X-bit labs - Print version

We reviewed the new Caviar drive from Western Digital boasting record-breaking storage capacityof 200GB, fluid dynamic bearings mechanism, 66GB platters, 2MB cache-buffer and much higherreliability. Learn more about this monstrous solution in our new article!

Western Digital Corporation has been pleasing us for a year with "delicacies": fast (7,200rpm spindle rotation speed) hard disk drives of huge storage capacities.

Not so long ago, it was IBM who wore the title of "king of monsters". There could be up to five platters in company's HDDs, but as the data density was growing, it became less profitable to produce multi-platter giants. When coming over to 20GB platters, IBM reduced their maximum number in a HDD to 3, while most manufacturers, Fujitsu, Maxtor, Quantum, Seagate and Samsung, thought it proper to limit themselves with only two platters. WD still sticks to the three-platter design.

In July 2001, Western Digital announced the highest-capacity 7,200rpm hard disk drive ever. That was the fantastic 100GB Caviar WD1000BB. Storage capacity like that could be accommodated in the drive by using non-standard 33GB platters. About the same time, there appeared competitor products with 40GB per platter density, but their maximum capacity was only 80GB. WD mastered 40GB platters in October 2001 and notched a new capacity record: 120GB.

We shouldn't forget one more "extraordinary" event that happened after the release of Caviar WD1000BB. Western Digital equipped its record-breaking hard disk drive with an 8MB cache-buffer. The drive was named WD1000BB SE. A bit later, 120, 80 and even 40GB models got a cache-buffer like that, the first of the two "B" letters in the marking got replaced with a "J" in this case. Until the fall of 2002, the 8MB buffer remained a kind of business card of Western Digital.

Closer Look

And in June 2002, WD announced its new hard disk drive family with storage capacities ranging from of 120GB to 200GB. That is, the data storage density doubled in less than a year! The 120GB and 180GB models are built of 60GB platters, while the one more colossus again uses non-standard 66GB platters. It's not quite clear, though, why the company didn't want to produce a single-platter 60GB model. :)

The new Caviar drives have something to boast besides the record-breaking storage capacity. They are the first HDDs from Western Digital to use fluid dynamic bearings. This is optional, though, while Seagate Barracuda ATA IV and V as well as Maxtor Diamond Max Plus 9 have the bearings mechanism as a must.

Of course, it would be impossible to exceed the 120GB bar without modifying the drive electronics. The new Caviar drives support commands with 48bit addressing, but the fastest data-transfer protocol is still UltraDMA100, not UltraDMA133. Let us remind you that these are two quite different things. The 48-bit addressing is a standard (included into ATA/ATAPI-6), while UltraDMA133 is under consideration as a part of the upcoming ATA/ATAPI-7 standard. More details about this can be found in our Maxtor D540X-4G Review called "Neptune: Missing Piece of the Maxtor HDDs Evolution".

And of course, Western Digital couldn't do without its own brand name 8MB buffer :). It is available by all the three new models. But the first one to arrive into our lab was a model with a classical 2MB buffer, but non-classical 200GB storage capacity.

One more thing distinguishing the new Caviar drives from their predecessors is the warranty period: it has been reduced to 1 year. The 8MB-buffer models still come with a three-year warranty, though.

Testbed and Methods

Before the tests the AAM register of all HDDs was set to OFF position (FAST mode) with the help of IBM Feature Tool 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.

Specifications

After we have got the next biggest hard drive in history, we quite naturally decided to compare it with the previous capacity champions. Thus, we will see what progress the Caviar drives have made in the passed year. Unfortunately, we couldn't get a WD1000BB model with 33GB platters (this model was transferred to 40GB platters), so we had to be happy with the results of WD1000JB (8MB buffer). Well, but nevertheless we will be able to see if double density can beat the four times bigger buffer :).

