Anandtech: Understanding SSDs and New Drives from OCZ
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
-
- Patron of SPCR
- Posts: 749
- Joined: Sat Nov 11, 2006 6:02 pm
Anandtech: Understanding SSDs and New Drives from OCZ
If I may say so, an outstanding article: The SSD Anthology: Understanding SSDs and New Drives from OCZ.
It covers why for some time, the only real options were the expensive Intel drives. It covers the basics of SSDs and why some are no improvement over HDDs at all (extremely high average latencies).
OCZ was pissed after Anand's former article, but they came back strong: they sent him some drives again, and after testing took the findings to heart and implemented the advice extremely fast (page 20/21 - a must read).
Respect to OCZ.
Very interesting and fun read, but for the synopsis: Intel still the best and expensive, OCZ Vertex most bang for the buck and acts like a slower Intel, yet much faster than current HDDs.
It covers why for some time, the only real options were the expensive Intel drives. It covers the basics of SSDs and why some are no improvement over HDDs at all (extremely high average latencies).
OCZ was pissed after Anand's former article, but they came back strong: they sent him some drives again, and after testing took the findings to heart and implemented the advice extremely fast (page 20/21 - a must read).
Respect to OCZ.
Very interesting and fun read, but for the synopsis: Intel still the best and expensive, OCZ Vertex most bang for the buck and acts like a slower Intel, yet much faster than current HDDs.
-
- Posts: 15
- Joined: Tue May 08, 2007 9:00 am
-
- Posts: 376
- Joined: Tue Jul 22, 2008 6:05 pm
- Location: Calgary, AB, Canada
I have to disagree with the article on one thing:
They say that the drawback of the reduced random write performance (RWP) is outweighed by the increased performance overall.
I object to this if the reduced RWP does result in noticeable stutters. Because I do expect my computer to slow down if I start an application/load a game level/... This does not disrupt my workflow.
However, seemingly random stuttering while actually doing work will disrupt my workflow.
If the computer takes 20 seconds longer to boot - well, I am usually getting some water during that time, anyway. If it takes 15 seconds longer to load a game - well, time to drink something. I can anticipate those disruptions, and fill them with all those little things that need to be done. Blowing my nose, scratching my head, ...
However, if I am coding/gaming/... and the computer repeatedly hangs for half a second (worst number quoted in the article), this will seriously irritate me.
They say that the drawback of the reduced random write performance (RWP) is outweighed by the increased performance overall.
I object to this if the reduced RWP does result in noticeable stutters. Because I do expect my computer to slow down if I start an application/load a game level/... This does not disrupt my workflow.
However, seemingly random stuttering while actually doing work will disrupt my workflow.
If the computer takes 20 seconds longer to boot - well, I am usually getting some water during that time, anyway. If it takes 15 seconds longer to load a game - well, time to drink something. I can anticipate those disruptions, and fill them with all those little things that need to be done. Blowing my nose, scratching my head, ...
However, if I am coding/gaming/... and the computer repeatedly hangs for half a second (worst number quoted in the article), this will seriously irritate me.
Maybe a little. Still, I prefer this to having to second guess overly positive reviews.Bar81 wrote:It was nice info but too self fellating.
Edit:
@K.Murx
The new OCZ drives don't have the stuttering problem anymore. Their random writes are much slower than Intel can do, but they are still much faster than the Velociraptor results. Take a look at the random read/write tests page of the article, and you'll see why this improved Vortex was considered acceptable.
Last edited by Modo on Thu Mar 19, 2009 1:35 pm, edited 2 times in total.
-
- Posts: 1406
- Joined: Tue Feb 13, 2007 7:28 pm
- Location: USA
Then you don't disagree with the article. Maybe you didn't read the whole thing or just missed the specific section (what was it like 30 pages long!?), but Anand was pretty emphatic about stuttering being unacceptable. There was the whole story of how he refused to give the initial release of the Vertex a passing grade, because it stuttered. It was not until they fixed this problem, by sacrificing some raw throughput, that he approved. The actual conclusion of the article is that it's ok to have mediocre RWP, not pathetic RWP.K.Murx wrote:I have to disagree with the article on one thing:
They say that the drawback of the reduced random write performance (RWP) is outweighed by the increased performance overall.
