LSI 9211-8i versus LSI 9260-8i versus Areca 1231ML-2G

[Computurd]

once you load the game, after you actually start loading it, hit the tilde button. just once. it is this button ~ but you dont have to hit shift.
i would recommend some more targeted results for sequential reads, maybe at a QD of 4 and 8 as well, because they can have an impact on load times etc under certain scenarios. for instance anandtech suggests that 69 percent of reads during gaming are sequential. so it has more impact than just file transfer application..
i think everyone here is interested in gaming performance in particular. so not only 4k random is going to come into play.

Only 20% of the accesses are 4KB in size, nearly 40% are 64KB and 20% are 32KB. A whopping 69% of the IOs are sequential, meaning this is predominantly a sequential read benchmark

some maybe some sequential/random at larger file sizes as well. i would suggest 32 and 64kb as well at some lower queue depths, perhaps 4 6 and 8. i know this only adds to complexity of the benchmark, but really it will only take a few minutes

[SteveRo]

once you load the game, after you actually start loading it, hit the tilde button. just once. it is this button ~ but you dont have to hit shift.
i would recommend some more targeted results for sequential reads, maybe at a QD of 4 and 8 as well, because they can have an impact on load times etc under certain scenarios. for instance anandtech suggests that 69 percent of reads during gaming are sequential. so it has more impact than just file transfer application..
i think everyone here is interested in gaming performance in particular. so not only 4k random is going to come into play.



some maybe some sequential/random at larger file sizes as well. i would suggest 32 and 64kb as well at some lower queue depths, perhaps 4 6 and 8. i know this only adds to complexity of the benchmark, but really it will only take a few minutes

I think I can create an iometer run for this "Only 20% of the accesses are 4KB in size, nearly 40% are 64KB and 20% are 32KB. A whopping 69% of the IOs are sequential, meaning this is predominantly a sequential read benchmark" but what about the other 20% - 128K? - something like this? Run at queue depth of 4, 6 and 8 - no higher?

[SteveRo]

OK so a little teaser of what is yet to come - in the pic below one of these controllers is $233. the other is $770 (both scores are top 10 all time).
I think Areca had better get on the ball -

[Computurd]

LOL I LOVE IT!!!! that is a beautiful result steve-ro! i told ya this little HBA is IT! my testing has confirmed much the same results. i can post tons of file sizees stripes etc etc etc, but they all say the same thing. for 230 bucks you cannot even get close to the 9211. this is a awesome device. it is only marginally slower than the 9260-8i, and for a fraction of the price. the performance is outstanding, to those who claim that it is only as good as ich10r, lol at you! that is the whole premise of my personal foray into the 9260 series, who knows how they perform with 6 gbps devices? will arecas devices be better? maybe. we wont know until we see them. HOWEVER the point is that the processors on these cards are so wonderfully overpowered if you are using them for 3gb/s devices that they just trash it. this is a small tiny little watered down version of the 9260-08i and look at that performance.
i cannot wait to see your results with the 9260-8i.
for your results you are using hte onboard processor for the 9211 right? so this is a bootable array you are testing with?

@back to business...the profiles that you are making have a problem, they will run those file sizes at the same time, correct? that is the problem when you are running mixed files with the iometer. it runs them concurrently instead of consecutively. so, they are getting ran at the same time. they need to run in order is kinda what i am saying, so you cannot run them all at once. you will have to break it up into different tests and run them separately. one hertz already provided a similar test to try to mirror it but that is the same problem that we ran into with it. also you cannot run them partially sequential, partially random at the same time either. that is also why the actual loading times are gonna come into play.


LSI's project name for the 9211 is "The Falcon", very fitting id say!

[SteveRo]

All of my vantage runs are with vantage run from the boot drive (with the array on the controller as the boot drive) - no softraids (yet).
I haven't figured out how to use a softraid for vantage - I suspect it can be done though.
The best pcmark vantage I could get from the 9260-8i was 20636 - respectable but not keeping up with the other two (so far).

[Nizzen]

OK so a little teaser of what is yet to come - in the pic below one of these controllers is $233. the other is $770 (both scores are top 10 all time).
I think Areca had better get on the ball -

I want to see the numbers from the Vantage test

[Computurd]

excellent work this is awesome stuff, makes me wanna get some acards! i cant wait to see your game load times as well~!
i would like to see you results with the vantage run as well.

[SteveRo]

I want to see the numbers from the Vantage test

Merry Christmas Mr Nizzen,

I still have a long way to go to match your score but these are pretty good - particularly the 9211-8i (Provantage - $233.) -

http://service.futuremark.com/compare?pcmv=259094

The 1231ML run is here - http://service.futuremark.com/compare?pcmv=259105

[Nizzen]

Merry Christmas Mr Nizzen,

I still have a long way to go to match your score but these are pretty good - particularly the 9211-8i (Provantage - $233.) -

http://service.futuremark.com/compare?pcmv=259094

The 1231ML run is here - http://service.futuremark.com/compare?pcmv=259105

Merry Christmas SteveRo! Thank you very mutch!

