Or just add the info you have now to Bikers list at the link below...
http://www.xtremesystems.org/forums/...d.php?t=230922
Or just add the info you have now to Bikers list at the link below...
http://www.xtremesystems.org/forums/...d.php?t=230922
"High Quality is never an accident. It is always the result of excellent intention, intelligent direction, sincere effort and skillful execution. It represents the wisest choice of many alternatives"
System Specs:
R3E + custom 1601 bios / XEON X5650 6-Cores @ 4ghz / G-Skill RipJaws 12GB Kit / Asus Radeon HD EAH5870 / Crucial C300 x 2 R0 + 2 x Samsung 840 Pro's on ICH / Logitech Z-5500 5.1 / Ultra X3 1000w PSU
Water Cooling by:
Swiftech Apogee GTZ / EK FC5870 Acetal+Copper / Swiftech MCP655 / ThermoChill PA120.3 / 6 Gentle Typhoons AP-15 / XSPC Dual Bay Res
PrimoFlex Pro LRT UV Green 1/2" / XSPC Black Chrome Compression Fittings
Misc:
SilverStone TJ07 / 24" Dell UltraSharp / Yadda Yadda lol...
tilt those are some awesome results i cant match that with the controller...your latency on there is killer!ok here is something to see... LSI 9211i8 windows 2008 drivers and 7 intels-m
there is a 20% off because of overhead i dont know how to make it lower but this is the best that my mind can think
the first on is 256k file seq. on 32outstandings IOS
the second one is 4k file seq 1 oustanding ios
PCIE latance at 89 (best result) and PCIE packet @ 128
this is the best result that i can take from the 5520 chipset
"Lurking" Since 1977
![]()
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
Good morning and happy new year!
Game level load times by controller - below.
Results are likely to vary greatly from one system to another - there so many variables: disk array used, processor, mobo, memory, os, apps that the os decides it needs to run without asking.
Also keep in mind these results are for only one game - other games may provide totally different results.
Test configuration:
i7 965 (stock clock/ no oc)
Gigabyte x58 Extreme mobo
6GB memory at 1066
Disk array used was 8xAcard 9010 all set to Raid 0 (so 8xAcardR0) except for Ich10R which is 6xAcardR0 (mobo max).
Comments/summary/opinions: If these results are representative (a big IF) and could be generally applied to other systems and if the primary use of your system is Crysis gaming - stick with ich10R.
I suspect that Crysis level loading might be bottlenecked elsewhere - perhaps processing?
Different games might provide a totally different result.
![]()
Last edited by SteveRo; 01-01-2010 at 06:35 AM.
<deleted>
MainGamer PC----Intel Core i7 - 6GB Corsair 1600 DDR3 - Foxconn Bloodrage - ATI 6950 Modded - Areca 1880ix-12 - 2 x 120GB G.Skill Phoenix SSD - 2 x 80GB Intel G2 - Lian LI PCA05 - Seasonic M12D 850W PSU
MovieBox----Intel E8400 - 2x 4GB OCZ 800 DDR2 - Asus P5Q Deluxe - Nvidia GTS 250 - 2x30GB OCZ Vertex - 40GB Intel X25-V - 60GB OCZ Agility- Lian LI PCA05 - Corsair 620W PSU
in server motherboard is everything in manual mode... automatic or optimum is not an option... so u change the setings u boot u test and do the same thing again and again until u will find the golden cut. pci latance is the time that u r giving to ur PCIE lane to stick to the card and then change lane. some card need more latance to boost performance like LSI. the thing with packet size it that if u put a big number there like 256 then u chooke the card.
