Hard drives, tests

The drives

For this survey we found six 250 GB drives, which use the Serial ATA 3.0 Gbits /s interface: the Hitachi T7K250, Maxtor DiamondMax 10, Samsung SpinPoint P120, Seagate Barracuda 7200.9 and the Western Digital Caviar WD2500JS and WD2500KS (WD Caviar SE and SE16, respectively). You will notice the presence of two Western drives. The first has 8 MB of cache and the second 16 MB. The latter size is also the one of the Maxtor DiamondMax 10. Except for the Western, all support NCQ.




We noticed that if Hitachi, Samsung and Seagate drives have two platters each with 125 GB of information (62.5 GB / face), with Maxtor there are three platters. For one only a single disc is used (100 GB/platter, 50 GB per side). With Western it’s even worse because they officially say that three platters are entirely used, which means 83 GB/platter and 43 GB/ side. These densities have a direct impact on the drive sequential transfer rate.






Manufacturers decided to use various controllers. There is Infineon for Hitachi, Agere for Maxtor and Seagate, and Marvell for Samsung and Western.

Different power connectors

Amongst the 6 drives tested, only the Hitachi and Western have standard Molex power connectors in addition of course to the Serial ATA power supply. The benefit of this type of input is its presence on many power supplies, even the oldest ones. If you don’t have a SATA power connector you will have to use an adaptor.


A word of warning; using this older type of connector isn’t without consequences. Only the new Serial ATA connector makes it possible to benefit from the improvement of this norm, the hot plug function.

Test procedure

For this test we measured several parameters. First we looked at the drives “synthetic” performances; cache and sequential transfer rates and average access time. Then there were more applicative tests, which include an applicative performance index based on PCMark05, a simulation of a file server type of load via IOMeter. And also a writing, reading and close copy (in the same partition) and far copy (starting at 50% of the drive) of a group of files.

These files include 2 big files for a total of 4.4 GB, plus 2620 files for a total of 2 GB and finally 16,046 files for a total of 733 MB. The source or target for reading or writing on the disc are two Raptor 74 GB in RAID, capable of 110 MB/s transfer rates without restrictions in this area. Of course, this is of little interest, because if the sequential transfer rate gives an idea of performances during the copying of big files, things are different with small ones.

All measurements were taken with sound management deactivated and activated (except for Seagate, because the function isn’t supported).

Another parameter that lead to several measurements was the chipset. We made others tests on the Intel i975X and NVIDIA nForce4.

With the i975X, we measured with and without AHCI mode activated. With standard parameters, Intel’s controller is configured in IDE mode and the SATA advantages such as NCQ or hot plug aren’t supported. It has to be configured in AHCI (Advanced Host Controller Interface) to benefit from these technologies. With the nForce 4, we tested with and without installing NVIDIA IDE drivers. They make it possible to benefit from the NCQ with the nForce among other things.

In addition to the six drives in this test, we added for comparison Western Digital’s latest Serial ATA race horse, the Raptor 150 GB WD1500ADFD.

Synthetic performances

Synthetic performances

Cache transfer rate measurements made with h2benchw and the « Core Test », show that the interface’s theoretical 300 MB /s are far from being used to their maximum, because the figure oscillates between 148 and 201.3 MB /s depending on the manufacturer. Hitachi displays the best performances whereas Samsung and Western drives are slightly slower. For the Raptor, we have to point out that unlike other drives it only uses Serial ATA 1.5 Gbits /s.


Sequential transfer rates are the first measured in the graph, which shows the evolution of transfer rates according to the advancement from the first to the 233rd Giga-byte. Samsung is the fastest for the whole drive followed by Maxtor and then Seagate. We noted, however, a difference of transfer rates between the two models, which leads us to believe that the platter may not be identical. This is in contradiction with Western’s announcement. Could the WD2500JS (SE, 8 MB cache) also have 100 GB platters? Lastly, we noted that Hitachi’s drive finishes in fourth position for the beginning of the drive and is last for the end.


This graph reports previous values with maximum, average and minimum transfer rates. Remarks are identical to the above. For comparison, we included the performances of the Raptor 150 GB. More than the maximum transfer rate, we appreciate the smaller drop as we get closer to the end of the drive.


Finally, access time is the best for Hitachi followed by Seagate and Western. Coming in last are Maxtor and Samsung. Synthetic performances, however, are not worth much if they aren’t validated by good application results.

Performance index

Performances index

We start with a performance index based on figures obtained with the « XP Startup », « Application Loading » and « General Usage » profiles of PC Mark 2005.


The first thing to notice is that all drives benefit from the change to AHCI or the installation of NVIDIA’s driver. These two parameters allow the activation of NCQ…even for Western drives that do not support this functionality! With Intel’s platform, the most important gains due to AHCI are noted with the Samsung, Maxtor and Seagate drives. With NVIDIA it’s mainly the Raptor 150 GB and Samsung. For Maxtor and Hitachi, however, they are just noticeable.

