Slashdot Log In
Are SSDs Really More Power Efficient?
Posted by
CmdrTaco
on Wed Jul 02, 2008 09:12 AM
from the how-will-i-green-up-now dept.
from the how-will-i-green-up-now dept.
Bakasama writes "Tom's Hardware compared the power performance of several available SSD cards with a Rotating HDD that was chosen specifically for its poor power efficiency.
The results seem to fly in the face of current wisdom.
'Flash-based solid state drives (SSDs) are considered to be the future of performance hard drives, and everyone seems to be jumping on the bandwagon. We are no exception, as we have been publishing many articles on flash-based SSDs during the last few months, emphasizing the performance gains and the potential power savings brought by flash memory. And there is nothing wrong with this, since SLC flash SSDs easily outperform conventional hard drives today (SLC = single level cell). However, we have discovered that the power savings aren't there: in fact, battery runtimes actually decrease if you use a flash SSD.'"
Related Stories
[+]
Notebook Storage SSDs and HDs Compared 149 comments
The Raindog sends us a particularly timely showdown article comparing seven 2.5" mobile hard drives, four of them HDs and three SSDs, across a wide range of application, file-copy, power-consumption, and noise-level tests. Tom's Hardware was recently forced to issue a correction to a claim, which we discussed here, that SSDs aren't actually much more power-thrifty than HDs. The Tech Report's in-depth comparison provides some data points on the question of whether solid-state storage is ready to supplant traditional mechanical hard drives, but notes that the price disparity is still substantial.
This discussion has been archived.
No new comments can be posted.
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
Full
Abbreviated
Hidden
Loading... please wait.
More power but only while being hammered? (Score:5, Interesting)
So if your not a laptop user and aren't currently benchmarking your drive how long will it last?
What is the power usage for real world office/ web browsing type use?
Re:More power but only while being hammered? (Score:4, Interesting)
Parent
Or even real world intensive use (Score:4, Interesting)
Even when I use something that hits my drive hard, it is far from continuous. In fact flash drives might have a chance to shine here as well as often one of the things that increases activity is the need for random access, which magnetic media isn't all that fast at. For example my harddrive works much harder than it has to when I'm doing audio mixdowns because it has to access multiple wave files at the same time. Most of the time is spent seeking from file to file, it takes comparatively little time to read the actual data needed. Flash, of course, has much faster random access. Thus for something like this it'd be loaded lighter, despite it being an "intense" use.
Parent
Re: (Score:3, Funny)
Obviously that cannot be! (Score:5, Funny)
Impossible! Those results are obviously wrong!! Now go back and do the experiment. Keep doing different experiments until we get the desired results!
How dear you try and endanger my stock portfolio?!
Re:Obviously that cannot be! (Score:5, Insightful)
Parent
Re:Obviously that cannot be! (Score:5, Interesting)
Interesting, I didn't notice that. But one thing I did notice is that it didn't seem to take into account real world use, things such as at what point does the frequency of the extra power needed to spool up a HDD outweigh any power savings -- laptops are turned on and off a lot more than a desktop, and if the responsiveness is there, then people are going to put their computer to sleep more often to save power. None of that was taken into consideration. So technically, he may be right in some circumstances, but that might not be the way most people use them in the real world.
Parent
Re:Obviously that cannot be! (Score:4, Insightful)
I'm not so sure your point is quite that significant. There is a graph, towards the end, that seems to be multiplying throughput by total battery life, giving a sort of "megabyte-minutes" rating for the different drives. In terms of simple hard drive throughput, this seems to indicate that the work per watt of the traditional drive was still superior (albeit by a small margin over the most expensive SDD), despite your complaint. But obviously it's not quite that simple--no real life usage would cause non-stop disk access like that.
The claim that the CPU stays more active with the faster drive, while technically true, is a little misleading and not nearly as clear cut as you're making it out to be. The only time the CPU would really be more active with a faster drive is under circumstances where it would be waiting for some kind of blocking I/O from the drive, which in my experience (at least under mundane use) isn't all that much. Most of the time you're much more likely to be dealing with system RAM than hard drive storage during program use (unless you run out of memory and start swapping things out, but then you've got other problems).
In short, while you point out perhaps an interesting oversight, I don't think it is quite as serious as you make it out to be.
