by Nikita Nikolaichev
04/29/2003 | 11:09 AM
To tell the truth, I have been long waiting for this sacred moment. The first ATA (or almost ATA) hard disk drive with 10,000rpm spindle rotation speed is lying in my palm. It’s so nice to feel its weight (by the way, it is notably heavier than the common ATA drives) and its ribbed case. It’s so pleasing to realize that you’ve just touched the live legend, because this one is your FIRST one!
At every seminar with the manufacturers’ reps I asked the same question: Aren’t you going to release an ATA drive with 10,000rpm spindle rotation speed? And every time I got the same answer: We can do it technically, but the market is not ready for these drives yet. And every time I thought to myself: I don’t care about this damned market, I need it! :)
But, finally the market got into a really revolutionary situation when the users no longer want to buy 5,400rpm HDDs, and the manufacturers are anxious to start selling something faster than that (because it can be sold at a higher price). And we could clearly see that this revolutionary situation was about to grow into something bigger. However, the manufacturers were afraid to make the first step, because every pioneer can get not only glory and popularity, but also all complaints and public reproach. So, we admire those brave one who dare!
Well, Western Digital released their first SATA hard disk drive. But it was not the interface used that made this event an absolutely extraordinary thing. Besides, the first SATA drive has already been launched by Seagate Company (see our Seagate Barracuda Serial ATA V Hard Disk Drive Review). The major peculiarity of the new WD drive is its spindle rotation speed, which is equal to 10,000rpm. Really, only SCSI drives used to have this spindle rotation speed. And now Raptor dashed onto their territory…
Yes, the new WD hard disk drive bears exactly this name: Raptor. Well, I don’t know what WD marketing people really implied, but I will do my best to find this out today.
The new hard drive, like a full-grown predator, should rush into an unsuspicious herd and catch the closest victim. Rip it into pieces and then get to the next one :)
Aggressiveness is one of the typical traits of all raptors, and it is exactly what WD expects its Raptor to be like in the today’s storage market. Why? Why should this drive be so aggressive? Let’s recall some history first.
A few years ago Western Digital Company gave up its SCSI business oppressed by grave technical problems and pushed by the competitors. The financial situation in the company was not very favorable in those days, and they couldn’t provide for the expensive HDDs production. It was a hard decision to make, but as the time passed we all saw that it had been the right decision. WD Company has fully recovered after that and sometimes they even afforded such naughty things as purchasing a fab from their competitor, Fujitsu. By the way, Fujitsu also gave up the fab not for fun. ;)
As we have already mentioned, WD started adding more high-capacity ATA solutions to their product range lately. Besides, they were the first manufacturer to equip their hard drives with a large cache-buffer, which improved their performance in some tasks quite tangibly. This couldn’t remain unnoticed by the customers and even the sales of the solutions with the regular 2MB cache-buffer grew up. Well, it looks as if the rule “first you work for reputation and then the reputation works for you” proved true again.
But there is one more rule, which the company should stick to if they want to become a success. In our rapidly changing world you should always run very fast, even if you want to stay at the same position. Well, hopefully WD managed to calculate the best time for a start.
And where does the red arrow in the WD marketing plans point at? According to the headquarters directive, this HDD should first of all aim at the corporate market, i.e. Low-End servers and high-performance workstations. As you see, this is not the tiniest goal...
As we are quite experienced generals already, we will estimate the armory first. What does a server HDD need? First of all it is high reliability. From this point of view Raptor looks very attractive: the claimed service time is 1.2 million hours, which is the same as by any up-to-date SCSI solution. And if something still happens to the drive, WD’s five-year warranty will come to rescue. The company approached this issue very seriously, which is a definite plus.
Secondly, the users expect a server hard drive to be fast, as the fewer HDDs will be needed to ensure a certain performance level, the better. Raptor should theoretically be quite fast with its 10,000rpm. Our tests will show if this is enough for success.
The third key feature of a server hard disk drive is the interface allowing to connect or disconnect the drive “on the go” and to build multi-drive arrays easily and at a relatively low cost. At first glance the new WD drive complies with this requirement as well. SATA interface supports HotPlug and allows building disk arrays very easily.
