Slashdot Mirror
Samsung Announces Fastest 64-GB SSD
XueCast writes "The new solid-state drive from Samsung can write data at 100 MB/s and read at 120 MB/s. This handily outperforms other SSDs now on the market, which typically feature only 50-80 MB/s read/write rates. Samsung's SSD will come in two form factors, 1.8" and 2.5", and will be running on the SATA II standard. It will only consume 50% of the power of current SSDs. There is no information yet about price."
This drive doesn't outperform MTRON (http://www.mtron.net/english/). They announced 120 MB/s read, 90 MB/s write drives and they are shipping 100 MB/s read, 80 MB/s write drives already. The SSD-based Fusion IO card (http://www.fusionio.com/) at the claimed 800 MB/s read and 600 MB/s write speed would beat both them handily. Still, it's good to see a major manufacturer up its speeds.
Texas Memory Systems http://www.superssd.com/benefits.htm says can saturate Fibre Channel (GBs/sec) and this one caps out at 100s of MB/s. Perhaps not quite so unequivocally outperforms as this statement makes it out to be.
How about outperforms other flash based SATA SSDs now on the market???? What is surprising is that more of the SSDs don't max out the SATA pipe.
yeah they are in different price classes but it isn't like SSDs haven't been around for long time now. Inexpensive ones that you can put into your sub $1,000 computer... perhaps that is new. Yet another sensationalized copy in a Slashdot story abstract. Oh so surprising.
In what price range are we talking ?
Yes, I'm left. You have a problem with that?
Savior of the Universe!
Thanks to algorithms that spread written data across the chip, MTBF's of SSD are much higher than those of regular HDDs with similar usuage patterns.
Furthermore, A simple buffering scheme sounds likely to solve most of the problems you're talking about (Assuming it's constantly many small writes done by the OS... for say, log file keeping or file access-time updating).
Sigs are for the weak.
distributors are definately in the process of getting io down. So is Linus himself. quote from http://lkml.org/lkml/2007/8/5/171 : "change relatime updates to be performed once per day." It's not only the livetime of flash memory that benefits from this. also power consumption and noise goes down for hdds. and overall io performance benefits fromsuch improvements,too. About the swap: just keep it big enough so the Kernel can free the ram of some unused data, but not a lot bigger. Twice the size of the ram is nonsense these days.. if you run out of buffers and cache you don't have enough ram. if you have enough ram swap is hardly used.
Cheap, fast, good - pick two.
.... will only consume 50% of the power of current SSDs"
.... Ah, shit!
"write data at 100 MB/s and read at 120 MB/s."
Hey cool, that's pretty fast.
"64GB
Good, good.
"There is no information yet about price."
She's built like a steak house, but she handles like a bistro....
The announcement was in March, mass production in June and availability in September.
I haven't seen a price yet but it's going to be at least close to a grand.
thegodmovie.com - watch it
http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2013240636+1421430848&name=64GB
And bigger, 128GB:
http://www.newegg.com/Product/ProductList.aspx?Submit=ENE&N=2013240636+1421430849&name=128GB
Yes, the prices are exorbitant. Just wait, patience is a virtue. At least we can actually see and purchase the current status of SSD, and at the rate they are increasing it will phase out hard disks in both capacity and price.
Is this what you call something that has been duped twice?
Today, plus...
Oct 28: http://hardware.slashdot.org/article.pl?sid=07/10/28/1337207
Oct 25: http://hardware.slashdot.org/article.pl?sid=07/10/25/149202
I shall make this SSD my flagship.. and I will call it the Executor.
They cost $920 when added to a Dell laptop. The 64-GB SSD is available initially on Dell's XPS M1330 ultraportable notebook Relevant Products/Services, and, later this year, on other models in the XPS line, as well as on Latitude corporate notebooks and Dell mobile workstations. For Alienware, users can choose dual 64-GB SSDs in RAID 1 or RAID 0 configuration, or a 64-GB SSD in combination with a magnetic drive for the Area-51 m9750 high-performance gaming notebook. Prices start over $1,000 for the SSD additions. As far as price is concerned. I would rather get this. http://www.engadget.com/2007/08/21/toshibas-320gb-2-5-inch-hard-drive-a-worlds-best-for-laptops/
And if battery life really concerns you probably getting external battery from electrovaya or batterygeek may eliminate that worries.