"High Quality is never an accident. It is always the result of excellent intention, intelligent direction, sincere effort and skillful execution. It represents the wisest choice of many alternatives"
System Specs:
R3E + custom 1601 bios / XEON X5650 6-Cores @ 4ghz / G-Skill RipJaws 12GB Kit / Asus Radeon HD EAH5870 / Crucial C300 x 2 R0 + 2 x Samsung 840 Pro's on ICH / Logitech Z-5500 5.1 / Ultra X3 1000w PSU
Water Cooling by:
Swiftech Apogee GTZ / EK FC5870 Acetal+Copper / Swiftech MCP655 / ThermoChill PA120.3 / 6 Gentle Typhoons AP-15 / XSPC Dual Bay Res
PrimoFlex Pro LRT UV Green 1/2" / XSPC Black Chrome Compression Fittings
Misc:
SilverStone TJ07 / 24" Dell UltraSharp / Yadda Yadda lol...
Good afternoon and happy new year again,
Iometer results for the controller compare are below.
Remember this is on my setup – YMMV!
Some of the more interesting numbers are color coded red and blue.
Observations –
The LSI 9260-8i:
Performs much worse then the other controllers at small file random io – both read and write. I used the latest f/w and driver.
The LSI 9211-8i:
4KB sequential read is rather low at 215 MBps compared to the other controllers – including Ich10R (Ich10R with not 8 but 6 drives).
Seqential read at 128K chuck size is very good – 1318MBps compared to approx 1400 if perfect scaling was possible.
All large file xfers at 1MB and higher chuck size falls off significantly – how much (if any) xfers happen at this chunk size?
The Areca 1231ML-2G:
With the exception of the 9260, the 1231ML has the slowest 4KB random xfer speed - over 90MBps slower than Ich10R
All large file random reads at 1MB to 4MB size falls off a cliff – how much (if any) xfers happen at this chunk size?
Question:
Can anyone tell looking at these results that the highest scoring PCMark Vantage HDD score would be the Ich10R array (Ich10R with not 8 but 6 drives) and that the highest PCMark Vantage (overall) score would be using the 1231ML? I couldn’t tell.
Pics -
![]()
Last edited by SteveRo; 01-01-2010 at 11:20 AM.
Thanks a bunch Steve. Very interesting!
I noticed that the ICH10R and the 9211 follow the same patterns in benches such as seq. reads and random write. For example in seq reads, they both peak between 16K - 128K files sizes, then drop off quickly. The 9260 and 1231 don't suffer from this. I wonder if this is because of the cache?
With the exception of large files sizes, I wonder why the 9211 does so much better than the 9260?I thought I read somewhere that the raid processor on the 9260 was clocked higher than the 9211.
![]()
MainGamer PC----Intel Core i7 - 6GB Corsair 1600 DDR3 - Foxconn Bloodrage - ATI 6950 Modded - Areca 1880ix-12 - 2 x 120GB G.Skill Phoenix SSD - 2 x 80GB Intel G2 - Lian LI PCA05 - Seasonic M12D 850W PSU
MovieBox----Intel E8400 - 2x 4GB OCZ 800 DDR2 - Asus P5Q Deluxe - Nvidia GTS 250 - 2x30GB OCZ Vertex - 40GB Intel X25-V - 60GB OCZ Agility- Lian LI PCA05 - Corsair 620W PSU
wow steve tons of results there GREAT job man! gonna take awhile to digest all of the numbers, however the game load times is where i want to start....
9620-8i price is roughly 500 bucks. the 1231 is $769 for 2GB and $1050 for 4GB.
in the load times the 1231 outperforms the 9260-8i in game loads slightly, less than a second, and also being last gen and pretty much maxed in firmware and software, i think it is a no-brainer that among the cards the 9260-8i is the best , especially considering it's forward mobility. i believe if you ran that at a 64k stripe or even 16k stripe the 9260-8i would beat the 1231 pretty easily, i have made tremendous gains with those stripe sizes, i am running some different tests for the 9260 at different stripe sizes and will give you numbers in the difference in load times shortly, before the end of the weekend. one of the largest differences is going to be the 16k stripe as it is at the other end of the spectrum from the 1mb of course. i do have numbers for that. basically about .5 sec faster with the 16k than the 1mb. doesnt sound like much but of course when load times are around 3 seconds on consecutive loads a .5 is a big deal.
the 9211's poor showing in game load times did not suprise me much, even though i would hardly call that performance 'poor'. hell for 230 bucks you literally cannot beat it. and considering that it is a virgin firmware AND driver jesus there is only one way to go...up.