And finally, the last thing: server drives should be of large storage capacity. Unfortunately, WD cannot suggest an alternative to the 146GB SCSI monsters. Raptor drive is now available only as a 36GB version, so it will be a tough task to build a high-capacity array. However, the price of the newcomer, which is lower than that of SCSI drives of the same storage capacity, is a consoling factor. As well as the price of SATA RAID controllers, which is much lower than that of SCSI RAID controller cards.
Don’t you think it’s time to look this Raptor into the eyes? :)
Here it is. It features a completely new case with the ribbed surface providing better rigidity (at the same time these ribs help to dissipate the heat more efficiently). The motor has been moved from the location typical of all 7,200rpm drives, because the platters of a 10,000rpm drive should be of smaller diameter. Instead there appears more space for a more powerful positioning mechanism. Sure, as the seek time needs to be lowered.
And this is our Raptor without the case cover:
And now let’s take a look at the connectors:
Note that Raptor features two power supply connectors: a standard SATA connector and the older four-pin connector. By the way, WD recommends to avoid using both connectors at the same time. Just in case...
The PCB, which can be seen on one of the pictures above is placed with the chips facing the inside, which reduces the risk of EMI damage if the HDD will be installed into the PC case not quite professionally. But not al the chips are inside. One of them is left on the outer side of the PCB. What chip is that?
This is the ATA<->SATA converter chip from Marvell. Will WD give it up soon? Otherwise, why did they place it in such an awkward location?
The table below presents the specifications of the new WD Raptor drive compared with WD400JB:
As you see, both drives are of almost the same storage capacity and they feature cache-buffer of the same size – 8MB. These are the only similarities between them. Raptor marked as WD360GD boasts an impressive advantage over the other drive in the access time/
We will test Raptor with our standard test set and on our regular test platform.
The testbed was configured as follows:
ATA/100 hard disk drives were tested with Promise Ultra100 TX2 controller (BIOS: 2.20.0.14 Drivers: 2.00.29), and SerialATA drives were tested with two controllers: Promise SATA150 TX2 Plus (BIOS: v1.00.0.20 Drivers: 1.0.0.16 ) and SiliconImage SiI CP3112SATA150 (BIOS: 4.1.50 Drivers: 1.0.0.41).
We used the following software for our tests:
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:
We tested WD360GD hard disk drive with Promise SATA150 TX2 Plus in three modes:
As you may have noticed, the previous WT and WB set now includes one more mode, when Promise SATA150 TX2 Plus controller theoretically shouldn’t process the requests additionally.
In our article we will refer to these modes as follows:
What will we compare the new WD Raptor with? First, we will take the results for WD400JB, because it would be interesting to see how greatly the performance differs by the drives from the same manufacturer but with different spindle rotation speed.
Second, we will take the results of two SCSI HDDs and two IDE RAID arrays built of Maxtor D740X-6L from our article called “IDE RAID vs. SCSI: Who Will Win?” . The SCSI drives from Seagate and Fujitsu will be our reference point for server performance, and IDE RAID arrays will allow us to estimate the effect of a faster spindle.
And third, WD360GD drive will be tested four times with different SATA controllers and with different caching settings.
The first benchmark to discuss is HDTach. In the previous test session we discovered some issues about it, as you remember (see our Seagate Barracuda Serial ATA V Hard Disk Drive Review for more details). I wonder if it copes with the task this time (although after our investigation last time I offer you these numbers only for the sake of tradition).
So what do we see here? The claimed average access time of the WD360GD drive evidently proved to be absolutely correct. We only see a slight deviation in case of Promise (WB) controller. As we know, in WriteBack mode the Promise controller drivers do process the requests, which definitely requires some time. If these are random requests, the driver fails to benefit from requests caching and the time the CPU spends on requests history processing simply increases the overall access time.
However, if the driver receives multiple read requests for the same data block, like in case of HDTach read burst speed measurements, then the Promise drivers improve the HDD performance quite significantly. :)
Too bad, but HDTach test seems to be unable to work properly with these tricky drivers.