1. Get sixteen, 4 GB SDHC, Class 6 or 8 innards
2. Strap the lot in parallel, giving 64 GB
3. 6|8 MB/sec/innard x 16 innards begets 100 MB/sec
4. Profit !!
Each 4GB innard is $20 currently, so 16 by 20 is 320. Figure $10 for plumbing. 1% margin for OEM (335), 50% markup by distributor (500), and another 50% by retailer (750), and there you have it $750 for 64 GB.
Thank you !! Come again !!
Yeah, like that'd help. Do you think your USB-chipset can handle much more than 500mbps concurrent traffic? Doesn't matter how many ports it has, it is unlikely to be close to S-ATA speeds seeing as the USB-chipset is on the regular (1gbps?) PCI-bus...
...what does a large google show?
Is your $24 card a high speed one?
No sig today...
First, you mean MBps. We're talking bytes, not bits.
Second, your hard drive can sustain 60MB/s on the fastest part of the drive. Its average is probably much less than that (due to different linear speeds on the inside and outside of the platters).
That speed drops catastrophically in many real-world scenarios. Small random reads, for example, become dominated by seek time and rotational latency and the high transfer rate doesn't help very much. Small random writes are only slightly better.
It is really not "only double". It has a real-world speed that is about twice a high-end hard drive's theoretical maximum speed.
MTBF doesn't mean what most people think it means, and is less useful than most people treat it.
There is no read limit. The write limit is about 100,000 writes (really erasures) per cell.
These devices will have wear leveling. That means that if a cell is close to running out of erase cycles, the drive will move data that has not changed in a very long time into that cell. A few cells will be kept as spares in case some cells don't last as long as they are predicted to.
If you do the math, and figure a typical use scenario as a laptop's primary drive, you get that these drives should outlast mechanical hard drives by many years. For example, a 64GB hard drive with an endurance of 100,000 writes should be able to tolerate about 5 million GB of writes before it fails due to wear.
How long it will take you to run that out depends on your average write rate. But with a reasonable rate (10MB/s) that works out to about 15 years.
Maybe I've been an elitist geek for too long, but I clearly remember "real" SSDs being a heck of a lot faster than 100mb/sec. Of course, they used actual DRAM instead of flash, and they'd lose everything if your battery ran out. It was essentially a hardware Ramdisk, with the (then-tremendous) benefit that it doesn't depend on the PC's memory controller, so back when the average PC had 16mb ram, you could have a 640mb SSD that pwned everything without breaking a sweat.
A few years ago there was this bizarre Gigabyte i-Ram gadget that took four DDR dimms of any size and connected by SATA, it was relatively cheap too at ~$125 (sans Ram). If they had made a larger model, say 8 or 16gb, I'd be all over it! There's also this FusionIO company that's kind of spinning its wheels right now, in true dot-com style, but they're at least trying to bring the concept of DRAM-based storage back into the spotlight.
Even with 15k drives and RAID, there are some things that just take forever on my workstation (random access stuff). Consumer equipment is getting really fast, but the high-end has been stagnating for years. With more and more people taking advantage of quad-core processors, dabbling with audio/video editing and hi-def content, not only do we need larger capacity, but we need massively increased transfer rates to match. What good is a terabyte disk if it takes 10 hours to read/write the whole thing ? Where are my 150mb/sec transfer rates ? Why design high-speed SATA interfaces if the actual drives can't even use a third of its juice ?
These flash drives serve a purpose, yes, but I think it's safe to say their target market is less concerned about transfer rate and more about battery life and shock-resistance. For the other 98% of the world, we want more speed dammit!
-Billco, Fnarg.com
Does anyone around here know of any numbers backing up the claimed high values for MTBF? I'm not unwilling to accept that the values are indeed high, but I'm looking for something closer to reality than the Wikipedia article arriving at an expected lifetime of 26,600 years.