If you focus just on drive performances, the Western SE16 dominates followed by Hitachi. Seagate ends up last behind Samsung.

Perfs - file copy

Performances – file copying

The next test is file copying. We measured reading, writing, and also the copy of the following files on hard drives: 2 big files for a total of 4.4 GB, plus 2620 files which total 2 GB, and finally 16046 files which weighs 733 MB. The source or target for reading or writing on the disc are two Raptor 74 GB in RAID which are capable of ensuring a transfer rate of 110 MB/s without restrictions.

This type of information is uninteresting of course because if the sequential transfer rate gives an idea of performances during the copy of big files, things will be different with small files. We copied the files in two different ways: whiting the same partition in the beginning of the disc and from this partition to a second one that begins at the middle of the disc.


First surprise, for unknown reasons the activation of AHCI on the installation of NVIDIA drivers notably reduces some of the reading performances: it is the case of the Raptor 150 GB in AHCI and the Seagate and Maxtor drives with NVIDIA’s drivers. These two use Agere controllers. Only the Western SE/SE16 really benefit from the modification which is in fact the installation of NVIDIA’s drivers. Except for these cases we noted that the Maxtor and Samsung provide the best transfer rates here whereas Western and Seagate are the slowest.


For writing however, AHCI/Driver NVIDIA gains are significant except for Seagate and Hitachi’s drive with NVIDIA’s platform. We noted the excellent performances of Samsung’s drive and the opposite for the Maxtor, which was really good for reading. Fortunately, gains obtained by the AHCI and NVIDIA’s drivers for this disc compensate for this.


Whether it’s for close or far copy, one drive is here very surprising when used with an AHCI controller. The Samsung drive is much faster than a Raptor and 30% ahead of the closest 7200 Rpm competitor. We made several tests to confirm these results and we always obtained the same figures. The other surprise was the performance drop recorded on the Maxtor with NVIDIA’s drivers.
We also noted that a better access time doesn’t guaranty better performances for extended copies, because Hitachi’s performance reduction is more significant with a close copy than other discs. Except for Samsung and the Raptor, it’s the Western SE16 which is the fastest here.

Perfs – IOmeter per platform

Performances – IOmeter per platform

IOMeter is used to simulate a load in a multi-user environment. It’s a load type file server comprised of 80% reading and 20% writing, which is 100% randomly accessed on the disc. In this case, the NCQ can be particularly useful. We tested the IOMeter with a number of concurrent commands from 1 to 128 and of course for a single command, the NCQ doesn’t bring any improvements. The performance gap with the Raptor is much more significant, because this drive is particularly well suited to this type of use.

As the volume of data is rather significant we decided to represent it first per platform then below per drive with and without activation of the AAM on Intel’s platform.


For Intel and without AHCI, Western drives are ahead whatever the number of simultaneous accesses. Seagate is very close but loses ground beyond 16 accesses, even though the Hitachi and Samsung are relatively close. The Maxtor’s performances are clearly lower, however, despite a gap that tends to diminish as the load increases.


The activation of AHCI changes things. The Maxtor drive sees its performances increasing from 2 concurrent accesses to come in first (except of course for the Raptor). Maxtor can say thank you to NCQ. If Seagate also benefits from the NCQ, it isn’t really the case for Hitachi. Of course, it doesn’t change the Western SE/S16’s performances, because it simply doesn’t support this function. However, this doesn’t prevent them from reaching very high performances.


Without NVIDIA’s drivers, performances with NVIDIA’s platform are relatively similar to those obtained with Intel.


With NVIDIA’s drivers, the situation is different, because the NCQ really seems to work with 8 concurrent accesses and above. This is quite a lot and it doesn’t often happen in the case of a single user. Maxtor’s drive is strongly impacted even if it doesn’t stop it from being in the lead with 128 accesses (5% faster than with Intel’s platform).

Perfs – IOmeter per disc & AAM

Performances – IOMeter per drive & AAM

Here are now performances obtained per disc. In addition we added results obtained with AAM (Automatic Acoustic Management) activated on Intel’s platform.


For Hitachi, the NCQ doesn’t bring much whether it’s on an NVIDIA or Intel platform. We noted that with this type of access the AAM strongly reduces performances. The gap tends to reduce with the increase in number of simultaneous accesses.


We noticed with Maxtor’s drive that if NCQ based performance gains are visible from the start on the Intel platform, they only appear beyond 8 concurrent accesses on the nForce. It then immediately catches up and even results are even better. It’s rather unfortunate however, because in practice this type of drive will be of use for personal computers or with a limited amount of concurrent accesses. Here we can easily see the benefit of the NCQ to counter the negative effect of the AAM as from 4 simultaneous accesses the NCQ compensates completely.


