Sign in with
Sign up | Sign in

How Can Battery Runtime Be Shorter?

The SSD Power Consumption Hoax
By

It’s important to answer this question, because it doesn’t seem like this should be the case. Most flash SSDs show power requirements in idle and when under load that are comparable to those of conventional 2.5” notebook hard drives. A typical 2.5” hard drive based on rotating magnetic platters usually requires between 0.5 W and 1.3 W when it runs idle, and from 2 W to around 4 W when it is under maximum load. Peak load occurs when the actuator has to move the heads back and forth on the disc surface due to lots of random accesses.

Let me emphasize a part of my last sentence: “when the actuator has to move the heads.” It’s important to understand that a conventional hard drive will only reach its maximum power requirement when you’re requesting random data that’s distributed all over the medium. In the case of sequential reading or writing, hard drives will not require much more than the idle power, as there is no energy-intensive acceleration and deceleration of the actuator.

Flash SSDs Only Know Two Power States

In contrast, flash SSDs only seem to know two states: active or idle. We don’t have specific information on this, but we received confirmation from two vendors that many flash devices don’t feature power saving mechanisms yet. On the one hand, the entire industry is looking to improve MLC flash to enable larger flash SSDs at sufficient performance levels. On the other hand, wear leveling algorithms are more important than power saving features, as durability may be an issue with SSD drives.

So while conventional hard drives may operate at relatively low power when little movement is required — such as during sequential read access — flash based drives do not. They will draw their maximum power level constantly when in use, and as a consequence, simply spend more total time drawing maximum power than conventional drives.

Display all 76 comments.
This thread is closed for comments
Top Comments
  • 17 Hide
    durdenet , June 27, 2008 3:14 PM
    Wow, I've always respected Tom's hardware as the most reliable company on tests... well not anymore after this article...

    Two of the BIG mistakes have already been mentioned in this comment page:
    1- "additional tasks you would not have normally done on a laptop (media processing, HD video, etc...)" and the possibility that "Windows doing more background stuff (desktop search, adware/av, etc...) because it doesn't take as long to finish maintenance tasks." would DEFINITELY influence consumption of a whole system but it's not even considered.
    2. "The SanDisk SDD drive at LOAD requires 1.0 mW while Hitachi HDD requires 1.1 mW at IDLE". So clearly, either the graph in "power consumption" is WRONG or the whole article is wrong. So I ask the editors to either take off the article or correct the graph.

    While they correct the errors of this article, do not believe the conclusions of this article. SSD do consume less for any X amount of tasks on a Y amount of time, and that is the only factor that counts. full stop.

    This article measure Y amount of time with a different amount of tasks...... very clever..... :-s
  • 14 Hide
    mastrom , June 27, 2008 11:13 AM
    I agree with Fritz, I think it's very important to make sure the work the laptop does with SSD and HDD is the same. And I'm not sure having a laptop run a benchmark until it runs out of battery comes close to real world usage. Playing a movie again and again would probably be a simple and clean test that is not questionable. At the moment I'm not convinced that your conclusions are correct.
  • 11 Hide
    Anonymous , June 27, 2008 10:34 AM
    There could be a systematic error in the benchmarks shown: if the flash based "disks" are faster then the whole system CPU/MEM/Chipset would draw much more power with flash "disks" compared with conventional disks - just because the benchmarks could run more often in the same time.

    Maybe one should compare something like playing video from disk where it is assured that the systems do precisely the same work?
Other Comments
  • 0 Hide
    Anonymous , June 27, 2008 5:49 AM
    This is interesting, but makes sense. So many power saving features have been implemented into HHDs, but SSDs are very young. I'm sure that at least some companies understand and recognize this, and are developing new features to help with this.

    Technically, it may not be that the claim on "energy efficiency" is indeed true. the efficiency could be measured in mW spent per Kb of information retrieved. For example, a HHD in sequential reading uses up very little power, but also retrieves only the information needed. It could be that the SSD's have to check many more bits of data in order to find the information requested. This is just a theory.

    I believe the salespeople took a logical step in claiming energy improvements, due to the fact that it uses no visible motors. However, this testing (at least in my eyes) shows that they have simply "forgot" to check up on the actual statistics.

