LSI 9211-8i

[Computurd]

I have been doing alot of testing with the LSI 9260-8i and believe me, I love it!
I feel very deeply that it is probably the fastest pci-e raid card solution available today. of course some disagree....
However, recently LSI has released the 9211-8i, and upon doing a little research and comparison, I came to the conclusion that this card is merely the 9210-8i, but with a different port configuration. For people who have read the very popular toms hardware article, this controller was also featured in its testing model. Not available to the general public at the time, even thought the 9260-8i was. (I had the LSI for months before Tom's Hardware got ahold of it BTW)
Note the very impressive speed and characteristics of the 9211 compared to the full featured 9260 in this link...http://www.tomshardware.com/reviews/...id,2388-5.html
The controller listed as the 9210 on there is the one i suspected of being the same as the 9211-8i. So i checked the egg and this controller is very cheap,, very reasonably priced.http://www.newegg.com/Product/Produc...-112-_-Product
so considering that a top flight raid card can set ya back upwards of 600 USD, $250 is a paltry amount LOL. I have ordered it and am having it overnighted here, and after checking its status tonight i am confident i will have it early tomorrow for testing. The reason this card is faster i believe is because it has a stripped down raid set, basically there is no raid overhead, it is just a straight shot to the board, with no cache overhead either. it does not do raid 5 or 6, only a very limited raid set.
If it is what i think it is, its basically a hotrod! Should be fun to toy with, but not have as much of an impressive feature set as the 9260-xx versions. However, for those guys looking to raid up some cards on the cheap, and get speeds WELL in excess of the ICH10R speeds this should be a great entry level/cheap product. the fact that it is SAS/SATA 6GB/S compatible should give it plenty of forward mobility, yet also with the processing power to absolutely monster the high bandwidth of current 3gb/s drives. After bouncing some ideas off of a few other 9260 owners, a few of us have gotten this controller for testing, cant wait to get started! The greek connection (tiltevros) has confirmed with LSI support that this is in fact the same controller as the one in the review, so kudos to him for gettin that squared away! should be interesting to see how this shakes out!
My testbed will be an 8 drive array consisting of OCZ vertex.
tiltevros testbed is going to have upwards of six intel SSD,
Biker will be testing also with his uber array, consisting of agility drives, so we should have all the bases covered, hopefully its good, but either way it should be fun!!
Here is some more info on the controller for those interested...
http://www.lsi.com/DistributionSyste..._PB_092409.pdf

NOTE the 290,000 IOPS potential

http://www.lsi.com/DistributionSyste..._PB_092909.pdf

and yes this device is bootable, we shall see how it pans out!

[XS Janus]

hehehe Oh yea!
no cache though...

[Tiltevros]

some extra news from LSI

Search for the HBA was ended when LSI showed their 9210-8i series (Code named as Falcon) which has been rated to perform 300K IOPS. These are entry level HBA's which can be configured to hook up up to Eight drives to eight Internal ports.

the result for hooking up 5 9210-8i is http://communities.intel.com/communi...lion-iops-game

[Tiltevros]

hahahahahaahah bring me an 8 PCIEx16 motherboard and we will have a RECORD :

[Computurd]

LOL yes, no cache janus, however with most applications for SSD usage now you really do not need one. I get the best results, as do others, with direct I/O with the 9260-8i. This also explains some of the move to the pci-e based ssd's, most most of which have very small caches. the card relies on the drives having caches of their own, the vertex has cache and so do intels, and since its raid, you can actually already have a large storage pool of cache that is better suited for the ssd's themselves already at their disposal. that is one of the issues facing raid cards and SSD's imo. the devices themselves already have cache and then you have the cache from the raid card as well and this introduces latencies.

[Tiltevros]

i dont know what to say..... JUST MASSIVE!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

[Nizzen]

LSI 9211-8i is not a controller but a Host Bus Adapter :p

I have 2 of the old sas host bus adapters. LSi 3000 series Just using them in a fileserver with Seagate 15k drives.

But we want numbers here, not blablalbalablabl...

[Tiltevros]

the first bench will be in 30 min. just post what program or bench do u like to use

[XS Janus]

Can't wait to see how these will far in the latency department... hopefully closer to the AR1231 as they can be...
Waiting for the first results! 3 more min
Like we need to tell you guys what we need.
CrystalDisk 2.2 or 3.0b
As Disk and its copy benchmark
HD tune Pro, reads AND writes please!
iometer numbers

-and some real world stuff, like copy-paste a large number of files or iso to see how long it takes.
installing a game or two from .iso residing on the array, and last, maybe some level loading times just for good measure Compared to your "old" setup or single SSD setup

[Tiltevros]

i cant do write test in hd tune cause its my boot up disk

[Computurd]

I will be doing very limited write testing myself. I hate to beat on the array too much. I know what to post! the card is on the bench now, we shall see very soon.
@nizzen- Yes, i know its an HBA, but thats splitting hairs

[Nizzen]

Where are the numbers?

Was it so disappointed that it not going to be posted? Was it epic fail?

[SteveRo]

status? ...

