The shortage of 16Mbit memory chips we wrote about in our Western Digital 1200JB HDD Review has grown more urgent during the last year. There are so few 16Mbit chips produced now that most manufacturers have to use 64Mbit chips in their hard disk drives. Samsung only stands its ground, as it owns production facilities to provide the smaller cache-buffer for its HDDs…

OK, let's get serious! :)

Some time ago we tested hard disk drives in different UDMA modes and witnessed a curious phenomenon: the performance of a drive didn't greatly differ in UDMA66 or UDMA100 protocol. Then we tested Maxtor drives supporting UDMA100 and UDMA133 interfaces and once again found no perceptible difference in performance. So, we wondered if the cache-buffer size could be the cause of faster interfaces being less efficient than they could. Just think: at 100MB/sec bandwidth, the data stored in the 2MB cache-buffer can be transferred in 20ms. And it takes only 15ms, if the interface bandwidth is 133MB/sec!

As the HDD processes much smaller data blocks than 2MB, the difference in transfer speed of such a data block from the cache-buffer to the main memory via different interfaces didn't affect the overall performance of the hard drive. Making the cache-lines longer resulted into a smaller number of these lines (as the size of the entire cache-buffer remained the same) and this negatively told on HDD performance in modern operating systems where more applications are competing to access the hard drive first.

Right now the things are slightly different. IBM, Maxtor, Seagate and Western Digital all have IDE HDD models with an 8MB cache-buffer. But still there is only Maxtor, which produces drives with UDMA133 electronics, all the rest are quite sure UDMA100 speed is enough for their products.

We have seen throughout the year 2002 that 8MB-buffer drives from WD were an effective irritant to other companies. Western Digital caught the spotlight of test labs and, accordingly, customers. At the same time, the advantages of hard disk drives with an 8MB buffer compared with the ordinary ones were not so evident. We only saw the old 8MB-buffer HDD from WD outperforming the new WD2000BB model in our file-copy FC-Test (see our Western Digital Caviar WD2000BB HDD Review). But now we have got an opportunity to compare the performance of the 8MB-buffer drives from WD with analogous products from competitor companies.

Testing Participants

There are four hard disk drives from three manufacturers in our today's review. They are WD1800JB and WD2000JB from Western Digital, Deskstar 180GXP (IC35L180AVV207-1) from IBM and DiamondMax Plus 9 (6Y160P0) from Maxtor. They are all equipped with an 8MB cache-buffer and have similar storage capacities. We didn't include Seagate Barracuda V SATA HDD here, as it features a new (and unexplored yet) SerialATA interface. We will dedicate a special review to this drive, so be patient. :)

Now, let's have a closer look at each of the hard disk drives:

IBM Deskstar 180GXP (IC35L180AVV207-1)

This is the front-man of the Deskstar 180GXP family, which should now be called Hitachi Deskstar 180GXP. We will still call the drive IBM Deskstar 180GXP in this article.

In the snapshot below you can see that the 180GB Vancouver2 model looks different from the 60GB Vancouver2:

In our article called IBM Deskstar 180GXP HDD: Quiet Revolution we reviewed the junior drive from the Deskstar 180GXP family. Now we see that drives from this family may have different cases. We kind of counted on that, to be frank.

   

Judging by the pictures, the design of the major 180GXP model resembles that of Vancouver drives (see our IBM Deskstar 120GXP HDD Review).

The case conceals three platters of pure aluminum covered with the magic Pixie Dust and ferromagnetic layers. The motor goes on a fluid bearing, while the electronics has been thoroughly re-designed and is now faster and "smarter". The drive owes its faster electronics to the new onboard processor (working at a higher frequency?) and the "smartness" - to the TCQ (Tagged Command Queuing) technology.

By the way, the interface-related part of the electronics has been changed, too. The Deskstar 180GXP drive communicates with the controller according to the ATA6 specs, that is, at 100MHz frequency, while still clinging to 48-bit LBA addressing.

The "1" in the end of the drive marking "AVV207-1" means the device features an 8MB cache-buffer. But we have already said that…

Maxtor DiamondMax Plus 9 (6Y160P0)

We have already introduced Maxtor DiamondMax Plus 9 drives to you and, unlike IBM products, they all look alike:

   

This case should conceal two most progressive 80GB platters in the industry! Actually, Maxtor people announced they would use 80GB platters in all DiamondMax Plus 9 HDDs, but the examination of DiamondMax Plus 9 of 60, 80 and 120GB storage capacities proved that it was not so. In all drives smaller than 160GB Maxtor used 60GB platters, while the 200GB model seems to consist of three 66GB platters (some sources say they are 68GB platters). Unfortunately, we don't have the top model of the family, but are looking forward to meeting it one day. :)

DiamondMax Plus 9 uses fluid dynamic bearings and it's the only drive in our today's roundup to feature ATA133 electronics.