Like with Samsung’s drive it is only from 16 concurrent accesses that the interest of NCQ is noticeable. Performances from this figure are much higher however than the one obtained with Intel’s platform. We noted that the activation of AAM leads to less significant performance reductions than with the Hitachi or Maxtor in this area.


With Seagate’s drive it isn’t possible to activate the AAM, which isn’t too critical. Once again, we have to wait until more than 8 accesses for the nForce NCQ to bring performance gains.


NCQ isn’t on the two Western drives and this is the reason why performances are similar between NVIDIA and Intel with or without AHCI / Driver. Once more we noted that even without NCQ, the gap due to the AAM tends to diminish with the increase of the number of concurrent accesses.


Here now is the Raptor 150 GB, which is in a totally different league except for NVIDIA’s NCQ performances.

Perfs – Perfs index & files with AAM

Performances – Applicative index & file copy with AAM

If performances with the IOMeter are strongly affected by AAM activation (because of the 100% random nature of disc accesses) we should take a look at the level of performances in less extreme situations.


As you can see, the performance drop is almost invisible on the Hitachi and isn’t huge on the Maxtor and Samsung. It’s greater for Western Digital. The Seagate 7200.10 doesn’t allow the user to modify the noise level whereas its adjustment with the Raptor doesn’t change anything.


We could have expected such a result. Whether it’s with reading or writing, performances don’t change much with or without noise management as heads don’t have to move a lot.


For copying the performance loss is notable for some of the discs such as the Western and slightly so with the Samsung. Hitachi and Maxtor’s performances remain almost unchanged. At times we even recorded a slight performance increase but this is most likely a testing error.

Noise - With and without AAM

Noise level with and without AAM

We saw in the previous pages that the discs that suffered the most from the activation of AAM were the Western SE and SE16. For Hitachi, the impact wasn’t very significant except with the IOmeter, whose best results were obtained with Samsung. What about simple access time?


Maxtor suffers the most from the activation of Automatic Acoustic Management with an additional time of 6.3ms. After that are Western and Hitachi at 5ms and Samsung at only 1.8 ms. This explains Samsung’s good results with AAM activated and with the IOmeter. However, we noted that despite the 20.9 ms access time with AAM, the impact on the DiamondMax 10 wasn’t the most significant with the activation of the AAM. In practice, performances were influenced by cache, NCQ or other optimisations included in the firmware, which more or less mask access latency.

A hard drive test with the AAM function would be incomplete without a measurement of noise levels here taken from a distance of 2 inches.


Just for spinning the most discreet drives are the Seagate followed by Hitachi and Samsung. Logically these are the 3 drives with two platters that are the least noisy. A step above is the Western and especially Maxtor. The Raptor gave us is a good surprise because its noise level is rather low despite the 10,000 Rpm.

With intensive disc accesses the Seagate is no longer discreet and its noise level is much higher. It is all the more unfortunate as it doesn’t have an option to change it. This is the same for the Raptor but because of its performances it can be excused. Seagate is followed by Maxtor, then Western and finally Hitachi. In this world of intensive accesses Samsung’s performances are from another planet and it has a rather discreet seek. Its access time is not as good compared to other drivers but is at the same level as the Maxtor, which is much noisier.

Noise level management gives some relief to silent computer fans. The gap isn’t that significant for Samsung, and this time Maxtor ends up second. Access time is increased to 20.9 ms! As the noise level is rather high in spinning, we prefer a more discreet drive in this domain even if it is less so in accesses.

Conclusion

Conclusion

The choice of a drive won’t be made only on performances. You also have to take into consideration the noise level and price, which is usually rather unstable (we won’t look for the reasons here). Warranty is also an important factor and we point out that only Seagate is guaranteed for five years compared to three for the others. And this is the only reason why you would choose the Seagate compared to other models of this survey. The warranty of our Seagate drive expires (according to the manufacturer) the 8th February 2011.

In the case of a “silent” computer, there is no real alternate choice to the Samsung SpinPoint P120. Discreet while it spins, it is mainly characterised by a much lower noise level than the others for access. Of course, the counterpart is that it isn’t the first in performances except for writing, and most of all intra-disc file copy in AHCI mode.


For performances, except for the Raptor which doesn’t belong to this competition, our preference goes to a drive that doesn’t support BCQ. It compensates for this absence by performances that are already very high without this function (see with the IOmeter) and this is of course the Western Digital Caviar SE 16 250 Go (WD2500KS).

Hitachi’s’ drive is good but doesn’t set itself apart from the others in a specific area, other wise we could have recommended it for a specific use. It is however a very good compromise. Unlike the Western drives, there is NCQ for Hitachi but its efficiency is very limited. The situation is the opposite for Maxtor, because it is with the Diamond Max 10 250 GB that the NCQ brings the most significant gains. This mode only compensates for performances that are much lower in certain cases. If the only advantage of the Seagate drive is the warranty, for Maxtor we will only point out that disc space after formatting is 233.8 GB as compared to 232.9 for the others.