    It could also be that as cell density increases, read and write speed will increase as the power consumption stays almost static. Since the 1.8' SSD eat up almost the same number of watts for the larger SSDs, it could be said that the power itself is not being consumed in the actually memory itself, but in reality the controller that access the information. IT is very very interesting to see the small difference between the different sizes of SSDs, and might be a window into seeing the full potential of these devices.
  • 9 Hide
    Anonymous , June 27, 2008 7:46 AM
    Strange picture comes from this article.
    The overview seems to be reasonable but comparing to the benchmark results it looks quite different.
    Just look at the Battery Runtime and Energy Consumption charts!.
    The SanDisk SDD drive at LOAD requires 1.0 mW while Hitachi HDD requires 1.1 mW at IDLE (for me it means in POWER SAVING MODE).
    So if laptop work time with Hitachi HDD is longer than with SanDisk SSD the benchmarks seems to be VERY unreliable.
  • 0 Hide
    mitch074 , June 27, 2008 8:10 AM
    Some improvements could be done to "disks" themselves (although can we still talk about disks in that case?), but I think there would be mquite a lot to gain with a software solution - like was done with Native Command Queueing on SATA disks.

    For example, better working read and write caching would be a boon: since SSDs have pretty much no access time penalty, a large write cache becomes much less interesting - grouped writes, optimized for the SSD's internal cache, may reduce the time the system stays on. Deactivating the swap file (or not making use of it as often as Windows does) would also probably reduce 'active' time quite a lot.

    It would be interesting to see the difference when using a more disk efficient OS: I dunno, any Linux-based distribution? After all, Dell provides Ubuntu on some laptops and UMPCs are currently often provided with SSDs - so it would actually be quite relevant.

    Why the Linux comment? Typically on these systems, the swap gets written to only when the RAM reaches around 60% (or more) use, while Windows preemptively copies 'dirty' RAM pages to swap and frees RAM only when actually needed (leading to faster RAM pages freeing, but also frequently unnecessary disk access).

    Running these tests again (with all disks and some extra RAM) with:
    - Windows using no page file
    - Linux with a swap partition
    - Linux with deactivated swap
    may give us a better idea of SSD advantages.

    About those who would complain that relying on RAM instead of swap is not possible:
    - current systems (except lower Windows editions) can handle more than 4 Gb of RAM; Windows 2003 for example, does support PAE and extended memory sizes (64 Gb)
    - if you have cash to pay for an SSD, you have cash to pay for 4 or 8 Gb of RAM

    Of course, that doesn't prevent SSD makers from looking for extended power schemes (reads should require less power than writes, and idle should require no power) and a low power SATA mode (allowed by shorter cabling) could be designed.

    Mitch
  • -1 Hide
    randomizer , June 27, 2008 10:07 AM
    Man, my laptop would be lucky to last 1 hour :lol: 
  • 3 Hide
    skittle , June 27, 2008 10:25 AM
    This SHOULD have been included in the article... since its the basis of of your claims. Copied from MobileMark website:

    Quote:
    MobileMark 2007 incorporates the following applications:

    * Adobe ® AcrobatReader 7.0
    * Adobe® Illustrator® CS2
    * Adobe® Photoshop® CS2
    * Apple® Quicktime 7.1
    * Intervideo® WinDVD® 8
    * Macromedia® Flash 8
    * Microsoft® Office® 2003 Pro
    * Microsoft® Project 2003
    * Winzip® 10.0


    Quote:
    Recognizing the increasing diversity in the use of notebook computing, MobileMark 2007 features DVD, Wireless and Reader modules in addition to the existing core Productivity module. Each module includes a robust and refreshed set of applications and can be run individually to show battery life in a greater variety of specific scenarios.
  • 11 Hide
    Anonymous , June 27, 2008 10:34 AM
    There could be a systematic error in the benchmarks shown: if the flash based "disks" are faster then the whole system CPU/MEM/Chipset would draw much more power with flash "disks" compared with conventional disks - just because the benchmarks could run more often in the same time.

    Maybe one should compare something like playing video from disk where it is assured that the systems do precisely the same work?
  • 2 Hide
    JonathanDeane , June 27, 2008 10:36 AM
    I am thinking the idle power of an SSD could in theory be zero since it is a flash drive and there is no "spin up" this is interesting... seems the drive electronics (not the actual flash chips) are eating allot of juice.