I object to this if the reduced RWP does result in noticeable stutters. Because I do expect my computer to slow down if I start an application/load a game level/... This does not disrupt my workflow.
However, seemingly random stuttering while actually doing work will disrupt my workflow.
Oops. Seems like I skipped over the part that contradicted my too-fast assumption.
There are a couple of points, like
But he mentioned a couple of times in the latter pages (which I skipped on my first reading), that he considers the JMicron drives unfit for sale.
I make my apologies!
There are a couple of points, like
out of whom I drew the conclusion that the author was of the opinion that any SSD would be better than a HDD.I still believe that a SSD is the single most effective performance upgrade you can do to your PC; even while taking this behavior into account. While personally I wouldn’t give up a SSD in any of my machines, I can understand the hesitation in investing a great deal of money in one today.
But he mentioned a couple of times in the latter pages (which I skipped on my first reading), that he considers the JMicron drives unfit for sale.
I make my apologies!
Hi,
I can tell you just one thing:
I'm using OCZ Vertex since almost 2 weeks and sold already my Samsung F1. I will post benchmarks tomorrow - after 9 day of using + benchmarks after 'reseting' an aligning but still with 'old' firmware. You can expect something interesting. I really don't care about benchmarks anymore - maybe Intel SSD is better with benchmarks but I do not notice ANY problems with my 30GB Vertex.
In real life usage (home user, WinXP SP3, FF cache on RamDisk) I don't see how higher random writes can speed up PC beyond this what I see now - simply speaking, everything is NOW. In my opinion there is no more problems with writes for SSD's - at least as a OS's drive. Only thing which I'm still concerned is longevity of those drives.
Sorry if I sound too enthusiastic but I'm using one of those Vertex (not just reading benchmarks) and before I was VERY skeptical about SSD. Now I'm just looking for some money to earn to bay more Vertex...
I can tell you just one thing:
I'm using OCZ Vertex since almost 2 weeks and sold already my Samsung F1. I will post benchmarks tomorrow - after 9 day of using + benchmarks after 'reseting' an aligning but still with 'old' firmware. You can expect something interesting. I really don't care about benchmarks anymore - maybe Intel SSD is better with benchmarks but I do not notice ANY problems with my 30GB Vertex.
In real life usage (home user, WinXP SP3, FF cache on RamDisk) I don't see how higher random writes can speed up PC beyond this what I see now - simply speaking, everything is NOW. In my opinion there is no more problems with writes for SSD's - at least as a OS's drive. Only thing which I'm still concerned is longevity of those drives.
Sorry if I sound too enthusiastic but I'm using one of those Vertex (not just reading benchmarks) and before I was VERY skeptical about SSD. Now I'm just looking for some money to earn to bay more Vertex...
-
- Posts: 1406
- Joined: Tue Feb 13, 2007 7:28 pm
- Location: USA
No. If you read the comments on the Anandtech article you'll see that there has already been a revision (maybe even two) released since the firmware he approved . This revision is supposed to increase sustained write without hurting random write.Modo wrote:According to AnandTech, all Vertex drives sold through normal channels have the modified firmware, which effectively removed the stuttering problem. By "old firmware," do you mean a pre-production sample that still could break?
That's my point exactly: All officially sold Vertex drives have firmware that essentially removes the stuttering problem.jessekopelman wrote: No. If you read the comments on the Anandtech article you'll see that there has already been a revision (maybe even two) released since the firmware he approved . This revision is supposed to increase sustained write without hurting random write.
Its a great read, and i was sold on getting a vertex over the x25-m until the last page where he briefly mentioned hardware compatibilities. And to check out OCZ's forum for more info. Well i did, and i didn't like what i saw. Too many postings for my liking which include data corruption and just plain hardware compatibility issues.
