Sign in to follow this  
imsabbel

Intel X25-M

Recommended Posts

There is a review of Intels new SSD tech online at:

http://techreport.com/articles.x/15433

Seems like they werent kidding. While the low write STR hurts in a lot of benches, random IO, all kinds of reads and the power consumption looks rather well.

For an MLC-drive, even sensational...

P.S.

I nearly forgot: Seems like we need 6GBps SATA faster than anticipated :D

Edited by imsabbel

Share this post


Link to post
Share on other sites
Too bad write performance, especially random writes, still blows. :(

Um, what makes you think that?

How about over 10,000 random 4k writes per second:

http://www.anandtech.com/cpuchipsets/intel...?i=3403&p=8

Note, that the random writes are EXACTLY as fast as sequential writes on the Intel SSD. ~0.09 ms per 4k write, random or not.

I have already detailed exactly how this can be on this message board. All writes are sequential, there simply IS NO RANDOM. Blocks are written in order, and that 16MB ram chip it has keeps track of that order. The flash blocks themselves persist the order, there are extra header bytes per block in flash memory for such data and ECC.

In ALL flash drives with wear leveling, the logical block addresses are virtualized over the physical ones. The next generation controller chips just go _all the way_ and don't just do simple wear rotations.

Here is a white paper I discovered last week that talks about these sorts of techniques -- the sort that I pointed out here last month:

http://research.microsoft.com/users/vijaya...sd-usenix08.pdf

http://forums.storagereview.net/index.php?...st&p=252341 (and the followups later -that thread before is full of bad information).

One of the other contentions about complex wear leveling schemes that is completely false is that as such a drive gets closer to being full, its performance will get a lot worse because there will be less 'space' for the wear leveling to work in. This is based on an assumption that the wear leveling schemes can somehow tell how full the disk is. No block device has any clue what is on it, its not a file system. By definition, it is ALWAYS 100% full from its perspective. However, any time you write , you displace the addresses you wrote over and "free" the overwritten bits. If every transaction has an id (ever erase block being filled), then one only has to find the blocks mapped to the same address more than once, and on such duplicated addresses only the most recent transaction is valid. This info is kept real-time, tracking all the blocks and their status is what that 16MB of RAM on the Intel SSD. (and the similar RAM bits on some of the older SLC drives out there serve a similar purpose -- and that is the main reason why they have faster random writes!).

Wear leveling performance will get worse as more and more bad erase blocks get removed years down the road, and there is less overprovisioned flash to go around for the pre-erasing. Also, certain workloads will cause the algorithms to have to do more compaction, which is less efficient.

Note that ordinary hard drives also virtualize the logical LBA's to physical platter positions -- how else do you think they deal with bad blocks? 99.9%+ of the LBA's occur in sequence on the drive, but not all of them.

Share this post


Link to post
Share on other sites
There is a review of Intels new SSD tech online at:

http://techreport.com/articles.x/15433

Seems like they werent kidding. While the low write STR hurts in a lot of benches, random IO, all kinds of reads and the power consumption looks rather well.

For an MLC-drive, even sensational...

P.S.

I nearly forgot: Seems like we need 6GBps SATA faster than anticipated :D

Another good review: http://www.anandtech.com/cpuchipsets/intel...doc.aspx?i=3403

Share this post


Link to post
Share on other sites

Sorry, I missed the Anandtech article.. thanks for pointing that out-- I just hit TR's article yesterday, which showed X25-M performance being much closer with other SSD's.

That said, I'm happy fellowing impressed. :D

Unfortunately or not I am (well, my customers in question) in the 0.1% or whatnot that hammers drives with small writes day in and day out so SSD lifetime is still an issue for us, although the performance increase is bloody nice... Ah well. And killing a $1k flash drive in a couple months is better than killing a $3k one... and if Intel's performance really does add up we might get a long couple months instead of a short.. hmm...

Share this post


Link to post
Share on other sites

Any guesses as to pricing on the SLC models? Is the 32 GB X25-E model going to be more expensive than the 80GB X25-M?

I'm only looking for something to hold OS and favourite apps, so 32 GB should be plenty, but I suspect it's not going to matter that Intel will have these out within 6 months, it's going to be a couple of years before I can consider one affordable!

Having read the Anandtech article, I wonder what will happen when all the other MLC SSDs get new controllers? Hopefully no more 1 second pauses, but will Intel still be streets ahead?

Maybe we should all chip in a to a "buy SR an X25-M" fund, and one lucky contributor drawn at random gets the drive when Eugene's done with it!

Share this post


Link to post
Share on other sites

Ah yeah.