the ICH10r is impressive, however i think the 92xx series will surpass it soon with future improvements,. and thats not even begging to mention sata 3 and sas2 devices. they will positively torch the ich10r. i am not surprised that the ich10r does better than the 1231. the 1231 was designed how many years ago? the ich10r is a more developed solution imo. hell it was designed when SSD's were actually on the market. im not sure when the 1231 was designed in 06 and released in 07. not sure.
that does bring into the equation that your results are using acards and not ssd....
one conclusion i can draw from the 9211 results for sure is that considering the 9260's 'poor' 4k random and the 9211's superb 4k random, the 9260-8i beating the 9211 could help prove that the 4k random is not as big of a determining factor as many think, especially in light of the fact that there have been other sources pointing to it as relevant, but not nearly the end-all performance indicator. people read far far too much into 4k.
(i have obtained warhead, last night off of steam, i am going to run it on the different controllers to compare to your load times with acards...i bet you smoke me)
thansk for your hard work and effort man.
Last edited by Computurd; 01-01-2010 at 04:55 PM.
"Lurking" Since 1977
![]()
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
Last but not least – FYI - AS SSD and CDM for all controllers.
Test configuration:
i7 965 (stock clock/ no oc)
Gigabyte x58 Extreme mobo
6GB memory at 1066
Disk array used was 8xAcard 9010 all set to Raid 0 (so 8xAcardR0) except for Ich10R which is 6xAcardR0 (mobo max).
Default controller settings and stripe sizes are used except 1MB stripe is used for the 9260-8i
Ich10R with 6xAcardR0 –
LSI 9260-8i with 8xAcardR0 –
LSI 9211-8i with 8xAcardR0, I could not get AS SSD to run the 4k/QD64 - AS SSD kept hanging –
Areca 1231ML-2G with 8xAcardR0 –
![]()
Well you definitely proved what a lot of us has thought about games. Storage currently isn't the bottleneck.
i have been downloading crysis warhead from steam for several years now....goddamn it is slow.
"Lurking" Since 1977
![]()
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
I'm not sure if I did this correctly however I used the Win 7 Performance Monitor to find out the max MB/s whilst loading the last level of COD MW2. (I selected maximum on the report and histogram within the general tab on the Performance Monitor Properties. I then started the counter as I loaded COD. Once the level had loaded I closed it down and paused the counter.)
Results -
7200RPM HDD
Read Max = 44MB/sec
Load time - approx 10 seconds
Intel G2
Read Max = 96MB/sec
Load time - approx 6 seconds
The read seems to have maxed out on the HDD, but the read on the G2 is well below the theoretical max. Intriguing.
Now check out the AnandTech Storage Bench - Gaming Workload in this review. The Kingston SD Now (JMicron 602 - 170MB/s sequential read) is just as fast at loading as an Intel X25-E (250 MB/s sequential read)
Assuming it takes around 400MB to load a game and the load is predominately sequential reads the difference in my loading times between the hdd and ssd is close.
The question is why was the ssd read limited to 96mb/s? Neither the cpu or the ram broke into a sweat.
EDIT:
L4D2 1st level.
Intel G2
Read Max = 108MB/sec
Load time - approx 7 to 8 seconds
Edit 1: and if I run AS Benchmark I get the correct read reading on the Performance Monitor so I think I am doing this correctly.
Last edited by Ao1; 01-02-2010 at 06:18 AM.
good question this is why i am wondering how queue depth is assigned, either by os or program. i have been told it is by the program by one individual, however i am not so sure. i am not entirely convinced of that.
IF it is assigned by the operating system, either by part, or in whole, if someone could somehow alter that to where things load automatically at higher queue depths....could be fantastic.
is there a way to alter QD and 'force' the system into doing it?
i think most coding is done for hdd, both at operating system and at program level. this sucks LOL. some applications however do use higher queue depths etc...interesting stuff.
maybe thats why raid cards work so well? delivering higher speeds at lower QD automatically? i dunno. it still would take a lot of convincing to persuade me that the storage aspect has no major involvement in loading times etc...