I'll pay a little extra and get the Intel drive which i KNOW for a fact will work, without issue, the way a mass storage device should.
Still, its nice to see that at the very least people now know what causes these studdering problems and are working to correct it.
I'll pay a little extra and get the Intel drive which i KNOW for a fact will work, without issue, the way a mass storage device should.
Still, its nice to see that at the very least people now know what causes these studdering problems and are working to correct it.
Modo:Modo wrote:That's my point exactly: All officially sold Vertex drives have firmware that essentially removes the stuttering problem.jessekopelman wrote: No. If you read the comments on the Anandtech article you'll see that there has already been a revision (maybe even two) released since the firmware he approved . This revision is supposed to increase sustained write without hurting random write.
As far as I understand this technology, stuttering problem was solved by adding cache and Indilinx controller. Firmware is more about right balance between I/O, read, write etc.
Reading the article makes it clear that while cache might have something to do with it, the stuttering is caused by low random write speed and the latter firmwares sacrifice sequential write speeds for random write and thus fixing the stuttering. In the article, the first Vertex he had was maxed for sustained write and exhibited stuttering despite having lots of cache.
Firmware is, by definition, the software that physically sits on a device. In this case, this means the controller and any OCZ software put together.wojtek wrote: As far as I understand this technology, stuttering problem was solved by adding cache and Indilinx controller. Firmware is more about right balance between I/O, read, write etc.
-
- Friend of SPCR
- Posts: 454
- Joined: Tue Aug 02, 2005 3:32 am
- Location: SF Bay Area, California
Question - does anyone know / has anyone seen benchmarks for OCZ Vertex drives vs. the new Samsung 256Gb ones? I could swear that I saw benchmarks for a Samsung-based 256Gb SSD (rebranded by one of the memory companies) in the Anandtech article, but I couldn't find it when looking for it a second ago... Hmm...
I'm thinking of getting a couple 250Gb SSD's. My first choice was the Samsung 256Gb drives, and alrightdeals.com has them for a little over $700 after shipping with an ETA of later this week. But I see that 250Gb Vertex drives are just a little cheaper from tigerdirect, and they're in stock now. So, I can wait until later this week if the Samsungs should be better, but should otherwise cancel my pre-order and go with the Vertex drives.
By the way, note that the "Samsung SLC" mentioned in the Anandtech article has very little to do with the new 256Gb drives, which are MLC and have new controllers.
I'm thinking of getting a couple 250Gb SSD's. My first choice was the Samsung 256Gb drives, and alrightdeals.com has them for a little over $700 after shipping with an ETA of later this week. But I see that 250Gb Vertex drives are just a little cheaper from tigerdirect, and they're in stock now. So, I can wait until later this week if the Samsungs should be better, but should otherwise cancel my pre-order and go with the Vertex drives.
By the way, note that the "Samsung SLC" mentioned in the Anandtech article has very little to do with the new 256Gb drives, which are MLC and have new controllers.
-
- Friend of SPCR
- Posts: 454
- Joined: Tue Aug 02, 2005 3:32 am
- Location: SF Bay Area, California
Still didn't find any real data, but I think the Samsungs are theoretically nicer. I believe they have more cache, and I'm betting the controller isn't half bad.shleepy wrote:Question - does anyone know / has anyone seen benchmarks for OCZ Vertex drives vs. the new Samsung 256Gb ones? I could swear that I saw benchmarks for a Samsung-based 256Gb SSD (rebranded by one of the memory companies) in the Anandtech article, but I couldn't find it when looking for it a second ago... Hmm...
I'm thinking of getting a couple 250Gb SSD's. My first choice was the Samsung 256Gb drives, and alrightdeals.com has them for a little over $700 after shipping with an ETA of later this week. But I see that 250Gb Vertex drives are just a little cheaper from tigerdirect, and they're in stock now. So, I can wait until later this week if the Samsungs should be better, but should otherwise cancel my pre-order and go with the Vertex drives.
