20 Modern Series — SAS, SCSI, Serial ATA
Power consumption and heat dissipation of modern hard disk drives, which usually have a much narrower range of operating temperatures (from +5°C to +55°C, more rarely from 0°C to +60°C) than the majority of other computer components, are one of the problems that often draws attention of users. HDD performance grows as well as CPU clocks or video card frequencies. But fortunately, there is no rapid growth of heat dissipation here (as performance increases), which can be seen in CPUs and GPUs for the last ten years. Nevertheless, general requirements to power saving and limited power and cooling capacities of certain chassis models often make users wonder how much power their hard disk drives consume. Such questions are asked not only by users (and manufacturers) of notebooks, where each half Watt can influence not only HDD temperature in a thin and badly ventilated space, but also battery life of the notebook (usually the ultimate objective). Besides, the same issue concerns users and assemblers of desktop PCs, where hard disk drives have to content with crumbs of power due to take-off in power consumption of processors and video cards.
The problems of power consumption and heat dissipation become more pressing for those who work with high-performance professional data storage systems based on hard disk drives (so-called Enterprise segment). In addition, reliability and durability of these drives depends much on their operating temperatures. According to our research, increasing HDD temperature by 5°C has the same effect on reliability as switching from 10% to 100% HDD workload! Each one-degree drop of HDD temperature is equivalent to a 10% increase of HDD service life. Powerful cooling systems are not always justified as they are noisy and expensive. On the whole, economy is the factor not to be overlooked when you make a choice. That's why our next attempt to dwell on practical aspects of HDD power consumption and heat dissipation is not only cognitive but also of a purely applied character.
We have already analyzed power consumption and heat dissipation issues of 3-inch desktop hard disk drives and of high-performance 2-inch notebook hard disk drives. And we shall return to this topic many times. But today it's time to speak of the most expensive and failure-critical (including overheating and power-shortage problems) Enterprise hard disk drives, which include 3.5" and 2.5" 10K and 15K hard disk drives with Ultra320 SCSI and Serial Attached SCSI (SAS) interfaces (we'll skip Fibre Channel for now). As well as certain professional 7K2 models with Serial ATA interface (later on SATA 2.5) and of high capacity (400-500 GB, not yet available to SCSI models), based on existing desktop hard drives from the same manufacturers, but with slightly overhauled design and firmware in order to increase reliability and improve performance in professional tasks. To the latter (that is to professional Serial ATA 7K2 hard disk drives) we refer traditional Maxtor series MaXLine III and MaXLine Pro 500 (as well as the older MaXLine II), the recently appeared Seagate NL35 (professional counterpart of top Barracuda 7200.8 and 7200.9 models), as well as Western Digital Caviar RE and RE2 (in particular, the recent 400-GB WD4000YR model). Unfortunately, Hitachi GST does not single out its Deskstar 7K400 and 7K500 models (400 GB and 500 GB correspondingly) into a professional series, though lots of their properties qualify them for this title. That's why they will be included into this review together with the above-mentioned 7K2 models and all current SCSI series, reviewed for example in our recent article. Besides, one of our contenders is the first (that actually appeared in Russia) hard drive with SAS interface — Seagate Cheetah 15K.4 SAS.
Detailed substantiation of our approach to analyzing power consumption and heat dissipation of hard disk drives (and the reasons why they are practically the same in power units) can be found in our previous review. So we proceed straight to figures. I will just remind you that we'll deliberately avoid using HDD temperature as a measure of their heat dissipation, as in our opinion it's useless in typical cases (read the substantiation at the link above). Besides, we get some additional information by measuring power consumption (instead of temperature).
Power Consumption Specifications
As a reference, Table 1 contains power consumption data for the main professional HDD series, provided in their specifications.
Table 1. Power consumption (Watts) of hard disk drives for professional applications (according to specifications).
Unfortunately, in most cases specifications provide very little information, which is not enough to get a complete view of the real situation. Sometimes manufacturers specify only upper limits, sometimes – typical values, sometimes these figures have nothing to do with reality, if you compare them with the readings taken from these drives. Nevertheless, specifications exist and we have to take them into account.
By the way, it would be a mistake to judge about power consumption of hard disk drives by the figures on their cases. Having compared these inscriptions with real figures below, we found out that they do not always match. Moreover, these parameters often mismatch even the specifications on these drives. It's often not so easy to understand the principles, which manufacturers follow to mark technical parameters of drives on their cases.