Well, i guess that SLC version will be significantly faster yet, and be value-based in its pricing (meaning if it can replace 8 15k scsi disks in random IOs, it will cost 2k$+...).

And yeah, SSDs will become a really hot topic, with all the headroom thats still available in controller /caching /tiered storage tech.

I really hope at some point SR will take care of this new field...

Share this post


Link to post
Share on other sites
Any guesses as to pricing on the SLC models? Is the 32 GB X25-E model going to be more expensive than the 80GB X25-M?

According to Anand speculation the 60gb should have a similar price to the 80gb MLC.

I'm skeptical.

Anyway I need 100+gb for OS and documents, which means 4x the 32gb model or 2x 64 gb.

Share this post


Link to post
Share on other sites
I would like to see a raid0 test with ichr9 and areca cards

Extremetech has raid-0 testing.

http://www.extremetech.com/article2/0,2845,2329593,00.asp

They list P5Q Deluxe mobo (which is p45 chipset and has ICHR10), but say p965 chipset (ICHR8) so there is an error/typo somewhere...

STR doesn't scale very well (as measured by HD TACH), but performance according to other benchmarks scales better. Another example why measuring STR is not particularly useful (unless your usage patterns are actually dependent on STR)

Edited by poisondeathray

Share this post


Link to post
Share on other sites
I would like to see a raid0 test with ichr9 and areca cards

Extremetech has raid-0 testing.

http://www.extremetech.com/article2/0,2845,2329593,00.asp

They list P5Q Deluxe mobo (which is p45 chipset and has ICHR10), but say p965 chipset (ICHR8) so there is an error/typo somewhere...

STR doesn't scale very well (as measured by HD TACH), but performance according to other benchmarks scales better. Another example why measuring STR is not particularly useful (unless your usage patterns are actually dependent on STR)

Considering those drives can get >200Mbyte/s even in semi random reads (anand had several application test patters that were >>100Mbyte/s), i would guess that current motherboards arent really made to cope with the amount of IO that 2 or more in Raid0 would produce (For example if the SATA-Controller is an internal PCI-Express 1 device...)

Share this post


Link to post
Share on other sites

I don't know about you, but i think 3sec of CS3 is pretty awesome. I get around 6-7 seconds on my ageing Fujitsu 15k MAU single drive (RAID-0 with 128mb cache didn't really help much), not even velociraptor can top that, but once this comes down to like $300, i'm getting 3.

Share this post


Link to post
Share on other sites
I don't know about you, but i think 3sec of CS3 is pretty awesome. I get around 6-7 seconds on my ageing Fujitsu 15k MAU single drive (RAID-0 with 128mb cache didn't really help much), not even velociraptor can top that, but once this comes down to like $300, i'm getting 3.

I get 4-5s and I dont see 3s as a huge improvement. Thats why i need the SLC units.

Share this post


Link to post
Share on other sites
I don't know about you, but i think 3sec of CS3 is pretty awesome. I get around 6-7 seconds on my ageing Fujitsu 15k MAU single drive (RAID-0 with 128mb cache didn't really help much), not even velociraptor can top that, but once this comes down to like $300, i'm getting 3.

I get 4-5s and I dont see 3s as a huge improvement. Thats why i need the SLC units.

So that you can save 1 second more per day for when you start up photoshop the first time?

Do you need the performance, or are you just looking for a way to waste money?

Share this post


Link to post
Share on other sites
So that you can save 1 second more per day for when you start up photoshop the first time?

Do you need the performance, or are you just looking for a way to waste money?

No, I dont even start photoshop everyday ;)

I'm looking for an upgrade that would be worth it, albeit expensive.

Word documents and big pictures (with vista viewer) are things that I do several times a day and where I would welcome a two/threefold improvement.

Share this post


Link to post
Share on other sites

Not getting it for my desktop, but my laptop which takes 12-20 seconds for CS3 and i use the entire suite, sometimes i dread opening them. I usually have 3-4 of them open along with 4 browsers and about 50 tabs each, so memory gets a bit low and even worst GDI resources and handles get used up. Its the reason people open notepad to write something simple rather than word, which is so much better and secure, its the speed in opening them. imsabbel, i don't know where you went to school, but 3 - 6 to 7 does not equal 1. This thing is twice as fast as the next fastest thing out there. That's nothing to sneeze at. This is pretty much the SSD revolution everyone was expecting when SSD drives were released but ultimately failed disappointed. I wonder how well it can pass the infamous 9 thread test.

Edited by HexiumVII

Share this post


Link to post
Share on other sites
imsabbel, i don't know where you went to school, but 3 - 6 to 7 does not equal 1.

Well, at least where i got to school i learned how to read, mr. deficiency.

Take a closer look at the post i replied to.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this