Now let’s try to measure the read and write speed with our “own” methods:
Of course, WD360GD is faster than its younger brother, WD400JB, when working with the data blocks of any size. But have you expected any other result here?
And if we analyze the performance of WD360GD with different controllers, we will have to point out stably high performance with SiliconImage controller and extremely high read speed with Promise (WB). From our previous article you know why the results look like this.
By the way, when we shift to larger data blocks (over 128KB), the read speed of Raptor + Promise controller combination drops down in all driver modes.
When we compared the read speed of all our testing participants we discovered that in case of a relatively small data block WD360GD yields only to SCSI drives and RAID 1 array. And when the data blocks are quite large, our hero is the second fastest after the RAID 0 array.
Promise (WB) controller shows even better results here, than during reads.
But look at this drastic write speed drop when we shift from 0.5KB blocks to 1KB blocks! Both WD HDDs seem to dislike the data blocks of this size. It is typical of all latest WD models, so we can call it a family feature :)
It is interesting that as the data block size keeps growing, the write speed onto WD360GD drive almost catches up with the write speed onto RAID0 array of two 7,200rpm IDE drives. And only with larger data blocks RAID 0 arrays managed to get ahead.
Now that we have already tested the Sequential Read and Write speeds of the new drive, let’s check its performance in the DataBase pattern.
The tables contain Total I/O values (the amount of requests processed per second) for different shares of reads and writes. Let’s start with the linear workload (queue=1):
At first, let’s compare the performance of WD360GD and other testing participants.
Not bad at all, don’t you think so? Excellent lazy write algorithms allow WD360GD drive to outperform one of the fastest 10K SCSI drives in mode tests! Not to mention Seagate Cheetah 36ES, which was completely defeated in absolutely all test modes. RAID 0 array of two IDE drives with 7,200rpm spindle rotation speed also appeared out of the actual competition. So, large cache-buffer, advanced lazy write algorithms and fast mechanics oust SCSI drives from the leading positions.
Now let’s see how the controller used can affect the performance of WD360GD:
As you see, Promise controller proves more efficient than SiliconImage when the writes share is bigger. It is especially noticeable in the results obtained for Promise (WB) controller.
Now let’s pass over to higher workloads:
This is shocking result! Look, how greatly SCSI drives improved here! Being able to change the requests processing order (command tag queuing), SCSI HDDs manage to get ahead of other testing participants, as soon as the queue depth grows over 1. WD360GD drive also got faster, but its success is nothing compared to the improvement of the SCSI drives. Well, it looks as if SCSI drives will remain most efficient under heavy workloads, unless the requests sorting is done by the hard disk drive and not by the ATA/SATA controller (in other words, unless the ATA/SATA HDDs learn to support command tag queuing). Although we have to admit that fast mechanics of WD Raptor helped it to beat both RAID arrays, and the gap between WD360GD and WD400JB looks more like the Grand Canyon. :)
To be fair, we should point out excellent implementation of lazy write algorithms by WD360GD: it managed to outpace Seagate Cheetah 36ES in case of high writes share.
Now let’s compare the performance of different SATA controllers:
It is evident that Promise controller ensures better performance of the raptor drive, and in RandomWrite mode WD360GD with Promise (WB) controller still managed to defeat Fujitsu MAN. And this is one more proof that proper controller matters a lot for efficient WD360GD performance.
Now the queue depth has been increased up to 256.
With this requests queue we get the maximum hard disk drive performance, which can hardly be ever obtained in real life. However, in this case you can clearly see the optimization of HDD firmware, or of the RAID controller driver, a sin our case:
Both SCSI drives are still ahead, but Seagate Cheetah 36ES outpaces Fujitsu MAN when the writes share is quite big. Both RAID arrays look pretty OK here, especially in case of low writes share. WD360GD drive lays itself out trying to beat RAID arrays and SCSI solutions. And although it manages to outperform RAID arrays when the writes share reaches 50%, the SCSI drives yield to it only in RandomWrite mode.