Contenders and Test Methods
Our today's contenders are 21 models of modern hard drives - 20 state-of-the-art series of professional storage drives. These hard disk drives are listed in Table 3. RAID with two Seagate Savvio 10K.1 drives comes in a separate line on the diagrams. A couple of these drives in RAID can compete with high-performance 15K models. (read our review).
We used the following testbed configuration to measure power consumption of hard disks:
We measured the power consumption of hard drives in various modes: Idle, ATA or SCSI Bus Transfer, Read, Write, and Seek (additionally in Quiet Seek mode, if supported by the storage drive; we do not publish these results here, as they are irrelevant). Besides, we measured maximum start-up currents. A package of these parameters renders the situation with HDD heating (a product of current and voltage gives the heat rate dissipated by a drive) as well as with its economy in the most complete way. Operating modes of a hard drive were controlled by the corresponding tests in AIDA 32Disk Benchmark, read and write modes were measured "in the beginning" of a disk (on the most frequently used outer tracks; power consumption on inner tracks is usually lower). The tests were carried out under MS Windows XP Professional SP2. We tested non-partitioned hard drives. Before the tests, we warmed the hard disks for 20 minutes using a utility with active random access.
We measured the +5 V and +12 V draw (accurate voltages at the output of the above mentioned unit were +5.02 V and +12.04 V) simultaneously with two digital ammeters of the 1.5 accuracy class with the resistance below 0.15 ohm (including the leads' resistance). The refresh rate of readings was approximately 0.3-0.4 sec. The table provides average values for several seconds (current fluctuations usually didn't exceed 30 mA), except for the Start-Up current - we published its maximum values (averaged for about 0.3 s).
Our readings are published in Table 2. The last column contains the data specified on a case of a hard drive.
Table 2. Current drain (mA) of hard drives in various modes.
The table holds a lot of numbers and there seems no point in commenting them all — they are self explanatory. But it should be noted that the +5 V current drain of the Seagate Cheetah 15K.4 SAS turned out a tad higher than in its SCSI counterpart — by 200-470 mA depending on an operating mode! We have already made sure that the current drain in Serial ATA models is higher by 100-250 mA compared to their UltraATA counterparts (depending on a manufacturer and a model). In case of 3 Gb/s Serial ATA, the gain is even higher. Serial Attached SCSI itself consumes much on the HDD side — additional 1-2.4 Watts (probably the same amount is consumed on the side of the controller). Progress comes at a cost even here.
The following diagram contains power consumption of hard drives at the +5V bus when transferring data via the interface (without accessing the platters) (that is the power consumed mostly by the controller, while HDD mechanics is powered by the +12V line). The hard disks are grouped by categories (by the spindle rotational speed).
Host Transfer Drive Power Consumption, Supply Voltage +5 V
Interestingly, the most economical SCSI controller is installed in 15K Hitachi drives (you may remember that they are based on a more progressive design than their 10,000 rpm counterparts). However, controllers of the Savvio 10K.1 (and Cheetahs in general) also consume very little. By the way, the power consumption of 4-5 Watts for data transfer via the interface (that is without reading/writing) is quite a high result, that is controller boards on their own (not equipped with heatsinks) can grow quite hot during operation. That's why I don't recommend installing professional hard disk drives close to each other, you had better fan their electronics (for example in a cage).
The most economical controller among all 10,000 rpm models is installed in the WD Raptor with Serial ATA interface. Strange as it may seem, it outperforms (nearly) all modern professional hard disk drives with the spindle rotational speed of 7200 rpm. The most economical controller appeared to be in 400 GB Seagate NL35 drives, based on Seagate Barracuda 7200.8. This couple (their controllers) consumes just 2.5 Watts each when transferring data along the interface.
Speaking of how the specifications stand to our measurements, the picture is rather odd. Some figures are similar, the others are noticeably different (it's more convenient to compare Table 1 with Table 3 below).
Now what concerns the correlation between the power consumption data specified on HDD cases and our readings in various modes. A perfect Babel! Frankly speaking, I will not even try to guess what each manufacturer meant by the figures on its HDD case. You may try to guess on your own. :) I can only assume that the values published on some HDD cases correspond to maximum current in idle mode (quite nominally at that). To put it simply, you cannot trust the figures marked on hard disks. In fact they are useless and even harmful as they misinform users. Moreover, you cannot use them to judge about the real heat dissipation of storage devices!