Western Digital WD1800JB

The exterior of modern WD drives hasn't changed, too. Why re-design a good thing?

When you take the drive in your hands, you are unconsciously expecting it to weigh at least a kilogram, but it turns to be surprisingly light for such a capacity.

Well, it's time we got used to 180GB accommodated in a 600-gram piece…

   

There are three 60GB platters inside. As for the bearing mechanism, Western Digital prefers usual, not fluid ones. As a result, the drive is rather noisy. Even when idle, the WD drive produces up to 3.5dB of noise. We need also to mention that the drive is very sensitive to shock during work. In order to reach higher data density, engineers pressed the read/write heads closer to the platter surface, so the gap between the platter and the heads becomes smaller. It's quite natural then for modern drives to be hypersensitive to hits and other shock during work. But if WD had used fluid bearings, like the competitor companies, its drives would have become more shock-resistant. One of the good features of the fluid dynamic bearing is its ability to dump vibrations (thanks to special channels inside the bearing where the incompressible liquid seeps during stress).

Western Digital WD2000JB

This product deserves a very close look. Western Digital is now producing the world's highest-capacity 200GB HDDs - WD2000BB and WD2000JB. They call them Drivezilla. This name is supposed to impress customers and terrify the competitors. :)

So, what is so interesting about this drive? First of all, it features 20GB bigger storage capacity than the competitor products. Maxtor also has a 200GB drive in their product range, but we haven't seen it yet, while WD2000JB is already in stores. The price is colossal, though…

The 2000JB model, just like the 1800JB, consists of three platters. But in order to notch the 200GB capacity, the platters were made denser: 66GB instead of 60GB. We will talk about the way they did it later on. Right now, let's look at the photos.

   

As you see, this HDD doesn't differ from the WD1800JB at all.

By the way, talking about hard disk drive storage capacity, we use the number specified by the manufacturer. And as we know, they think 1KB equals 1 thousand bytes and 1MB - 1 thousand kilobytes (by the same principle, 1GB = 1000MB). With lower-capacity drives we shouldn't notice this trifle at all, but now that the drive capacities have grown considerably, this "trifle" has grown rather big…

Look at the table:

As we see, the difference between the "real" and "tricky" gigabytes is over 7%! Thus, you can only write onto WD2000JB 186 "real" gigabytes, as there are 390721968 sectors on it. By the way, this is exactly the number (186GB) the Promise controller told us on testbed start-up.

Don't think we are seizing upon Western Digital: all HDD makers are now using these calculations for telling their drives storage capacities.

And we also don't say the makers are telling lies to customers. They notify everyone that they use the "1KB = 1000Byte" formula when specifying HDD capacity. But isn't it time we returned to the old good kilobyte? The memory modules issue is quite evident, so I'd better leave it out. :)

But back to the testing participants. The brief specs of the drives are summed up below:

* - HDD electronics is compatible with the ATA6 standard.

The table above suggests that Maxtor DiamondMax Plus 9 (6Y160P0) is the most outstanding drive of all, judging by the specs. Firstly, it is the only one to use (supposedly) two 80GB platters, but as a result has the smallest storage capacity among the reviewed drives. Moreover, it is not all clear about the Maxtor drive. For example, the mysterious "<9ms" average seek time aroused our most dark suspicions. This time, it is all right, though. The Maxtor's Web site shows unified specs for all drives of the DiamondMax Plus 9 family. As we have already dealt with every drive from this family (except the mysterious 200GB model), we can say that Maxtor produced several HDD models in this family differing by the platter size.

Maxtor's claims that all DiamondMax Plus 9 drives use 80MB platters are far from being true. Right now, only the reviewed today 6Y160P0 model is supposedly built of 80GB platters.

It's quite natural for models with different platters to have different average seek time. That's why Maxtor didn't put exact numbers into DiamondMax Plus 9 specs. That is also the reason why there are no track-to-track and full stroke parameters (for example, they were specified for the D740X-6L HDD).

IBM drive boasts fastest mechanics, while Maxtor drive claims to have the lowest noise level in Idle mode.

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.

The following software was used:

• HDTach 2.61;
• WinBench 99 2.0;
• Intel 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 Hitachi Feature Tool Utility. For WinBench tests all the drives were formatted in FAT32 and NTFS as one logical drive with the default cluster (to format the drives in FAT32 we used Paragon Partition Manager software). 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.