[Computurd]

average read access measured in everest is .08 so that is a definite improvement. new device, testing takes time....there is an absolute lack of materials for this device, so some is learning curve...

[One_Hertz]

average read access measured in everest is .08

2-3x improvement? Definitely looking forward to more results.

[Computurd]

no the 9260 is getting .14 to .16 in measurement, so it is closer to 2x improvement

[One_Hertz]

no the 9260 is getting .14 to .16 in measurement, so it is closer to 2x improvement

Another way to look at it is consider that vertex access time is 0.08-0.09 to begin with so the 9211 is adding nothing (or near nothing) to access time while the 9260 adds 0.08

Please do some real world comparisons - anything - between the two controllers.

[XS Janus]

Take your time, please. Leave as little questions as possible open for forum quarrels.
Thanks for the update. .08 looks pretty great.
Good night!

[Computurd]

preliminary results indicate like small file performance, with nearly half the access time of the 9260-8i. the only area that suffers is the tall sequentials, and that is because of the wrench in the plans so to speak. the stripe size is set at 64k and there is no modifying it. there is more work to be done but this is the gist of what we have found. ihave achieved over 300 mb/s in 4k random at half the access time of the 9260-8i. i will post real world comparisons later. boot time is faster but not by much.
the real thing is some weird results, for some reason 4k random is faster than 4k sequential. something is going on there.

[Computurd]

okay some more preliminary results here, i think something is wrong with the settings, we are ironing it out as we go. however, to settle a previous debate i ordered l4d last week to test load times in relation to the 9260 and the i.o extreme. this link shows the load times of some pretty high profile high cost devices.http://hothardware.com/Articles/Fusi...eview/?page=12
i have beat those load times with the 9211-8i. i really wish this game had came in so that i could test on the 9260-8i but i will put it back on later,. and see what happens
my load time for dead air is 5.6 seconds
load time for no mercy is 6 at conservative 4.0 overclock settings, and my drives need to let garbage collection run, so there will be more results shortly, after i do these...i will post youtube videos for nonbelievers this is without much tweaking so i can guarantee you we will be seeing some much faster results!

[One_Hertz]

damn still a bit slower than napalm's areca 1231 with crappy jmicrons. Can't really complain with a $250 tag I suppose. Nice results nonetheless.

A single X25-M will load those levels in roughly 8 seconds (subsequent loadups are roughly 3-4s) when hooked up to a fast machine (4.7 qx9650).

No need for youtube, that result seems reasonable. I don't quite understand how the areca is faster than the 9211 though considering access times are the same and the 9211 is supposed to have higher max bandwidth.

[Computurd]

lol one hertz as i said, still figuring it out 300mb/s at 4k random with access times of .08 is not bad at all also i haven't taken the time to vlite my os and tear it down basically running an unusable version of xp to inflate scores....lol. i am measuring performance on a fully functional OS that wasnt created what ten years ago?
funny how it is still faster than the i/o extreme LOL@@@ funny that in their testing hothardware couldnt get their x-25s to match the performance you are speaking of.
and the only thing the areca 1231 is faster than is the other old 3gbps controllers, i can guarantee you it will not best either the 9260 or the 9211
i have gotten them to load in approx five seconds now on the first load, at a setting of 4.2. so that is roughly a 37 percent improvement over the intels that you are speaking of, and an two fold performance of the ones they tested in the review, but running at 4.2 instead of 4.7.
getting into the area of diminishing returns of course. now that you get to these speeds you are pushing the limits of bandwidth that is even usable of course. the 9211 does not have the large file performance however, that the 9260 does. it will not hit tall sequential that are near the 9260. however, this is virgin silicon here, with much room to grow. also there is the limitation on stripe size which is particularly depressing

[One_Hertz]

lol one hertz as i said, still figuring it out 300mb/s at 4k random with access times of .08 is not bad at all also i haven't taken the time to vlite my os and tear it down basically running an unusable version of xp to inflate scores....lol. i am measuring performance on a fully functional OS that wasnt created what ten years ago?
funny how it is still faster than the i/o extreme LOL@@@ funny that in their testing hothardware couldnt get their x-25s to match the performance you are speaking of.
and the only thing the areca 1231 is faster than is the other old 3gbps controllers, i can guarantee you it will not best either the 9260 or the 9211

tearing down the OS doesn't change much. Maybe 0.1-0.2s faster. Napalm and I run a similar config. Very minimalistic, but runs everything. I do it because if you remove a bunch of stuff from your OS like cmd and internet explorer core, it is much more difficult for viruses and malware to stick to your PC.

Hothardware is using a 2.66ghz cpu... Can't quite compare to them.

Get faster than 4.5s then talk about the areca .

[Computurd]

hey one hertz, i have an idea. you should prove to us your loading time so that we have something to compare to. you yourself should put up a video of your load times. maybe some kind of storage manager so that we can see the devices used. maybe do it on a 24.7 system without ln2?

prove it!

[Tiltevros]

benches with 6 x25m will be post here soon so we will have a better look of the HBA card no reason to fight about it.. we have G1 and G2 intels so ... just wait to find out the best setup for this professional card