Start-up current of hard drives is a separate issue. While it falls within 1A along the +5 V line, the main load certainly falls on the +12V line, where peak currents (average for tenths of a second) reach 1.5-2 A or sometimes higher. It's OK if your controller and hard disks support staggered spinup. If not (all hard drives in a RAD start up almost simultaneously), PSU load is very high (for example, 4-drive RAID with top SATA drives from Seagate consumes up to 9-10A at start-up)!
Start-up power consumption of hard disk drives of three professional categories with different spindle rotational speed.
The most "humane" (for power supply units at startup) models are Hitachi Deskstar 7K400 drives and the tiny Seagate Savvio 10K.1 (start-up current does not exceed 1200 mA at the +12V line), as well as Maxtor MaXLine III (1400 mA) and single-platter Seagate Cheetah 10K.7 and 15K.4 SAS (the 5V load actually lets the latter down). The most voracious hard disks at startup are traditionally SATA HDDs from Seagate — Barracuda 7200.8 and 7200.9, as well as their successor NL35: 30 Watts and higher at start-up (and 2.3 A at the +12V line) — only two Savvio 10K.1 drives are up to it. :) About two amperes (over 28 Watts) are consumed at spinup by SCSI drives from Fujitsu (mid-capacity models from these series), as well as two Maxtor models of the highest capacity — 300 GB SCSI Atlas 10K V and 500 GB MaXLine Pro 500 (represented by DiamondMax 11 :)). Besides, 400 GB models from WD also cannot boast of mild start-up (unlike WD Raptor). These characteristics are very easy to explain: start-up current in some SCSI and ATA drives is spread over quite a long period of time (start-up current is limited by the electronics of a drive to a certain level during spin-up). But other models are designed for fast spinup, so their start-up graphs resemble an abrupt surge with a falling slope rather than a long plateau.
Heat dissipation of the drives
Current drains (especially on both power supply lines) are not very illustrative as far as heat dissipation is concerned. So we shall use them to calculate power consumption for each operating mode (see Table 3). Of course, the power in this case is calculated with regard to a voltage drop on the internal resistance of ammeters in power supply lines, that is it corresponds to this very case. The power may be slightly different with other voltages.
Table 3. Power consumption and heat dissipation (W) of hard drives in various modes.
Some data are more illustrative on diagrams. For example, the next picture shows power consumption of hard drives in idle mode (no data access and transfer, though platters are rotating).
Hard Drive Typical Idle Power.
Huge spread of values for 15K models is very interesting: the most economic model here is Fujitsu (average capacity) - it's nearly twice as better as the top Maxtor model. Representative of the latter is the hottest among 10K drives as well — though its capacity is higher than in the other models. By the way, 73 GB and 147 GB models of the Seagate Cheetah 10K.7 differ in power consumption by just 0.9 Watt. They are both the most economic SCSI drives with the spindle rotational speed of 10,000 rpm. (they are really noticeably cooler than the others). 7K2 drives are much less different in this parameter than SCS models — from 7.4 Watts for Seagate NL35 (exactly as in its specifications!) to 8.9 Watts for the 500 GB Barracuda 7200.9 (specifications probably lie here. :)). On the whole, the typical power consumption for such drives in idle mode is 8 Watts. But nuances (that is currents at the 5 and 12V buses) may differ noticeably from model to model (as a rule, 400-500 mA at +12V bus and 480-630 mA at +5V bus). The winner in power saving is the 2.5-inch SCSI Seagate Savvio 10K.1 — 4.5 Watts in idle mode are a very good result. That is even a couple of these drives dissipates less heat than the majority of 15K models and some 10K ones.
Economic Savvio takes the lead in active random seek mode as well. Here is the layout of forces in power consumption and heat dissipation:
Hard Drive Average Seek Power.
Top-capacity SCSI drives from Maxtor are the most voracious here (about 20 Watts — that's not a joke, they badly need active cooling). They can be justified only by the fact that they offer the highest performance in most profile applications. However, top Seagate Cheetah 15K.4 is also one of the hottest models in seek mode, unlike the junior SAS model of the same series. The most economical models in seek mode are SCSI drives from Hitachi GST. SCSI drives from Fujitsu again back up our rating that these are currently the best hard disk drives for PCs(!), because along with the best performance in desktop applications, they offer the least power consumption (in their classes). :) By the way, the spread of power consumption in seek mode among SCSI models is over 1.5 times in each of the classes. But in case of SATA drives with the spindle rotational speed of 7200 rpm, the spread is a tad smaller, most drives fall within the range 11-14 Watts. Interestingly, the most economical models here are Seagate Barracuda 7200.8 drives, while the 7200.9 model with more platters cannot boast of low heating. However, even this drive is less voracious in seek mode than the five-platter Hitachi 7K400 and top SATA models from Maxtor.