The reviewed hard disk drives had the following firmware versions:

• IBM Deskstar 180GXP (IC35L180AVV2-7-1) - FW: V260A63AF0
• Maxtor DiamondMax Plus 9 6Y160P0 - FW: YAR41VW0
• Western Digital 1800JB - FW: 63.13F63
• Western Digital 2000JB - FW: 63.13F63

Performance

HDTach 2.61

We traditionally start out with HDTach.

So, we have got much higher measured average access time values of the WD drives than the specified ones! The measured access time of the IBM Deskstar 180GXP of 180GB capacity is 0.4ms lower than that of the 60GB drive from the same family. But the result of the Maxtor drive is most amazing!

The measured average access time of this HDD is only 12.1ms. It's too good for a full-platter drive.

Let's compare the drives by average (throughout the whole platter) read and write speeds:

We see only the WD1800JB falling out of the bunch. The rest of the drives showed similar read and write speeds, with Maxtor being the fastest of all. We will find out later on why it could win this test.

WinBench 99 2.0

Lately, 8MB-buffer HDDs from WD used to win in WinBench. Now the competitor drives also have such a buffer. Wonder which one is going to be the leader now?

Here are the integral results in High-End Disk WinMark and Business Disk WinMark tests:

Here we are! Maxtor HDD is the leader now, while WD drives are at the end of the race. Note also that Maxtor showed different results working with different interfaces. And while UDMA133 positively impacted the performance of the drive in the High-End test, it also led to lower results in the Business test. Well, the conservation law still holds true. :)

You will see the diagrams for data-transfer rates in the end and beginning of the drive and for average access time a little later. Let's first check what we have in NTFS file system:

Well, the situation changes dramatically. The major action took place among the leaders, where the WD drives do not belong again.

IBM HDD has come in between the Maxtor's results shown with different interfaces. So, the most interesting thing is really this difference in Maxtor's results. We will see a little later why this happened. And now, the average access time diagram:

WinBench 99 produced slightly different results than HDTach, but the overall picture is the same. Maxtor has the lowest average access time, while the WD HDDs - the highest.

The next diagram shows the linear read speed in the beginning of the disk (the densest tracks) and in the end (the least dense tracks).

There is something to think about. Note the read speed of the Maxtor's last tracks: it is unnaturally high. When we see such a small difference in read speed between first and last tracks, we conclude that the drive has a "shortened" platter. That is, the inner, less dense tracks are cut off.

But in theory, the reviewed HDD should have two 80GB platters. If Maxtor DiamondMax Plus 9 (6Y160P0) does have two platters, then what was the original platter, which lost the tracks? Maxtor just can't have 100GB platters. :)

It's more probable that Maxtor used about 70GB (68GB?) platters in its HDDs with over 120GB storage capacity. Let's suppose that the 160GB drive consists of three platters. They cut off 50GB of tracks from the resulting 210GB (don't fear the number 50GB: it is just 17GB per platter). So, we have a 160GB drive. Its linear speed in the beginning would be higher than in the 60GB-platters Maxtor DiamondMax Plus 9 (6Y120L0). Moreover, this drive would have high read speed in end tracks as well as low average access time (as the heads will have to move at a shorter distance). We see all of these symptoms by the Maxtor drive participating in our roundup …

Now, let's dwell upon WD2000JB and WD1800JB. The first one has higher read speed both in the beginning and end. It means WD engineers didn't re-design the old 60GB platter to create the 200GB HDD, but developed a brand-new one. By the way, the platter of the WD2000JB has one zone more compared with the previous one.

Now you can have a look at the linear read speed graphs:

• IBM Deskstar 180GXP (IC35L180AVV207-1): Graph;
• Maxtor DiamondMax Plus 9 (6Y160P0): Graph;
• WD1800JB: Graph;
• WD2000JB: Graph.

Intel IOMeter: DataBase

The drives are processing SQL-like requests (data chunk size - 8KB, 100% random data chunk address).

At request queue = 1, both WD HDDs were at even when there were fewer writes to be done. But when there was more writing necessary, the WD1800JB shot forward. Note that both WD drives lose to IBM and Maxtor HDDs in the RandomRead mode, when the drive only receives read requests. It is directly connected with their higher access time.

Maxtor HDD behaves differently with different interfaces. Moreover, it does much better with the UDMA100 controller than with the UDMA133 one. Why so?

Suppose the HDD firmware chooses the cache-line size depending on the current UDMA mode. The higher UDMA, the longer the cache line. Thus, the drive connected to UDMA133 controller would have fewer cache-lines than the same drive connected to UDMA100 controller.