Someone may think that this is scarce info. But firstly, Western Digital is rather secretive about its products and, secondly, all the three drives are absolutely identical in their construction, so the only difference is the data density per platter. The density directly affects physical read/write speed only. And the speed is there, right? :)

So, the specified speed grew by 24.5% compared to the "40GB" generation. It looks rather nice. But then, we learn that 100% density increase gives us only 40% speed growth! The specified average access time remained the same.

One more new thing in the new HDD family is the firmware version. It's been radically changed:

So, we may expect the newcomer to differ a lot from its predecessors. We will see such great differences! Now, let's leave the official data and go over to benchmarking results.

Performance

HDTach 2.61

And here we encounter the first "complexity". The WD1200BB and WD1000JB do have the same average access time, but it's 0.6ms higher by WD2000BB! Is it a fortuity or a peculiarity of new models?

By subtracting the rotational latency, we will get the measured average seek time:

As we see, the last "flagships" fitted into the specified time, while the WD2000BB didn't.

One more surprise: the WD2000BB has much lower read burst speed. It's reduced to the level of Seagate Barracuda ATA IV and the HDDs of the now retired Fujitsu. It's all rather strange. The only thing to reconcile us with it is that burst speed doesn't greatly affect overall performance.

The next diagram shows us the evolution of the platter (let's not forget that read speed is directly connected with the linear density).

Let us remind you that the speed of the WD2000BB at the outer tracks was measured wrong. Actually, it's about 57MB/sec as you can see in the WinBench graph. But even if we take the correct speed measurement, we have no more than 15% progress over the WD1200BB. Where are the promised 24%? We can't be absolutely sure, but maybe WD cared more about higher data safety and increased the amount of service data per track. At least, the graphs of the tested WD2000BB and WD1800BB were ideally smooth.

WinBench 2.0

Note that WinBench shows lower read speed in the beginning of the drive for the WD2000BB and WD1000JB, while the "end" numbers are incorrect for all three HDDs. That's why we calculated read speeds in the beginning and in the end with the help of HDTach log-file. But this approach also started to give faults. HDTach also produces lower results for drives that have higher than 50MB/sec linear read speed.

So, it looks as if we will have to find something to replace these benchmarks with :(.

As we see, the WD2000BB can't boast anything tremendous. The Business result is lower because of higher access time. The High-End result grew up a little, but wasn't enough to overcome the one-year-old model with 8MB buffer.

The integral mark is a complicated matter. Let's better check every test one by one.

Nearly in all applications, the results of the WD2000BB and WD1200BB are practically the same. This indicates that the basic firmware algorithms remained unchanged. The only exception is Sound Forge, which is traditionally sensitive to lazy write implementation. It's pleasing that the worse read from the buffer speed didn't spoil the results of the new Caviar, even in FrontPage. The write to the buffer speed must be high by WD2000BB :).

In NTFS the new model definitely loses to its ancestors, although just a little.

Judging by the practically identical performance of 2MB and 8MB-buffer models, the main thing here is average access time.

Intel IOMeter

Well… The new one definitely copes better with reading. The WD2000BB outpaces the WD1000JB and WD1200BB quite noticeably.

The unaccountable write speed drops on 1KB blocks are transformed into full dislike of all that's smaller than 2KB by WD2000BB. In all other cases it is ahead, as we have expected.

Now, let's have a look at the graphs of the DataBase pattern. This pattern helps to reveal the HDD's character by means of the changing the write-to-read operations ratio.

This time we could limit ourselves with graphs for two workloads only: the maximum and minimum ones. Under all the other workloads the results are in fact interpolated.

Note the strange speed drop at 100% reads, which started under the lowest workload and reached the top under the maximum workload.

Hm, very strange behavior. The only explanation is some mistake in the firmware.

But as soon as there appear write requests, the results of the WD2000BB get back to the norm.

The numbers for mixed requests are calculated by averaging the results for every read/write requests ratio, excluding 100% read and 100% write cases.

We can see clearly that the WD1000JB with the oldest firmware is much worse than the other two Caviar drives. Western Digital did a good job on optimization of lazy write algorithms.