Last edited by Computurd; 01-02-2010 at 08:18 AM.
"Lurking" Since 1977
![]()
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
Look at the read speeds that SteveRo is achieving. Multiply the read speed (lets say 500MB/s) by the time it takes to load a level (lets say 27seconds) and if everything was working perfectly Crysis would have just loaded a 13GB sequential read file.
I don't know how many MB's load on a typical Crysis level but I doubt its 13GB.
If it was loading @ 100MB/s over 27 seconds it would be 2.9GB and even that seems a bit high. Even if it loaded 1GB that is only 37MB/s.
AnandTech got a load time 38 seconds with a G1 Intel. Depending on which level that was it is close to SteveRo's time on a storage system that is infinitely faster.
good points, but also you have to realize it isn't going to be a sequential load pattern in its entirety. it is going to be doing a mix of randoms and sequentials at different file sizes, so maybe how it is handling that loading has a lot to do with it. but even with a lower expectation as in your second example it still does seem off.
wish we knew what level they were loading with that in anandtech, the crysis levels load way faster in the first game than the second one. we need someone with a x-25m and the game to post up some load times. it is really going to depend upon which map i load the majority of them in under 20 seconds in game mode. none of them take over 23.
but also, you have to look at its impact of the games performance itself, that page you linked also shows the effect that the faster storage solutions have on FPS and the like. the entire purpose of fast storage does not begin and end with load times. it is a small easily measurable aspect, of course. but what about when you are actually playing the game?? your lower frame rates are where you spend alot of your gaming sessions at. that is where the performance really comes into play. that is where you are really seeing the performance truly.
load times can be too much of a focus of people, much like 4k. people just seem to focus in on it too much. most people dont drop the serious cash just for a few seconds in load times![]()
Last edited by Computurd; 01-02-2010 at 09:30 AM.
"Lurking" Since 1977
![]()
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
people say that they cannot see the performance of their gaming scale as well with the benchmarks, in regards to raiding storage solutions etc. but can they really? or are they just measuring one thing? game loads?
what about scaling of the performance IN the game? you may spend a few seconds loading maps for a game. but how much time do you spend in that map?
Last edited by Computurd; 01-02-2010 at 09:36 AM.
"Lurking" Since 1977
![]()
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
I'm puzzled by that as well because once the game has loaded the disk activity drops to next to nothing and is mostly just writes. (Under 1mb/s!)
It's easy to see the disk activity in the Performance Monitor. Just change the view to "report" and set the "report and histogram" to Maximum.
I've got Crysis somewhere so I will install it to see how quickly it loads.
that is why they are focusing on the minimum frame rates. they usually occur when you are going into a new area of a map or something of that nature, that few seconds where your disk is loading textures. it is only a blip on the radar when it comes to ssd, however for hdd they drop frame rates considerably. these are the hangs and lags that you get in your game. non-existent with extremely fast systems.
again we are looking at an area mos people do not focus on. people always go on and on about maximum frame rates. however what percentage of your gaming sessions are at max frame rate??? barely any. the true performance indicator is the lower frame rates. they show the true weakness of any system.
"Lurking" Since 1977
![]()
Jesus Saves, God Backs-Up *I come to the news section to ban people, not read complaints.*-[XC]GomelerDon't believe Squish, his hardware does control him!
Computurd,
The only application I can think of that lets you set QD is iometer
User selectable QD in normal applications is not possible, like One_Hertz says, it requires different "coding" and many applications would not benefit from this.
(due to the nature of how applications process the data)
On top of this, SSD's have changed the picture a bit, using HDD's the "CPU" was more likely to wait for the HDD than the other way around.
Today it is more likely that the "SSD" is waiting for the "CPU" to process the data.
note:
For an application to send "disk" commands in parallel it needs to be coded differently (multi threaded).
The only other possibility is that multiple applications run i parallel.
(of course it can be a combination of the two)
-
Hardware:
Bookmarks