In the DataBase pattern, the data block address is taken at random so there is very little probability that the HDD receives requests to read or write two adjacent 8KB blocks in a row. So if the controller's firmware is aggressively pre-fetching data blocks of a larger size, that is, opens up a separate cache-line for every data block with a "new" address, the number of cache-lines will end up earlier in a drive with fewer lines like that. And all subsequent requests will oust the previously-cached data from the buffer, thus the HDD processor will have fewer chances to choose an optimal heads route. One way or another, the requests processing speed will drop down, as the HDD will not perform the requests in a "proper" moment.

Note also that Maxtor drive showed equal speed in the boundary modes with both controllers.

Back to the facts: IBM HDD is leading at small write operations share, but then loses its ground. It starts to "speed up" only after the write operations share reaches 40%.

At request queue depth = 16, the HDDs from WD don't lose to Maxtor in the RandomRead mode anymore, in spite of the Maxtor's advantage in average access time!

WD HDDs are far ahead when there are more write requests in the queue, although IBM Deskstar 180GXP (180AVV207-1) is a little better when the write operations share is rather small.

Maxtor DiamondMax Plus 9 (6Y160P0) performs worse than other drives in case of small write request share, but outruns IBM Deskstar 180GXP when there are more writes. It is interesting that the results of Maxtor's drive with the two UDMA controllers are nearly identical at this request queue depth.

The tendency gets stronger on further request queue increase. IBM Deskstar 180GXP (180AVV207-1) is the best at re-ordering requests in case of small write requests share, while WD HDDs boast the best lazy write algorithms. Notwithstanding its excellent average access time, Maxtor can't say anything where special optimization of the firmware is necessary. Well, we may expect Maxtor to prove better under some other type of workload …

Intel IOMeter: Sequential Read & Write

All as usual: the HDD is bombarded with a stream of read/write requests. The data block size is constantly increasing. The request queue depth is fixed at four simultaneous requests.

Sequential Read

IBM HDD was faster processing smaller data blocks (maybe the "new onboard processor" did the trick). WD drives were the slowest in "speeding up". They only notched their maximum reading speed at 16-32KB blocks. It's not a very good result, considering that IBM and Maxtor did it with 8KB blocks already.

An interesting fact: Maxtor hard drive with UDMA133 controller was a little faster than the same drive with the UDMA133 controller.

Sequential Write

WD drives were the fastest processing 512Byte blocks, but once again went behaved highly outstanding with 1KB blocks (see our Western Digital Caviar WD2000BB HDD Review). And WD1800JB outperformed WD2000JB in this mode. On further data block increase, WD drives followed the curve of IBM until WD1800JB reached its maximum at 16KB blocks and fell behind WD2000JB and IBM Deskstar 180GXP.

Maxtor once again preformed differently with different UDMA controllers. But now it did best with the UDMA100 one. It's curious, but looks as if the Maxtor were better reading in UDMA133 and writing in UDMA100 protocol.

Intel IOMeter: WorkStation

The Workstation pattern tries to simulate intensive work of a single user in various applications in NTFS5.

The results show HDD performance under a given workload. We use the following formula to calculate the overall HDD performance rating:

Performance rating = Total I/O (queue=1)/1 + Total I/O (queue=2)/2 + Total I/O (queue=4)/4 + Total I/O (queue=8)/8 + Total I/O (queue=16)/16 + Total I/O (queue=32)/32

As the HDD in a workstation isn't likely to have a long request queue, the "heavy" workloads have diminishing coefficients.

We got the following ratings after all calculations have been completed:

WD1800JB quite unexpectedly got ahead of all. Well, its excellent lazy write algorithms make up for its not very good average access time. Maxtor working in the UDMA100 mode got the second rank. Not surprising if we recall its performance in the DataBase pattern at queue = 1.

WD2000JB follows behind the leaders. Its big average access time must have caused the slowdown.

Intel IOMeter: FileServer & WebServer

The following patterns, as their names suggest, emulate a disk subsystem workload typical of File- and Web-Servers.

According to the results, IBM drive is the best choice for entry-level fileservers. It got beaten by WD1800JB at request queue = 1, but won all the other workload modes.

As we see, the excellent average access time of the Maxtor drive didn't help it here. It only worked in the WebServer pattern at request queue = 1. But as we increase the workload (increase the request queue depth), this HDD loses all its advantages. The drive is only performing reads in the WebServer pattern and we saw above, in the DataBase pattern, that the IBM HDD is beyond competition in reading.

As all workloads are equiprobable for a server, we calculate the performance rating here by averaging the Total I/O values:

The IBM HDD won both patterns. Its advantage over competitors is most evident in the WebServer pattern.