Now come the controllers:
As you see, all controllers performed almost equally fast, except Promise (WB). The latter needs extra time for requests processing, which slows it down noticeably with such a long requests queue.
So, it’s high time we tested WD Raptor in server patterns.
When we compared the performance of Raptor drive with that of SCSI drives and RAID arrays built of 7,200rpm drives, we discovered that it was faster than RAID arrays under linear workload (queue=1), and is equally fast with them under higher workloads. But as for SCSI drives, our Raptor can compete with them only under linear workload.
If we transform the results obtained into our rating (we consider all workloads as equally probable and average the Total I/O values), we will get the following picture:
According to this rating, the WD Raptor drive got a bit more points than RAID arrays (due to an unbelievably large gap between them under linear workload), but yields quite a bit to SCSI solutions.
Now we are going to take a look at WebServer patter, which is especially interesting because WD Raptor will lose its major trump here: efficient lazy write algorithms (WebServer doesn’t include and writes at all).
At first sight the performance of WD360GD changed only under linear workload. But if we compare its result with the performance shown by other testing participants, the picture will strike us a something really curious:
Look, how fast RAID 1 array has become here! It’s true that reads interleaving onto both drives of the mirrored pair can improve the performance rather significantly. Only under linear workload WD360GD leaves both RAID arrays far behind and competes with Seagate Cheetah 36ES SCSI drive, which boasts almost the same access time.
Now please check our rating:
Well, when WD360GD works only with read requests, it yields one step to RAID 1 array of two HDDs.
Well, the tests in server patterns revealed a very interesting thing: WD360GD hard disk drive is a worthy competitor to SCSI solutions only under linear workload. Keeping in mind that the disk subsystem of a small server is not too loaded, then why not take a Raptor drive? Besides, each HDD forming the array is usually loaded less than the entire array in general.
Well, in the server field WD360GD wasn’t quite comfortable under high workloads. Let’s see if it manages to stand out as a workstation solution.
And Raptor did it! :) Under linear workload it outperforms both SCSI drives. Although later the defeated competitors win their leadership back, it is exactly the performance under small workloads that matters for workstations. So, we calculate the performance rating accordingly:
Performance = 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
Note that WD360GD managed to be ahead of Cheetah 36ES with any controller card involved. And with Promise (WB) controller it even outpaced Fujitsu MAN. RAID 0 and RAID 1 arrays rolled back to the very end of the rating list as they have never been very fast under linear workload. However, they are anyway faster than the single WD400JB drive.
We will try to estimate the performance of WD360GD in WinBench99. This benchmark has been serving for test in “desktop” applications for years already. Of course, the applications as well as their files have changed a lot since the times this benchmark was introduced, but unfortunately, we do not have any alternative benchmark at our disposal.
Here are two linear read graphs for your reference:
As usual, let’s compare the HDDs in two integral tests: Business Disk WinMark and High-End Disk WinMark:
Wow, this is impressive! The performance growth is astonishing! If we compare the results shown by WD360GD and WD400JB (keeping in mind similar cache-buffer and close firmware), we will see that a shift to faster rotating spindle resulted into a 7% performance growth in Business Disk WinMark and in over 21% performance growth in High-End Disk WinMark. And since WD drives with 8MB cache buffer have always been very fast in these tests, we can conclude that WD made impressive progress.
Depending on the SATA controller used and the driver working mode, WD360GD results change, but they anyway remain very high. The maximum result in Business Disk WinMark is achieved with Promise (WB) controller and in High-End Disk WinMark – with Promise controller (disabled caching).
By the way, we can notice the performance difference between WD360GD with Promise (No) and Promise (WT) only in this test. In all other benchmarks the performance difference lies within the measuring error.
In NTFS the indisputable leadership belongs to Promise (WB). WD360GD proved the fastest with this particular controller in both: Business Disk WinMark and High-End Disk WinMark.
Well, no doubt that WD Raptor (WD360GD) hard disk drive looks like a very nice desktop solution.
In conclusion we would like to run a benchmark measuring how fast the files can be created, copied and read. This is our FC-Test (see this article for more details about this benchmark). The HDDs performance was measured with 5 preliminarily prepared file sets. The test names let you guess very easily which folder served as a prototype: Install, ISO, MP3, Programs and Windows.