By the way, note that the "Samsung SLC" mentioned in the Anandtech article has very little to do with the new 256Gb drives, which are MLC and have new controllers.
Besides, it's a moot point because the Samsungs are shipping as of today, apparently (got an email with a tracking number from the vendor). These will be my "storage" drives, while my boot drive will be 3 RAID0'd Intel X25-M's. Boy, I'm going to have some fast, silent storage!
-
- Posts: 78
- Joined: Mon Mar 16, 2009 2:06 pm
- Location: East Midlands, UK
I'm very jealous!shleepy wrote:Besides, it's a moot point because the Samsungs are shipping as of today, apparently (got an email with a tracking number from the vendor). These will be my "storage" drives, while my boot drive will be 3 RAID0'd Intel X25-M's. Boy, I'm going to have some fast, silent storage!
Hello everybody. I've been reading this wonderful site for years (and learnt a lot), yet this is my first post.
Don't you think 3 intel SSD in raid will be bottlenecked by the SATA controller itself, and also your data will be unsafe? In my opinion this is completely unnecessary.my boot drive will be 3 RAID0'd Intel X25-M
-
- Friend of SPCR
- Posts: 454
- Joined: Tue Aug 02, 2005 3:32 am
- Location: SF Bay Area, California
Depends on the controller, doesn't it? I think it should be fine. I'd go with just a couple of them, but I'd want more space than 160Gb (and the 160Gb version is more $/Gb). As for unsafe data, that doesn't worry me a bit, especially with SSD's. Besides, any data worth backing up will be on other drives... The RAID'ed Intels will just be for the OS, games, and any other intensive programs.zborbas wrote:Don't you think 3 intel SSD in raid will be bottlenecked by the SATA controller itself, and also your data will be unsafe? In my opinion this is completely unnecessary.my boot drive will be 3 RAID0'd Intel X25-M
As for it being unnecessary - well, of course. "Necessary" is a bit of a subjective term, though. Based on my current setup, I use around 200Gb of space for the boot drive. I wanted to get the fastest possible SSD with a reasonable size and not too ridiculous price in its place and avoid the crappy cheap ones, so X25-M it was.
-
- Friend of SPCR
- Posts: 454
- Joined: Tue Aug 02, 2005 3:32 am
- Location: SF Bay Area, California
I believe that ICH10R's RAID bandwidth limit is around 600-650 (based on this forum thread, for example).zborbas wrote:I thought you would use an ordinary on-board SATA controller, with a max bandwith of 300 MB/sec.Depends on the controller, doesn't it?
I also had a couple of Areca ARC-1210's which would probably work quite well (had them as server pulls). But after trying one out and realizing how much it slows down boot-up time and whatnot, I decided against doing that. If I find a cheap-ish option that does not add the additional hassles and would improve performance, I might buy it. But for now, I think that I'll be fine with the onboard.
By the way - welcome to SPCR!
What a great article! It sounds like confirmation that SSD is the future, but it won't overtake magnetic platters for at least a few more years most likely, which is what everyone had figured anyway. Sounds like that "TRIM" command is also the next big thing for SSD performance and extension of life. Also it seems that Intel really is best, even with all the work going into the OCZ drives, they only really kept up in 3 out of 4 categories of performance - the random writes were still 1/10th as good as the Intel drives, but even at that, it was still 50% higher than the VelicoRaptor, hence why SSD is still an attractive option, at least in the case of the Intel or OCZ's summit/vertex series.
There's an update to the article, in the mean time there were at least two firmware updates.
One was good performance-wise, but buggy, resulting in data loss, the other fixed that, and kept the extra performance.
Link to Anandtech article: http://www.anandtech.com/storage/showdoc.aspx?i=3535
I wonder how can you find out which firmware revision you'll get BEFORE buying the drive?
One was good performance-wise, but buggy, resulting in data loss, the other fixed that, and kept the extra performance.
Link to Anandtech article: http://www.anandtech.com/storage/showdoc.aspx?i=3535
I wonder how can you find out which firmware revision you'll get BEFORE buying the drive?