Parent
Still too new (Score:5, Insightful)
I believe that much of the problem is that SSD's are still a new technology compared to rotating disks. Right now, engineers are more concerned with increasing capacity and just making the damn things work. These are much more important than efficiency. As time goes on and the technology gets more mature, efficiency will get more attention from engineers.
Re:Still too new (Score:5, Interesting)
Parent
Re:Still too new (Score:4, Interesting)
The technology already exists to reorder disk blocks based on usage and it could easily be adapted to reduce the number of flash chips which need to be activated in an SSD.
Uhhh... wouldn't that defeat the purpose of using wear leveling algorithms?
A more relevant technique would be to avoid turning on flash chips that do not have data you're accessing... but that is not as easy as it sounds when your data is (purposely) fragmented all over the place.
Parent
Re: (Score:3, Insightful)
Uhhh... wouldn't that defeat the purpose of using wear leveling algorithms?
No not if you activate only those chips that you need to execute that read-write-write operation. But then again I'm sure if it really was as simple as switching some thing on and off then we would have good results by now..
btw, look at the IO ops per second graph [tomshardware.com], it's interesting to note that not all SSDs are better than the disk. Though the best SSD beats disks with three orders of magnitude in webserver load..
Re: (Score:3, Interesting)
it's different when you're just reading data, VS writing data, and these benchmarks compared power usage while writing gobs and gobs of data.
flash memory chips can use 5X as much power to 'write' data as they do to 'read' data, oh and hey if you're reading the same old data over and over, why not just have it in ram, and just not spin up a drive or read flash memory at all?
oh and wear leveling can be designed around keeping as few chips powered up as possible, you just need to reserve a bit of flash memory
Re:Still too new (Score:4, Interesting)
Call me some guy mystified by brands but after using Seagate cheap stuff (SATA) and SCSI for years and never seen these things actually fail (besides stupid filesystems), I wait for Seagate, Fujitsu, Hitachi like known brands to ship their SSD rather than being abused by some memory vendor who has no clue about the hard disks to buy some overpriced flash memory fantasy.
I also don't know the actual reliability of SSD too. What about journaling? Can it handle? A journal is still needed on SSD drive, what if kernel fails or OS filesystem layer goes nuts? A journal will be in same area of disk and will be written over and over millions of times.
I could never buy the "speed" claims of SSD not just because I use very fast SCSI stuff but I actually see the horrible performance of them in my smart phone, HD Camera. It is like performance suicide if someone dares to put a very complex applications to "memory card" instead of phones built in memory.
They are trying to ship it before it is a technology fit to general use. Much like some video sites existed while everyone had to struggle with 56K modem.
Parent
Re:Still too new (Score:5, Informative)
I could never buy the "speed" claims of SSD not just because I use very fast SCSI stuff but I actually see the horrible performance of them in my smart phone, HD Camera.
Connect your very fast SCSI drive to your phone and see if it's still as fast.
Parent
Re:Still too new (Score:5, Interesting)
Parent
running a synthetic benchmark 100% of the time... (Score:5, Insightful)
I wish they would measure power usage under conditions that many notebook computer users actually use them in, which does not include running synthetic benchmarks on their computer 100% of the time it's running. Of course, if you keep the machine writing to the ssd constantly then it's not going to show power savings. But how many mobile users' usage patterns include constant reading from and writing to disk?
Re:running a synthetic benchmark 100% of the time. (Score:5, Insightful)
Exactly... The thing about spinning platters is that it takes energy to start up _and_ keep it spinning. So obviously doing read/write 100% of the time would bias towards the conventional hard drives.
Hell, even read/write 10% of the time is too much for normal usage.
Parent
Sweet (Score:5, Funny)
Now I'm even happier that I paid the extra $1300 for the SSD option in my MacBook Air.
That way it will run out of battery sooner, leaving me free to use a real computer.
Not so good benchmark (Score:5, Insightful)
I believe this comment from the article could explain some of this away.
'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?'
Re:Not so good benchmark (Score:5, Insightful)
Parent
Re:Not so good benchmark (Score:5, Informative)
Yeah, another post mentioned that a SSD's power consumption during load is higher than another magnetic disk on idle... here it is:
"The SanDisk SDD drive at LOAD requires 1.0 mW while Hitachi HDD requires 1.1 mW at IDLE"
Of course, benchmarks are always a better indicator. Also, to be clear, as mentioned in the article (which I'm sure nobody else read), the test was performed by repeatedly running a benchmark on the system until it ran out of battery, so the test with the SSD is likely to have run the test more often.