Why do we use so many different files here? You may think it could be more than enough to take one big file or a few hundreds of small files to measure all the speeds. The matter is that we wanted to take the “real” HDD performance, that is how fast it works with the real and not synthetic files. Since hard disk drives from different manufacturers cope with files of different sizes not equally fast, we need quite a lot of files of the same size to be able to detect the HDD’s native inclination.
Since the results in this table are given in MB/sec, it should be easy to compare different HDDs with one another. And for a more illustrative analysis here are some diagrams:
As for SATA controllers working algorithms, the results obtained in this test session do repeat everything we saw during the tests of Seagate Barracuda SATA V. so, now we have every right to state that the controllers performance differently because of different driver algorithms.
This way Promise (WB) controller speeds up the writing quite noticeably, but doesn’t stand out at all during reading. It is remarkable, however, that it still manages to perform well in copy tests with smaller files.
SiliconImage controller is stably fast in all modes. It outperforms Promise (WT) and Promise (No).
As for the performance of our hero against the background of Fujitsu MAN SCSI drive or WD400JB, the new Raptor is always faster than the rivals.
The change of file system didn’t tell on the situation at all.
I believe that everyone or almost everyone have already put up with the fact that WD hard disk drives do not support temperature monitoring in SMART. However, I decided to check this feature by WD360GD just in case and was pleasantly surprised to find out that my favorite DTemp reported success.
Having started the “warm-up” test I started watching the temperature very closely. The monitor indicated 43 and this number remained unchanged for an hour. After that I checked the HDD all over with the infra-red thermometer trying to detect high temperature zones and here is what I got in the end:
If this value displayed in the SMART register is a real temperature and not a trick of WD software guys, then this spot should be located somewhere around HDD mechanism.
As for the noise WD Raptor produces, I also have some great news for you: Raptor is not the noisiest HDD I have ever heard. The spindle produces just a tiny bit more noise than any regular 7,200rpm HDD. The heads are traditional WD heads, which means that they are very quiet and non-irritating. The only moment when you may feel not very happy with the way it sounds is the motor start. As Raptor drive is ready to work after 7 seconds already, it is very active during these 7 seconds. But anyway, you will get used to this sound very quickly, this is not a GeForce FX :) I am saying this from my personal experience, as I installed a WD Raptor and a GeForce FX card in one PC case.
Well, the trip to 10,000rpm started quite successfully for WD. WD Raptor drive appeared a very exciting product. Fast mechanics and excellent lazy write algorithms allowed it to star quite a bit in our tests. Sometimes, it even defeated SCSI solutions, which is not a trifle.
However, the actual numbers obtained during the tests indicate that WD360GD features a very big piece of desktop genotype. No command tag queuing support and a Marvel converter will prevent it from competing successfully with 10,000rpm SCSI drives. WD360GD is very powerful under linear workload, but as the load grows up, the performance gap between the newcomer and the SCSI drives increases quite a lot.
As for Raptor’s future in the server market, it looks quite vague to me still. Until the server cases with native SATA drives support come (i.e. with the corresponding backplane or hotswap racks) I wouldn’t count on high sales. On the other hand, someone should be the first anyway...
An additional argument in favor of this SATA 10,000rpm drive could be its large storage capacity, such as 146GB, for instance. In this case the price difference between a SATA and a SCSI drive could be quite attractive for many users. But in the meanwhile there is only one Raptor: 36GB.
Another niche where WD wants to promote its solution is high-performance workstations. And we should admit that this field is a very good application for Raptor. High data density, fast response time and advanced lazy write algorithms make this HDD the only good choice for a high-performance workstation.
And it is also very remarkable that you will have to pay not too much for this speed, because if you were planning to get a SCSI, you would have to spend another $150 for a SCSI controller, while all contemporary mainboards feature an integrated SATA controller chip, which also supports RAID functions. Well, if I were you I would spend this $150 extra for another Raptor and enjoy the higher reliability (with RAID 1).