beginning we thought that the system block simply died. Our first intention was to rush to the poor thing and to save it at any rate. However, it turned out absolutely safe and sound and even managed to quietly start the Windows while we were nerving around. At a glance it worked as fast as usual, all the applications opened without any delays.

In order to check how greatly this operation told on the HDD performance in the tests, we decided to carry out only three of them:

• Disk Winbench99 1.2;
• HDTach 2.61;
• Intel IOMeter.

It was not a random choice. WinBench shows how fast the application files are processed, HDTach allows checking the access time and IOMeter - the dependence of the heads fastness on the overall HDD performance.

The results were simply impressive: the average access time doubled and the performance remained as good as before!

The model with 5400rpm rotation speed showed a bit worse results here, while its faster counterpart suffered a drop within 2-3% only.

Graphs	Graph	Graph	Graph	Graph

The situation is just the same as in Disk WinBench: the change of access time doesn't lead to any lowering of the benchmarks results. Will the same thing happen in IOMeter?

Since in Linear mode IOMeter practically gets the average access time and that's it, the twice as slow operation of the heads block (significant increase in average seek time) results into a double performance drop. And then - no other linear dependence can be observed. Just look at the diagram below:

As you can see, slower operation of the heads block adds a certain constant to the general dependence (see the difference between the graphs along the Y-axis?). It is pretty curious that in case of maximum workload IBM DTLA 307020 with slower heads block proves as fast as DTLA 305020 in the standard mode.

Heat and Noise

We measured the heating with the external thermal diode connected to the mainboard. IOMeter takes over 2.5 hours to run just once and the HDD works quite tensely all this time that's why we may regard it as normal thermal conditions for the drive. The HDDs tested were installed into the PC case, into a free bay without any special fans. The room temperature was around 22 degrees Centigrade. As soon as the test came to an end we restarted the system and took the temperature from the system BIOS (1007A):

Well, as we have expected a dual-platter HDD was a bit warmer, but not too much. Since all the tests were run in the case, even a single-platter drive got 5 degrees warmer than in the previous roundup.

As for the noise level measuring, we didn't resort to any special means. We simply listened to the drives working. Here are our impressions:

• Without the AAM enabled IBM Deskstar (DTLA) drives appear louder than their competitors.
• With AAM enabled the heads move absolutely noiselessly and IBM Deskstar (DTLA) drives appear the quietest of all.

Conclusion

Strange as it might seem, but IBM Deskstar 75GXP (DTLA 307020), an older generation HDD with 15GB platters, turned out faster than its more modern competitors with 20GB platters almost in all the tests.

IBM Deskstar 40GV (DTLA 305020) performed not as brilliant as its elder brother and yielded a bit to WD Protege (WD200EB) with a four times larger cache-buffer. All the others were beaten in all the tests. We would also like to point out that this HDD showed very good results in a "server" test - Intel IOMeter. According to it, IBM Deskstar 40GV (DTLA 305020) is capable of running close to Quantum Fireball Plus AS and WD Caviar (WD200BB), i.e. performs just like the faster drives with 7200rpm.

IBM Feature Tool utility allowed bringing the noise to naught and hardly told on the performance, having caused a slight drop only in file applications. We wouldn't dare "promote" this utility, because, as for us, the sound of moving heads doesn't get on our nerves that much. However, it is a really pleasing thing that you can now adjust the noise-to-performance ratio of your hard disk drive in a pretty wide interval.