Parent
Re:Not so good benchmark (Score:4, Interesting)
Of course if the SSD is running under load the entire time it may actually run more often than a regular HD.
from real world benchmarks SSD's only gain is from random reads. writes, sequential reads, etc all prove that SSD's are only as good as a regular spinning disk if not far worse.
It is a trade off. both have advantages and disadvantages. Maybe in 5 more years SSD's will ultimately win but for now it can go either way.
Parent
Re: (Score:3, Insightful)
Re: (Score:3, Interesting)
Somebody correct me if I'm wrong, but isn't it true that with HDDs, the system can auto-power them down when they're idle, but with SSDs, this can't be done as easily due to the nature of SSDs?
Re:Not so good benchmark (Score:5, Informative)
Parent
Re:Not so good benchmark (Score:4, Interesting)
Parent
Re:Not so good benchmark (Score:4, Interesting)
That's an excellent point. I was wondering if the metric should simply be different altogether.
What if we used MB/Watt or some such? The "disks" are supposed to be really good at some things (random reads) but don't hold too much advantage over others (long continuous reads). So how many WattHours does it take to load a continuous 50 GB file? How about a random 50 GB of data off a 128 GB disk? How does that compare to the same measurements with a standard magnetic disk? How does power consumption change between reading, writing, and mixed disk loads? Writing flash takes far more power than reading, doesn't it? Yet on a physical disk it's not that different in power requirements.
There are other things too. Operating systems still aren't designed around these things, they are designed for physical rotating disks. Do these things even have native controllers (designed for this purpose) yet, or are they still using modified rotating disk controllers like the first models used? As time goes on, better power saving features will show up, especially as the OS cooperates to tell this disk more information about what's going on. A well managed flash drive may be able to shut off large chunks of it's self and only wake up the bits that actually need reading/writing. That would help quite a bit, I'm sure.
PS: First time I've been to Tom's Hardware in 6 months to a year. Nice to see they found a way to make it uglier. Used to be a nice site. I especially like the "you must login to see the printer friendly version" trick.
Parent
Re: (Score:3, Interesting)
PS: First time I've been to Tom's Hardware in 6 months to a year. Nice to see they found a way to make it uglier. Used to be a nice site. I especially like the "you must login to see the printer friendly version" trick.
I don't go there at all anymore; I've got my ipcop box running URL Filter with the Squidguard blacklist, and whatever tracking junk they run at Tom's just leaves me hanging with a 1x1 gif. It's not even worth my while to add them to the whitelist.
Been there, seen that, got the t-shirt (Score:4, Insightful)
A 32" Philips HD ready CRT was around 100-110 W at the time I looked.
However, this is highly dependent on brand as well.
Re:Engineering is a trade-off (Score:4, Informative)
That's 2ms grey-to-grey (or worse...).
At best you're looking at 4ms (white-to-black) which is 250 Hz.
Black to white is actually generally faster than grey-to-grey, not slower. That's why the whole grey-to-grey benchmarks started showing up. Because screens that could go black-to-white in 4ms were easily available, but they still had grey-to-grey times of 32ms for some level transitions.
Further, most 'fast' screens use 'overdrive' which actually overshoots the target destination color (because larger transitions are faster than smaller ones), and then brings it back down to the target color. (leading to 'sparkle' when whatching movies etc because a pixel on a small transition from 'almost black' to 'just a little bit less almost black' shoots through medium-grey to get there.
Parent
Good article.. BUT... (Score:5, Insightful)
The thing is that Tom measured battery life with the hard drives constantly working.
He glosses over this with the following statement:
No, Tom, no one turns on their computer and simply waits for the battery to die. However, no one turns on their computer and has their hard drive constantly thrashing either.
Typical usage patterns include document editing, movie watching, music listening, etc, which involve very, very small amounts of hard drive access.
Use a better battery benchmark that leaves the hard drive idle most of the time, then come back and let us know how these drives fare.
Re:Good article.. BUT... (Score:4, Insightful)
I think the article is still somewhat Valid, though, even if it does gloss over some of the details a bit.