FC-Test: NTFS

Our last test serves to check HDD performance at creating, reading and copying files. NTFS comes first:

Here are the diagrams for every mode:

The Maxtor HDD showed the fastest file writing speed. Moreover, it was faster with UDMA100 than with UDMA133 controller. The only unexpected thing is WD1800JB writing files faster than WD2000JB, which has higher data density per track.

IBM is the best in reading file sets with Maxtor close behind…

WD HDDs are once again lagging behind, although WD2000JB caught up with IBM and Maxtor drives on ISO files. Note that the "denser" WD2000JB was faster here than WD1800JB.

Maxtor was brisk copying "large" file inside one partition, while IBM was better with the sets of small files.

WD drives could only compete with the leader in the ISO pattern. And WD1800JB outperformed WD2000JB once more.

Copying from one partition to another brings about no changes. Maxtor is best with large files, IBM - with small ones. By the way, we found a weak spot by the IBM drive: copying ISO files. It does it not only worse than the competitors, but also copies them slower than the files of a smaller size (see Install and MP3 patterns)!

Let's see if another file system brings anything new to the picture.

FC-Test: FAT32

Maxtor writes files the fastest of all in FAT32, too. The first position is occupied by Maxtor drive with UDMA100 controller, and the second - by the same drive with UDMA133 controller. IBM is the third best.

IBM is ahead in reading. It wins three patterns with large files and loses to Maxtor in the patterns with small files. WD drives come right behind IBM and Maxtor in all patterns.

The same Maxtor and IBM are leaders in copying, too. The latter seems to be improving its performance with ISO files. WD drives are outsiders, especially WD2000JB, which was much slower than its "younger brother" in the ISO pattern.

Well, we were hasty to compliment IBM: it ruined the results in the ISO file copy test once again. :(

Maxtor drives go unrivalled…Remember we said they would show up somewhere? FC-Test helped us to discover the strong point of this HDD: it is excellent at writing, reading and copying files.

Temperature

This time it was a hard task to compare HDD temperatures. The two HDDs from Western Digital don't show their temperature in SMART and Maxtor is talking nonsense about its temperature. So, we turned to the old way of measuring: with a remote infrared thermometer. But there was a problem here, too. The PCB of WD drives is turned with the chips facing the inside. This lowers the chance of chips damage with electro-static discharge, but also prevents us from measuring the maximum temperature of the chips.

As WD drives don't show their temperature in the SMART, DTemp utility doesn't work with them. The textolite temperature of the PCB by WD drives is very low: 33^oC only. It's just a few degrees higher than the air temperature inside the case (30^oC, InWin A500).

The upper cover temperature is usually lower than the temperature at the side of the case, as manufacturers often lay some special materials from the inside of the upper cover to reduce the noise. The side temperatures were very close: they ranged from 42^oC to 44^oC.

We would also like to add a few words about the reliability of the drives. Yeah, it's not common to talk about reliability in reviews, as it would be wrong to draw any conclusions with the example of one given HDD. But this time we can't keep silent…

The reason is that during the test session we lost WD2000JB (after a re-start of the testbed it clattered with its heads and was no more recognized by the system), and Maxtor DiamondMax Plus 9 (6Y160P0) got covered with bad sectors. Two drives of the four is a little too much, in our opinion. If the user spends so much money as if he were getting a SCSI drive, he must get SCSI quality as well.

The benchmarking results belong to other drives we took instead of the dead ones, of course. Let's hope we were unlucky to have not very successful samples as the substitute drives still work well.

Conclusion

Well now, after a year since the arrival of the first IDE HDD with an 8MB cache-buffer, nearly all the surviving manufacturers introduced drives like that. The record-breaking speed of WD HDDs from the JB series is now surpassed by IBM Deskstar 180GXP (180AVV207-1) and Maxtor DiamondMax Plus 9 (6Y160P0) in WinBench tests as well as in file-copy tests.

The HDDs from Western Digital were still faster than the rivals when it came to writing. There is one supposition we would like to make. WD1800JB and WD2000JB drives have different platters, but the same firmware. This may have prevented WD2000JB from showing its real performance. It's most notable in FC-Test results.

The Maxtor drive did well in WinBench and file-copy tests, but the difference in its performance with different interfaces is hard to explain only by the different bandwidth of the interfaces. It looks like the firmware of the HDD adjusts for the UDMA mode offered by the controller. But it's still a mystery to us what it really does.

The HDD from IBM showed high performance in all the benchmarks. While it wasn't the best in every test, it didn't suffer the same drastic failures as some other drives did. By the way, it survived all the benchmarks :).