    An OS patch with an option to put the drive to sleep when inactive for something like 1 second seems like it would save power and since the power on should be instant have almost no negative impact.
  • 14 Hide
    mastrom , June 27, 2008 11:13 AM
    I agree with Fritz, I think it's very important to make sure the work the laptop does with SSD and HDD is the same. And I'm not sure having a laptop run a benchmark until it runs out of battery comes close to real world usage. Playing a movie again and again would probably be a simple and clean test that is not questionable. At the moment I'm not convinced that your conclusions are correct.
  • 0 Hide
    Anonymous , June 27, 2008 11:27 AM
    What about just turning on a laptop and leaving it running word and outlook and browsing the web for the complete battery life, now thats a typical session...
  • -4 Hide
    pschmid , June 27, 2008 11:41 AM
    As I pointed out in the conclusion, an application scenario that does not tax the drive very much can very well lead to increased battery runtime. But does this reflect user behavior?

    Think this way: You hibernate the system, because you have to move location. You start applications, you close them. Anti virus software does its job. The swap file causes frequent access. Indexing triggers more drive activity. A p2p client causes permanent access as well... the more we multitask, the more drive access we cause. This leads to power disadvantages for Flash SSDs; at least for the current product generation.

    I decided to run the tests because I had changed from a Seagate Momentus 7200.1 hard drive to the Mtron Flash SSD drive half a year ago - and I saw a decrease in battery runtime of my Lenovo T60 that could not be attributed to wear of the battery. Hence we followed up and found this.

    If you want to send me more specific suggestions on additional tests, such as specific applications, I'll be happy to run them when we compare the next Flash SSDs. Email pschmid@bestofmedia.com.

    Thanks for your comments,
    Patrick
  • -2 Hide
    okayman , June 27, 2008 12:25 PM
    In reply to neg@infostrade.com.pl:

    That is the trick in marketing SSDs. If you plainly compare IDLE and LOAD wattage, yes, SSDs SHOULD be using less energy than conventional platter drives. But what you have forgotten is the read/write performance. You have to count that in when measuring power consumption. For example, if the SANDISK SSD takes 2 seconds instead of 1 second for TravelStar to read or write the same data, tell me, which one would be in LOAD state longer? Lets say if a noname SSD takes 1 second to write 1MB and a noname platter HDD takes 1 second to write 10MB. The LOAD wattage of the SSD has to be 1W and HDD to be 10W to make their power consumption equal given the same amount of data to be written! If you only look at the Wattage, it doesn't give you any result at all.

    This is what we call "weighted".

    I'm also eager to buy an SSD, not because of power consumption issues, but instead is because it seemed to be less prone to shock than conventional HDD. I hate sudden deaths of HDD without much warning even sometimes you think you keep it working in good environment. SSD has a "life expectancy" you can predict of, at least.
  • 2 Hide
    okayman , June 27, 2008 12:56 PM
    To Tom's Hardware:

    I also partly agree with Fritz and mastrom and partly disagree! Why, because as long as the drives (both SSD and HDD) are fast enough to catch up with the system in playing VIDEO file, it is in theory, like READ ONLY performance to me. So I can't conclude, but I will assume that SSD drives will use less power than HDD, because we all see the IDLE and LOAD wattage for both type of drives.

    Instead, measurement of power consumption IMO should include a fixed but large file to process, see which drive could finish faster, given that we use the SAME MACHINE (just swap the drive). It is like a "control experiment" which we learn in high school science class. Just one difference in doing the same amount of job - the drive.

    Whatever you called "real life" is very subjective and really vary from individuals. For benchmarking purpose, I think we should just "benchmark" it like you used to do with CPUs - on the same board, same other hardwares.

    If you want to give different type of people the idea of which drive they should choose, haha, may be you should do another 2 tests beside "benchmarking". Which is plain idling (like web browsing, office applications, typing) that sort of less intensive work. And the other one is on intensive side like people who use Photoshop to edit and video editing for theeir whole office hour.

    Sorry, I know it's gonna take a lot of time to do these. It's a matrix of drives x 3 working environment x time you need to test each scenario.

    I don't see this article a conclusion at all, but, it's good work. It makes people think and reconsider.
  • 0 Hide
    sublifer , June 27, 2008 1:38 PM
    I'm surprised nobody touched on this: SSD drives scaling down to 1.8" takes about same power as 2.5", This would imply that scaling up to 3.5" (or maybe 5.25" remember the bigfoot drives?) would take same power (or nearly so) as the 2.5" drives.