The short end of it is that a SSD will use less power when idle, but more power at full load than a typical hard drive will, which may be a real factor when deciding which to purchase.
Personally, though, I don't see the real benefits to them in 99% of situations. Their performance is only marginally better than standard HDD's and, as you pointed out, MOST people wont regularly thrash the HDD anyway. Plus, they're expensive - you could spend a little extra on a bigger and/or additional batter and some more RAM, get a massive regular hard drive and probably still come to less than half the price of a 32Gb SSD.
Hopefully when the technology matures a bit, they'll be to HDD's what the common DVD was to CD's.
Parent
Re:Good article.. BUT... (Score:5, Informative)
Parent
Re:Swap (Score:5, Interesting)
Because in general folks know how to do it (so you're not adding code complexity in most cases) -- and it seamlessly handles the odd folks who *do* eat all of their RAM in a workload and end up needing it. (And since the folks with big working sets relatively to current states also tend to be the folks paying more money... they do get listened to -- and these are the same folks that would require a perfect patch in 24 hours when they're unhappy, so you're much better off having a little planning pre-release than trying to crowbar this in post).
The thing to do is to make the virtual memory subsystem as efficient as possible about handling swap statistics just in case you need it so that the folks who really don't need it aren't aggressively impacted.
(Note: I am a virtual memory subsystem kernel engineer -- but not on Windows. I make no claim about how efficient or inefficient Windows is at doing this as a result. I would seriously expect that since they're designing the core kernel to operate from laptops up through Windows Server Whatever --
they have to accommodate cases beyond the 4Gb in your laptop should be enough to keep everything in core, though. )
Parent
Re: (Score:3, Interesting)
Used/tried a Quad G5 with 16 gig memory installed (mine has 4,5 GB).
Even on Leopard, that mainframe like monster will still create a 64MB swap file amusing us. I also saw 6 swapfiles has been created on my 4,5 GB RAM installed Tiger (OS X 10.4, doesn't have auto cleaning) after days of heavy load uptime.
Is it a strategic thing that OS kernel does? I remember reading some real weird stuff about Microsoft's tactic to swap the applications to disk even while they didn't have to but I can't find the file now. Y
Re: (Score:3, Informative)
well In linux and OS X you can actually disable swap. Windows however relies on it heavily, and with Vista under 4 gigs of ram you need swap space to store the entire OS.
Re: (Score:3, Insightful)
Author = Clueless (Score:5, Insightful)
He makes the claim in the comments about the article that "well who just watches dvds? You have to keep the system busy and test that!" That's about as valid as setting the machine not to sleep and seeing how long it can idle there.
On an ultraportable especially, you aren't going to be churning the CPU with a benchmarking program. You won't be rendering animation frames. Mostly you'll be in a web browser or text processing program, waiting on the user to type something. With occasional spurts of OS and program start/stop. Good gosh it sounds like a MIX of tasks, rather than either extreme.
Re:Author = Clueless (Score:4, Informative)
At Apple (a few years ago), we would test typical portable battery life by scripting a set of tasks for the computer, both in OS X and in Windows. This way, we would be trying to simulate this common mix of tasks and obtain more realistic battery lifetimes (and comparison between similar Windows laptops and our own). Naturally, it wasn't always the case that our benchmarks were the ones put up on the web.
Stuff like this benchmark is really just an extreme corner case. As an engineer who relies on lots of hardware to help perform long and complex simulations, I know something about thrashing a system to death. And yet, I would never, ever, ever run such a continuous thrasher on my laptop (at least without plugging it into the wall). Their scenario seems (to me) as extremely unrealistic and may qualify as FUD.
Parent
Author = Clueless and Stupid (Score:5, Informative)
I have a 32GB SSD in my T61. My real life usage shows that I get between 30 and 45 more minutes of battery life out of my SSD-equipped notebook than on my other T61, which has a 160GB 7200rpm drive in it, when both of them are on the "medium" power saving setting in Windows.
Re:Author = Clueless and Stupid (Score:4, Interesting)
My experience matches yours. I had a huge increase in battery life when I swapped out the HD in my Fujitsu Stylistic 2300 for a CF-IDE adapter w/ a 2GB (booting Windows 2000) and 4GB (data, swap space) CF cards in it.
Ran much cooler and was absolutely silent (no fan).