In order to reduce the figures from Table 3 to common simpler and more intelligible denominator, we calculated two parameters, useful in practice: mean power consumption of hard disk drives under low load and during intensive (constant) HDD operations. I used two characteristic HDD profiles to calculate these parameters, though they do not pretend to being completely credible:
1. Model of the average hard disk power consumption for low HDD loads (for example, office work or editing graphics) can be described by the following formula:
P typ =(Idle *90%+ Write *2.5%+ Read *7.5%)/100%,
where lettered modes denote the power consumption of a drive from both voltage sources in the corresponding modes; digits (multipliers for these power values) denote percentage of the HDD mode duration (we take maximum power consumption values for reading and writing, which correspond to the beginning zones of a disk; Seek mode is actually metered here through reading and writing). This model is based on the assumption that read/write HDD operations make up 10% of the total time for the typical desktop usage.
2. In the same way, the average power consumption during intensive HDD operations (much more typical of a usual profile for professional hard drives) can be defined by the following formula:
P max =(Write + Seek + Read *3)/5
Calculated power consumptions are used to plot the following two diagrams.
Average power consumption of hard disk drives under low load (typical PC operation).
These results are obviously close to the layout of forces in Idle mode — the most economical models consume just 7-8 Watts, the coolest drives being Seagate and WD models as well as the 15K Fujitsu drive. Of course, Savvio 10K.1 is way more economical than the other contenders.
The mean power consumption of hard disks under intensive (constant) load is shown below.
Intensive Drive Use Power Consumption.
Disposition of our contenders in this case is close to that in active seek mode, but the average consumption is lower by several Watts. Interestingly, two Savvio 10K.1 drives consume as much as one 15K or even 10K model. It makes them real competitors (in performance as well) and even gives some advantage in the rating. Average power consumption of top SATA drives in active mode is currently 10-12 Watts, which is a tad less than 1.5 times as high as in idle mode. They do not require active cooling, if they are installed on a metal chassis. Unlike top SCSI models, which must have active air cooling during intensive operations. But not for all. For example, junior Seagate Cheetahs may do with passive cooling in some cases, when installed on a chassis.
Power consumption versus random access block size
A question arose in the process of experimenting with power consumption of hard drives — what's the relation of power consumption versus a random access block size? The fact is that the current drain at the +5V bus for reading and writing is noticeably higher than in seek mode (see Table 2). On the contrary, the current at the +12V bus is higher in seek mode. Seek mode is much more exacting to cooling, according to the average heat dissipation (see Table 3). At the same time, as the random access block size grows, the current drain gradually increases at the +5V bus and decreases at the +12V bus (as the time of moving between large blocks grows due to additional time expenses for reading and writing these blocks). There may be situations when monotony of the average power consumption relation is broken - power consumption for certain read or write blocks either gets maximal or on the contrary reaches some minimum.
So in search of such extremums we analyzed the relation of power consumption versus a random access block size for reading and writing. We've taken two latest hard disk drives from the opposite Enterprise segments: Seagate Cheetah 15K.4 36 GB with SAS interface and WD Caviar RE2 7K.2 400 GB with Serial ATA interface. By the way, the average power consumption of the spindle engine at the +12V bus is absolutely the same in these two drives in sequential read/write mode, though Cheetah's controller is nearly twice as voracious.
The average power consumption was measured depending on a random read or write block size from 512 bytes to 1 Mbytes during execution of the corresponding IOMeter patterns. At the same time we were registering changes in performance of these hard disk drives — average access time, IOps, and absolute read/write performance (MB/s). The results are published in the tables below and illustrated on summary diagrams.
The current at the +5V bus for random reading in Seagate Cheetah 15K.4 SAS gradually grows nearly by 20%, as the block size increases from 512 bytes to 1 MB, while the current at the +12V line drops from 870 mA to 540 mA. Full power consumption also drops, but this drop is smaller percentagewise — from 15.7 Watts to 12.5 Watts. The graph of power versus block size shows two extremums: one minimum at 64 KB blocks, followed by maximum at 128 KB blocks. That's exactly what we wrote about in the preamble of these experiments. At the same time we can see that IOps performance is strikingly similar to the situation with power consumption! Thus, as far as performance is concerned, operations with 64 KB blocks are far from optimal for this hard disk drive (together with the Adaptec 4800SAS controller). 32 KB or 128 KB are better, though power consumption with such blocks will be a tad higher.