    You'd be looking at huge power savings against 3.5" drives, not to mention performance increases. Then you could also add capacity almost indiscriminantly and power consumption probably wouldn't change much since it sounds like most of the power usage is coming from the controller which wouldn't likely change.

    No, we aren't talking mobile power saving benefits with this scenario, but think of all the desktop workstations and home PCs in the world and all the potential power savings there.
  • -1 Hide
    blueeyesm , June 27, 2008 2:10 PM
    Perhaps SSD makers need to seriously invest in the memristor technology HP has invented recently. This would solve their issue of power consumption.
  • 4 Hide
    gwolfman , June 27, 2008 3:14 PM
    I agree with fritz (above & quoted below). Maybe since the drives are quite faster it's actually doing more work in the same time, meaning CPU use is higher and power consumption is up. Just an idea, I don't know how the benchmark program is designed.

    "There could be a systematic error in the benchmarks shown: if the flash based "disks" are faster then the whole system CPU/MEM/Chipset would draw much more power with flash "disks" compared with conventional disks - just because the benchmarks could run more often in the same time.

    Maybe one should compare something like playing video from disk where it is assured that the systems do precisely the same work?"
  • 17 Hide
    durdenet , June 27, 2008 3:14 PM
    Wow, I've always respected Tom's hardware as the most reliable company on tests... well not anymore after this article...

    Two of the BIG mistakes have already been mentioned in this comment page:
    1- "additional tasks you would not have normally done on a laptop (media processing, HD video, etc...)" and the possibility that "Windows doing more background stuff (desktop search, adware/av, etc...) because it doesn't take as long to finish maintenance tasks." would DEFINITELY influence consumption of a whole system but it's not even considered.
    2. "The SanDisk SDD drive at LOAD requires 1.0 mW while Hitachi HDD requires 1.1 mW at IDLE". So clearly, either the graph in "power consumption" is WRONG or the whole article is wrong. So I ask the editors to either take off the article or correct the graph.

    While they correct the errors of this article, do not believe the conclusions of this article. SSD do consume less for any X amount of tasks on a Y amount of time, and that is the only factor that counts. full stop.

    This article measure Y amount of time with a different amount of tasks...... very clever..... :-s
  • -5 Hide
    warezme , June 27, 2008 3:32 PM
    Meeooww!!!, SNARF! Interesting how what I assume to be early SSD adopters viciously defend their prized and over priced new toys..., LOL!

    Face it you guys you where spoofed. Current generation SSD's are over priced and not ready for prime time. You are bankrolling R&D. The results are very valid as most users who actually do "work" on their laptop use multiple programs at once and have various services and applications running simultaneously. They don't just turn their notebooks on and occasionally surf the web or check email and no its not realistic as some goofus proposed to run the test on a linux system.., yea cause we all run Linux. Another guy was clutching at straws about saving a second here and a second there because of the awesome performance increase of the SSD will save so much time.

    LOL! This has been an intertaining comments thread to read. Keep em coming.
  • 6 Hide
    asdasd123123 , June 27, 2008 3:56 PM
    But if the flash drives are as much as three times faster in many aspects..
    That would mean that whatever you are doing, will take less time to complete. And the drive would stop reading earlier.
  • 4 Hide
    Anonymous , June 27, 2008 5:30 PM
    Why is there a huge spread in io operations per second on the charts? Why are the numbers for MobileMark07 Performance and "File write Performance (PCMark05)" identical? Where did they get the "Performance*Runtime/1000" numbers? I can't replicate them from any numbers provided. Not only that, but the runtime minutes were pretty close, so those results would suggest that the Hitachi drives performance was better than the SanDisk SSD, and... well, here are the performance numbers I recreate from the runtime minutes: Hitachi-284, Sandisk-274, MemoRight-290 etc. Go and look at the all those performance charts and tell me, does that look right to you?
  • 0 Hide
    Anonymous , June 27, 2008 6:23 PM
    Third last paragraph:
    "do not even provide convincing performance while they help to suck your battery empty quicker than before."

    Last paragraph:
    "you can at least be sure that you get the best drive performance"

    must....resist.....lol....
Display more comments