I was quite bummed when the system died.
William
Parent
Am I the only one... (Score:5, Interesting)
Who thinks that maybe the overall bettery life would be more affected by the HEAT the laptop is putting out?
After all, most laptops I've ever dealt with or owned are compact little furnaces that have their hard drives, cpu, gpu, and everything else shoved in as little space as possible in order to make room for things like batteries and keyboards.....
thus if they remove a high heat generating device (even more so with a 7200rpm drive) wouldn't logic also assume that you reduce the amount of heat needing to be removed from the system?
thus the fans would run less often, and therefore drain less battery power.
not too mention the proposed savings by simply accomplishing tasks "faster" when it comes to disk access.
Re:Am I the only one... (Score:5, Insightful)
Well, since power usage is directly related to heat output, we could benchmark SSD's versus standar drives much more scientifically; we could even use the author's broken benchmarking tool. Put the hard drive into a calorimeter, run the benchmark for 10 minutes and measure the total heat output. Repeat for the drive you are benchmarking against and presto, actual, scientific numbers on power usage for just that device, rather than simply plugging away until the battery runs dry.
Parent
Power Efficient? They can barely aim! (Score:5, Funny)
Seems unlikely to be all that power efficient to me... hauling around that huge bulk, and it sure didn't seem like Executor really made all that difference at the Battle of Endor from a capital ship perspective. (Probably because most turbolaser batteries seem to have really lousy guidance,
after all).
Oh... you meant those SSDs.... my bad.
Old and inaccurate (Score:5, Informative)
The article states nothing new - there are two very interesting blogs from Lenovo which already stated the same in August 07 (!). To quote:
Solid state HDDs promise to save power compared to traditional hard disk technology. And they will. However today's generation of SSDs have no power savings benefit compared to traditional HDDs. The big reason is that current SSDs with a Serial ATA interface are actually Parallel ATA hard disk drives with a serial bridge chip. They don't offer support for low power interface states and the architecture has a potential for data-losing error conditions when recovering from a low power state like suspend or hibernate. In the future, there will be native SATA solutions which will solve many of these problems and will at the same time offer a real power savings benefit which should increase battery life.[1]
An updated quote from a newer blog:
Power Consumption - All SSDs are going to save you battery life on your notebook, but some will save you more than others. Again, the native SATA drives will give you better battery life.[2]
To summarise: old news and mostly outdated with very recent SSD drives.
[1]: SSD part 1 (Aug 07) [lenovoblogs.com]
[2]: SSD part 2 (March 08) [lenovoblogs.com]
Check Page 14 (Score:5, Insightful)
In TFA, there is a graph on page 14 with power consumption measurements for the 5 drives tested.
The SanDisk SSD shows 1.0 watt active, 0.5 watt idle.
The Hitachi drive (magnetic) shows 3.2 watts active, 1.1 watts idle.
So even if the SanDisk drive spent 100% of its time in active mode and the Hitachi drive was always idle, the SanDisk drive should still provide longer runtime.
However, their runtime test (page 12) shows 7:03 runtime with Hitachi, 7:02 with the SanDisk.
All they have to say about this is:
Most of the power consumption measurements are in line with our results in Mobilemark 07. However, it has become clear that idle and maximum power do not provide the full picture when we talk about flash SSDs.
Well, something clearly is wrong here.
iPod counterexample (Score:3, Informative)
Obviously the use model for both devices is the same.
Already debunked.... (Score:4, Informative)
The testing methodology was flawed to draw any conclusions. The problem is the CPU may have been more active due to less IOWait states. AS a resulte, the drive consumption may be lower, and the benchmark was not throttled to the platter disk performance. The benchmark might have run many more times during the test.
Take with grain of salt (Score:5, Informative)
The title "Some SSD devices are inefficient" just does not drive page clicks. So Tom sensationalizes some facts, omits others, runs questionable synthetic benches and Voila... Slashdot delivers the page hits.
Tom sometimes takes known "problems" and dramatizes them. After all, an objective view should not be allowed to stand in the way of revenue generation.
As others have pointed out.
1. The reason for inefficient drives was usually power hungry interface logic chips.
2. The newer drives are all better according to a Lenovo blog (thanks jlp2097)
And to top it off the numbers don't even add up (see PJRC's nice post above).