PS: It looks like areca 1231ml suck in the harddiksscores compare to mine

Merry Christmas NapalmV5

[SteveRo]

I think I can create an iometer run for this "Only 20% of the accesses are 4KB in size, nearly 40% are 64KB and 20% are 32KB. A whopping 69% of the IOs are sequential, meaning this is predominantly a sequential read benchmark" but what about the other 20% - 128K? - something like this? Run at queue depth of 4, 6 and 8 - no higher?

Good afternon again Turd, make up an iometer .icf for what you want and when the site is back up - post it at -

http://dump.no/

I will run it this next week.

[SteveRo]

Merry Christmas SteveRo! Thank you very mutch!

PS: It looks like areca 1231ml suck in the harddiksscores compare to mine

Merry Christmas NapalmV5

Nope, it's not the 1231ML - it's the short legs of the acards vs the vertex in seq xfer's - compare yours to the #4 run - you will see he beats you (with the 1231) in many disk tests -

4. 23690 dj_ski69 - OCZ Vertex RAID 0

Vantage rewards high throughput much more than small file iops.

[Nizzen]

Nope, it's not the 1231ML - it's the short legs of the acards vs the vertex in seq xfer's - compare yours to the #4 run - you will see he beats you (with the 1231) in many disk tests -

4. 23690 dj_ski69 - OCZ Vertex RAID 0

Vantage rewards high throughput much more than small file iops.

Cool

Then I have to try the 1231ml soon too

But it is no proof of using the Areca 1231ml in vantage. You can use whatever you want it looks like. The "Hard drive model" is just the first boot device on the computer

[SteveRo]

Cool

Then I have to try the 1231ml soon too

But it is no proof of using the Areca 1231ml in vantage. You can use whatever you want it looks like. The "Hard drive model" is just the first boot device on the computer

The 1231 should be faster vs the 1680 for your vertex array - but both the 1680 and the 1231 should be overtaken by one or more of the new pcie 2.0 controllers in the next month or two.
The 9211 is very close even now and I suspect the 9211 FW (when compared to the 9260) could be improved significantly yeilding a significant performance boost.
It will be hard to beat the 9211 based on value.

[Computurd]

investing money in a 1231 right now is not a good idea. once areca releases 18xx series the value will go down significantly, and all the newer faster devices are going to be 6gbp/s, the next lines of ssd will not be 3gb/s. not to mention as ssteve is showing, a HBA for the 6gb/s is matching the 1231 except it has NO CACHE. so that means the processor is exponenetially faster, and that is three steps down on lsi's processor chart.

[HDCHOPPER]

Exellent testing this helps all I see a 9211 in my future

[Computurd]

yes right now if someone were to ask me which controller to buy it will be the 9211, hands down. BUT not the 9211-4i, it has a x4 bus, while the 9211-8i has a x8 bus. there will be those that say that the x4 bus wont matter, but they are stuck on the 3gb/s generation thinking. future devices with 6gb/s implementation will be able to saturate that x4 lane. i wonder why in the hell they did that on the -4i but oh well.
9211-8i it is!!! no one will come close to matching it on price v performance, and the best solutions out there will only beat it by a very very slim margin. the 1231 only beat it by 2 percent!

[lowfat]

I need to see some games!

[SteveRo]

yes right now if someone were to ask me which controller to buy it will be the 9211, hands down. BUT not the 9211-4i, it has a x4 bus, while the 9211-8i has a x8 bus. there will be those that say that the x4 bus wont matter, but they are stuck on the 3gb/s generation thinking. future devices with 6gb/s implementation will be able to saturate that x4 lane. i wonder why in the hell they did that on the -4i but oh well.
9211-8i it is!!! no one will come close to matching it on price v performance, and the best solutions out there will only beat it by a very very slim margin. the 1231 only beat it by 2 percent!

If LSI was to fix the 9211 FW for large seq xfers, add a monster cache (4GB) to the card and keep latency low ... wow.

[SteveRo]

If LSI was to fix the 9211 FW for large seq xfers, add a monster cache (4GB) to the card and keep latency low ... wow.

They need to fix this -

[Computurd]

yup over 128 file size it starts to run into problems, but realistically how many users will need performance at those file sizes? some might, but for others...it will be interesting. however, i think file sizes that large are not that large of an indicator of the real world performance. the sequential at and under 128 do look good though, but not near what the 9260 pumps out
i think cache is really only needed for devices with write issues. this proves that you can get performance without the large caches.

[One_Hertz]

I need to see some games!

+1. So far 1231>9211>>>9260 but thats just vantage. ICH?

[Computurd]

i am also very curious of the 1231 4k, 64k and 32k sequentials...oh we must be patient stevo-o is working!
im busy too though!

[SteveRo]

yup over 128 file size it starts to run into problems, but realistically how many users will need performance at those file sizes? some might, but for others...it will be interesting. however, i think file sizes that large are not that large of an indicator of the real world performance. the sequential at and under 128 do look good though, but not near what the 9260 pumps out
i think cache is really only needed for devices with write issues. this proves that you can get performance without the large caches.

Yep - the 9260 does not seem have the 9211's large seq xfer problem -

[Computurd]

9260-8i results for sequentials....

[SteveRo]

Turd - how many vertex do you have in your array - are they scaling in seq reads as they should?
For me - each acard gets approx 175 MB/s in seq read, so for 8 - if perfect scaling I would have 1400 MB/s but the best I have seen is 1125 MB/s, what are you seeing?