A similar extremum for 64 KB blocks can be seen for this hard disk drive for writing as well: we can see well the power consumption curve dip, the performance in this point falls out of the general tendency for the worse. On the whole, the dependence of power consumption on the block size for writing is significantly lower than for reading (power consumption itself is lower) due to effective write-back. Write operations demonstrate another tendency — power consumption is really lower at 32-128KB blocks and it again grows as the block size increases to 1 MB.
WD4000YR tendencies are a tad different. The absolute maximum power consumption for random reading is demonstrated at 16KB blocks (though we can see no performance anomalies). That is in this case it's the contribution of some increase in the average read current, while seek deceleration due to the block size growth has no effect yet. The situation with writing is still more unusual. Minimum power consumption is demonstrated at mid-size blocks (pronounced minimums are at 8KB and 64KB; or in other words, maximum power consumption is again demonstrated at 16KB (as in case of reading) against a background of general reduction in power consumption at 8-128 KB blocks). Power consumption greatly increases at 256KB blocks and larger, even though the access rate drops drastically. It has to do with a considerable seek deceleration, while the write current is much higher.
Analyzing power consumption and heat dissipation of hard disk drives by measuring currents at +5V and +12V buses in various operating modes provides additional information on HDD properties, which will be useful for general understanding as well as for actual practice. In conclusion of our analysis of power consumption in representatives of practically all series of modern Enterprise-class hard disk drives we should say that...
1. Specifications on power consumption of a hard disk or, moreover, this data printed on an HDD case should be taken critically. They will seldom give you an idea of the true power consumption and heat dissipation of hard disks! We can rely on our readings in real tasks.
2. There are three main categories of hard disk drives in the enterprise segment, which differ in their spindle rotational speed (15000, 10000 and 7200 rpm). Although power consumption and heat dissipation depend on the spindle rotational speed, this effect is not as pronounced as, for example, their relation to the average access time or some other performance parameters. It turns out that even a number of the fastest models may consume power on the level of slower hard disk drives and vice versa. Sometimes, power consumption of models in the same category may differ by 1.5 and more times.
3. Of course, it must be taken into account when designing professional data storage systems — hard disks from one manufacturer may do well with passive cooling, while their counterparts from the other manufacturer may require active cooling, and vice versa.
4. Moreover, some modern high-performance SCSI drives feature humane heat dissipation, competing here even with desktop SATA drives. The miniature Seagate Savvio 10K.1 is the most economical model among high-performance hard disk drives, having outscored all 3.5-inch ATA drives, even though it offers a very decent performance level.
5. In some cases you should pay special attention to providing proper load that does not exceed PSU capacities, when hard disks spin up — it even concerns some modern SATA models (for example, from Seagate), especially hard disk arrays. It turns out that maximum start-up current practically does not correlate with spindle rotational speed, as manufacturers sometimes use special limiting circuits. Some SATA 7K.2 models may be twice as voracious to spin-up current than their faster SCSI counterparts from the same manufacturer.
6. The appearance of Serial Attached SCSI (SAS) brings power consumption of controllers at the +5V bus to a new level. Unfortunately, these changes are not for the better, because hard disk drives with SAS interface consume at the +5V bus much more power (from 1 to 3 Watts) than their counterparts with the traditional parallel SCSI interface. In fact that's the cost for just several 3 Gbd buffers (transceivers) of the serial interface. The situation will obviously deteriorate as interface transfer rates grow in future generations of this interface, because reading and writing will also require more power (due to the increase in data transfer rates from platters). But the power consumption of mechanics at the +12V bus will most likely drop a little or stay on the same level.
7. In these conditions we are interested in the relation of power consumption versus data access method, random access block size in particular, which is more typical of professional HDD profile. As a block size grows, we can see a power consumption extremum, either maximum or minimum. You should take it into account if you use such hard disks and choose optimal operating conditions for data storage systems considering performance and heat dissipation.
And in conclusion I'd like to note that we have accumulated great experience in evaluating power consumption and heat dissipation of hard disk drives from various segments. While such issues were defined in our procedure for testing notebook hard disk drives long ago, they are still overlooked in our analysis of new products from the Desktop and Enterprise segments. But in our future (updated) test procedure for such hard disk drives we shall use power consumption measurements by the fine-tuned or a tad improved algorithm.
We'd like to thank Maxtor and Seagate
for the provided hard disk drives
Alex Karabuto (firstname.lastname@example.org)
March 2, 2006
Write a comment below. No registration needed!