So, what can we say about the results of the DataBase? The WD2000BB is always slower than the WD1200BB and we can't always explain it by the difference in average access time.

The coefficient of efficiency of write caching is calculated as the average read access time divided by the average write access time.

In its new firmware, Western Digital managed to make the lazy write a little more efficient, although it has already been unattainably efficient compared with the competitor products.

Now, after we have checked the hard drives performance, let's watch them go through the patterns that emulate real-life workload.

The WorkStation pattern serves to imitate intensive work of a single user in different applications in NTFS5.

WD2000BB is 8% faster than WD1200BB. But WD1000JB (with its 8MB buffer) spoils all the joy as it showed even better results.

There is one point to be mentioned. Watch the graph of the WD1200BB. It begins much lower than the graphs of the other WD models, but gets closer to them in the end. There is a simple explanation. In a new, out-of-the box HDD, the actuator movements are damped, so the average access time is somewhat high until the HDD works at its full. We first saw this peculiarity of Western Digital drives in the far-away 1994. So, later on we will watch HDD results more closely and use "break-in" procedures when necessary.

As we see, WD1200BB wins in this pattern, although it's quite similar to the previous one. That's why we supposed our sportsman wasn't simply warmed up.

The next pattern differs from the above two as the data is only read from the HDD, but is not written onto it.

Well, now we once again encounter an inexplicable performance drop of the WD2000BB, when the workload is on the rise. The drive ends up in the last place. How come that Western Digital tried to correct smaller flaws in the firmware algorithms, but at the same time made bigger ones?

FC-Test

While Intel IOMeter is more oriented onto random requests and its results greatly depend on average access time, FC-Test, on the contrary, favors linear read/write speed.

As it's hard to read the table, we prepared some diagrams for you. There are nine of them, as usual :).

The results in two file systems don't differ a lot. Thus, WD2000BB is an indisputable leader at writing. Overall, the HDDs got ranked exactly by their "age".

The same picture is with reading, although WD1200BB is closer to the leading positions here.

The advantage in the "raw" speed of the WD2000BB vanishes at copying. When copying large files, it has nothing to oppose the 8MB-buffered drive with.

The ranking remains the same when copying the data from one partition into another. WD1000JB is one third faster than its younger relatives. Strange as it might seem, but WD2000BB is the very last one here. At processing small and middle-sized files, the three HDDs alternatively turn up the last.

The next diagram shows applications start-up time. It's based on results of reading Programs and Windows patterns and shows the average number of files read per second.

The new WD drive is a little better than its predecessors in reading small files. That's the only solace after not very impressive copying results.

Noise and Temperature

The new WD Caviar is similar to its predecessors as far as noise level is concerned. The spindle rotation sound is quite noticeable, while the heads buzz very little when seeking. The case temperature (measured in the middle of the side panel) after intensive IOMeter work was 46^oC, so it would be good to use additional cooling for this hard drive. And again, the drive doesn't report its temperature through S.M.A.R.T.

Conclusion

It looks like the introduction of the 48-bit addressing pushed Western Digital to revising the firmware, which hasn't changed for a few years. Pursuing this noble goal, the company made some mistakes:

While these two issues affected the overall performance very little, the slump of random read speed in case of the workload exceeding 16 simultaneous requests immediately told on the web-server speed. I assume it might be the peculiarity of the first version of the new firmware and it will be corrected later on.

On the whole, WD2000BB (just like the WD1800BB and the new WD1200BB) is not faster than the previous WD1200BB model (and also the last WD1000BB and WD800BB models), but it has bigger storage capacity and must be more reliable.

Not everything is clear with the increased average access time. This may be an aberration of this given HDD or an undocumented feature of the whole generation. We also don't have the results of the new HDDs with 8MB cache-buffer so far. So, we will be searching for other models of the new drives from Western Digital to test.

Western Digital Caviar WD2000BB HDD Review