The flash memory modules I've encountered have guaranteed a minimum of 100.000 write cycles per data memory byte before failure (NDAs prohibit me from listing the specific devices, but I suspect that this number is nothing out of the ordinary).
With a page size of 1024 bytes, a 64GB drive would hold 64 million pages. If we assume that all updates require a full page erase-write, but that a clever algorithm distributes updates evenly, this leaves us with a guaranteed life-time of 6.4 * 10^12 (6,400,000,000,000) updates before memory failures start rolling in.
That's without doubt more than sufficient for desktop usage, but let's for a moment assume that you're able to max out the drive, writing at the rated speed of 100MB/s. With a page size of 1024 bytes, that's 100.000 page updates every second, so failure will set in after 64,000,000 seconds = 2 years.
Now, assuming that you're able to feed the drive at 100MB/s is probably way off, but on the other hand your wear levelling algorithm will probably be far from perfect.
I just had to go look at the Fusion IO page and their FAQ and... well, let's just say, does anyone have an URL for marketting-bullshit-bingo to English babelfish please?
By the half of the page I had developped an extreme allergy to the word "leverage". Two sentences out of three were just saying that the lever some (supposedly awesome) proprietary technology. And more importantly, I was none the wiser. There wasn't a single sentence that even said what it _does_. What makes that technology so awesome? What's the MTBF? You know, some actual technical data.
The more I think about it, the more I doubt that it was actually a Frequently Asked Questions. More likely just something that a marketter thought up, along the lines of:
Q: Are you awesome?
A: Yes, we leverage proprietary technologies to be uber-awesome. We leverage Buzzword(TM) and Uninformative Trademark(TM) and Tech-Sounding-Word-We-Made-Up(R) to be so awesome, that you can't even imagine how awesome we are. And we'll leverage that too. Leverage. Leverage. Leverage.
Q: Does it rock?
A: Yes, we leverage proprietary technology that really really rocks. We liberate enterprises from legacy architectures, we're scalable, we put enterprise-level SANs in your palm, we solve world hunger, cure aids, and probably filled your bullshit bingo card already. That's how much we rock.
Q: Will it rock my socks off?
A: Yes, our awesome leverage proprietary sock-rocking technology. We're that awesome. And did we mention "leveraging" yet? We leverage a lot.
Not exact quotes, but let's just call it an artist's impression. I haven't heard a more content-free text since someone accidentally sold us 100% tech-illiterate merketers when we thought we wanted a technical workshop.
Don't get me wrong, I don't doubt that their engineers probably know their shit. But that's what happens when you leave the FAQ writing to a marketer who doesn't know his arse from his elbow, and obviously think that using enough words will hide the fact that there's no information there.
And just to beat a dead horse some more, what annoys me isn't as much the use of buzzwords, but that they're used to obscure and mis-inform.
E.g., so they say it's "scalable"? How? Your typical motherboard has only one 4x PCI-Express slot, and on half of them it will be under the heatsink of any high end graphics card. So how _do_ you scale there? Throw the card away and buy a bigger one? How's that more scalable than buying a new hard drive? Even if you had more of those slots on some special motherboard, how's that more scalable than buying more hard drives? No, seriously.
E.g., the claim to replace an enterprise SAN and all the infrastructure... is omitting why that infrastructure was there in the first place. If anyone just needed more storage on their local machine, it's trivial to add more than 640 GB hard drives locally for a fraction of the cost. A hard drive, even on a card, is not a SAN replacement.
E.g., video games are hard-drive intensive? No shit? What video games were they playing there? Database Larry Rebuilds The Indexes 3D? Looks to me more like they wrote a list of every single use they could think for a computer, than actually having put some thought into it.
Etc.
Again, I'm willing to give their engineers the benefit of the doubt. I can see why such a card would be nice. Just saying that it would be nice if their good work was presented to the world by someone less blatantly clueless.
A polar bear is a cartesian bear after a coordinate transform.
"let's for a moment assume that you're able to max out the drive, writing at the rated speed of 100MB/s. With a page size of 1024 bytes, that's 100.000 page updates every second, so failure will set in after 64,000,000 seconds = 2 years."
2 years seems pretty impressive to me for beating the virtual snot out of your test subject testing in a completely unrealistic scenario. I would be surprised if my car's engine survived 2 years of running non-stop at 7,000 RPM.
Samsung claims an mtbf of 2,000,000 hours, which is only ~200 years, not 26,000.
I've seen some specs listing 300,000 program/erase cycles, minimum, which would boost your 2 years to 6, and note that that's their minimum guarantee, the average lifetimes are expected to be considerably (as much as 10x) higher. Presumably these devices just write off a page if it goes bad.
"Who is the Journal of Quantum Physics going to believe?" --Stephen Hawking
That 60 MB/s is almost never attainable in practice.
SATA drives have a seek latency of about 9ms. This means that the drive can perform 111 seeks per second. Assume a very pessimistic scenario of reading a 2KB cluster. Your drive's performance is now about 200KB/s.
For an expensive and low capacity SCSI drive, you can get 3.3ms, with about 600KB/s worst case scenario.
This is assuming you're actually reading data you're interested in. Some of that will involve reading filesystem metadata, which from the user's POV isn't what you're actually interested in. For a directory with lots of small files I imagine you could get maybe half of that performance.
I've seen SSD latency being quoted to be around 0.01ms. The same calculation above gives 195MB/s, assuming reading takes no extra time (which is false)
From this you can see that a hard disk is highly limited by seek latency, while a SSD is much more limited by read/write speed.
Airplane mechanics actually do keep track of flying hours as a maintenance target, but cars lifetimes usually get counted by miles. However, suppose you do look at times - my experience with Chevy engines has been that they last about 120,000 miles, so at an average speed of 30mph, that would be 4000 hours, or about half a year (other cars should of course run longer :-) While 7000 RPM is obviously not a good speed to run the engine at, running it continuously for long periods of time is likely to be much better for it than realistic operating cycles. Diesels would probably do even better on long continuous use applications.
Bill Stewart
New Fast-Compression-only CPR http://preview.tinyurl.com/dy575ks
How often do you feed a disk (not in a server) at 100MB/sec for any sustained period of time? Heck, how often do you feed a disk 100MB/sec for one minute much less an hour much less 2 years straight.
I suggest two things:
1) those so paranoid about drive life return to their handy array of 9.1GB disks in raid 50 and leave the thread
2) run perfmon (or the linux equiv.) and look at your overall disk writes for an average day, triple it and then calculate the number of years the drive will last and cut it in half for the hell of it. I'd guess the computer and storge of the drive will be long obsolete before the expected lifetime.
If you need to handle writing 100MB/sec of data at a constant rate for weeks/months/years then you don't need a 1.8" SSD. You need a couple pentabyte san. These drives are *perfect* for normal users, power users, heavy users. I'd gladly put one in each of our developer's PCs for doing coding and builds. Our AV guys would love them too.
You can get rich if you own a politician, but you have to be rich to buy one in the first place.
More than 10x higher. No one sustains 10MB/sec of writes 24/7 or even averages that on an individual's computer/laptop.
The only situation you might find to push that is a dedicated high-use AV workstation in a 24/7 media company. Oh, and never mind that workstation would be using arrayed drives for additional speed and redundancy isntead of a single drive...which would of course increase the expeted overall lifetime.
You can get rich if you own a politician, but you have to be rich to buy one in the first place.
Where are you getting your numbers? Block erase is 128K or 256K on NAND flash that I've looked at, and erase time is about 1500 usec. The large flash drives measure sequential I/O across multiple chips to get their meaningless performance numbers. Random writes are still painfully slow. The controller keeps an erase count for each block to do load leveling, and when a page gets used too many times it has to swap the whole page with one that isn't used much. The article claims 8GB per chip, which seems high to me. I think they might mean 8Gbit per chip, which would mean 64 chips to do 64GB.
Since there is an onboard controller with a RAM buffer, it can do a verify on every write. Flash tends to fail at erase or write time which can be recovered with no data loss, so MTBF depends on how many spare blocks you leave.
Intron: the portion of DNA which expresses nothing useful.