Printable View
http://hothardware.com/Articles/Fusi...Review/?page=1
EPIC WIN. Almost the performance of the SLC iodrive for just $900. If I didn't just blow all my money on a nice plasma screen then I would be ordering this right now and throwing my second 5870 out the window.
Basically the best way to get away from the performance equivalent of an X25-M is this thing.
If it had more than 3 months of warranty I would purchase one in a heartbeat.
I love the speed, but I'm not xtreme enough for it apparently. Neither is my wallet. :p
Looking good. Glad prices are coming down. That'll probably be my next ssd upgrade after my OCZ agility gets slow/old and the cost of iofusion is affordable. I really need to hide my credit card, I haven't even gotten my first SSD yet (its waiting to get my house as I type this) and I"m looking at this stuff.
Prices in 1 year might actually be affordable for most people!
That is the first time I have seen Intel at the bottom of benchmark scores. It's a shame it can't boot. That right there is a show stopper for me, but other than that it is very appealing. The first properly designed ssd/pcie solution imo.
Get any drive that holds the NTLDR, and have the actual installation on IOXtreme..
I think that's possible - otherwise someone (or all) at FusionIo are reeeeaaaaly stupid. (they made imba controller logic, I doubt they can't make a driver right)
B R U T A L
is the only word that comes to mind...
Looks like the warranty is now listed as 3 years http://smiliesftw.com/x/xmaseek5.gif
Still no place I can purchase in Canada. :mad:
Its fast, but there is faster out there, get a lsi 9260-8i and a few ssds you can blow that thing away. they are comparing it to a z-drive? lol the zdrive sucks. basically a rocketraid 3520 and four ssd's. the reason it blows so bad is the controller they have it hardwired to. (the z-drive that is). IOEXTREME small file performance is nice, but for it to not be bootable and at 11 bucks GB that is unrealistically high priced, at least for me.
Blow it away? Who needs 2GB/s sustained speeds vs 800MB/s when you got only 80GB of space? You buy this thing for random I/O, period.
Now, whether you need that much random I/O or whether a single X25-M or even a VelociRaptor is enough - that's a different question.
Pointless since it doesnt boot. I wonder with w7 boot partition if the OS can still be installed there.
Sad but true! Same here... :rolleyes:Quote:
Originally Posted by [XC] Synthetickiller
Yeah, no breakfasts for a year and I might afford it! :rofl:Quote:
Originally Posted by [XC] Synthetickiller
But damn, this thing is insanely fast. Wonder if you can see the difference in daily use, though.
The other way around. You can stack as many SSDs as you want in RAID and this thing will still blow them all away quite badly. The Zdrive uses some sort of LSI card btw not rocketraid and a single x25-m is faster than that. The 9260-8i seems to have similar performance to the card on the Zdrive.Quote:
Originally Posted by Computurd
Not bootable is a show stopper for me.Quote:
Originally Posted by One_Hertz
Also - while this is less $ - it is still not as fast as an areca 1231ML with a couple acards (4xR0).
Of course a major downer for the 1231ML/acard solution - only 28GB (formatted) if using 2GB sticks.
4gb ddr2 sticks are still just too much.
I have gotten somewhat proficient vliting w7!
The IOMeter results are impressive, wish they had done more access patterns with large variety of reads/writes than just the workstation pattern. Impressive, a damn shame it costs $900.
I actually think the ioxtreme may be faster @ reads than that setup due to limits of the 1231, but you are welcome to do some tests against the ones ran in the review.Quote:
Originally Posted by SteveRo
When I get a chance I'll run iometer with their w/s settings and we can compare.Quote:
Originally Posted by One_Hertz
well i will have to disagree with that quite vehemently, the statement that
1. the lsi is akin to a z-drive...not even close. and 2. that the i/o extreme outclasses the lsi.
first two are the i/o extreme, one is solo, the other in raid:
http://i517.photobucket.com/albums/u...-pro-vista.png
http://i517.photobucket.com/albums/u...ystal-raid.png
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
NOW my setup with LSI 9260-8i:
http://i517.photobucket.com/albums/u...smallfiles.png
http://i517.photobucket.com/albums/u...EVERPERIOD.png
http://i517.photobucket.com/albums/u...llie1/scre.png
and the thing is those are with vertex, there is not a drive out there yet that can push the lsi to its full specification. there is still much upward performance to be gained from this card, given the right drives. remember it tops out at 2.8 gb/s read. I dont think it is possible to say unequivocally that the i/o extreme outclasses it given the fact that 6gb/s ssd's havent even came to fruition yet.
the first revision of the z-card used the highpoint rocketraid 3520 http://www.madshrimps.be/?action=getarticle&articID=911
^^that was the card i was referring to when you mentioned the z-drive.
the newer version of the z-drive uses the SuperMicro 3Gb/s Eight-Port SAS Internal RAID Adapter, which features the LSISAS 1068E 8-port PCI-E SAS controller.
this is a 3gb/s card.
this card does not even begin to match the specifications or speed of the 9260-8i. you are comparing a 3gb/s card to a 6gb/s card. the LSI controller on the newer z-drive is very very similar to the 3520 on the original z-drive, which is to say it is woefully underpowered for the application it is being used for.
^^^
Sequential rates don't mean much, if anything. Everything you have posted is irrelevant. Run iometer @ the setting done on the review and see for yourself how you will get outperformed a few times over by a single ioxtreme. That LSI is complete crap at everything but large files. I think Steve maxed it out and only got like 20k 4kb random reads? That is less than a single x25 drive. The real world performance will mirror that.
far from being irrelevant my friend:p:
in the course of that review they talk of the fact that the i/o extreme was tested in formatted and non formatted setups, because the difference of a formatted drive in speeds of course is slower, so we shall use their own IOP/s info on the formatted drives.. the i/o "extreme" 80 gb results, as that is the single pcie slot solution, ie single device.
http://i517.photobucket.com/albums/u...orkstation.png
now look at the exact same setup i just ran with win7 installed, firefox browser going at the same time, normal 24/7 setup on my computer. i ran this moments ago. My iops beat the single i/o extreme solution, 80 percent read 80 percent random 8k file transfer, the exact same settings.
http://i517.photobucket.com/albums/u...dom8kfilet.png
another test, same settings with the i/o v lsi 8k 64 read 100 random. lets see a single x-25m do that.
http://i517.photobucket.com/albums/u...eter-dbase.png
http://i517.photobucket.com/albums/u...d100random.png
i believe in this test i even beat the raided i/o extremes specs...
http://i517.photobucket.com/albums/u...orkstation.png
http://i517.photobucket.com/albums/u...orkstation.png
Who knows about block-fragmentation with ioDrive and ioXtreme? http://forum.ssdworld.ch/viewtopic.p...eb591bf4bcc5cb
Are these drives now bootable?
btw:
A old benchmark (IOmeter 2006) with 2x Acard 9010 and Workstation 8k
I must run a repetition with IOmeter 2008 soon.
http://www.abload.de/img/hc_807vistax64fw1.4712q0p1.jpg
If this thing didn't cost as much as buying a good hooker for the night it would be a hell of a deal. Plus whats with the warranty???? That is a company that believes in its products for sure.
Congratulations, you just benched your raid controller's cache :rolleyes:Quote:
Originally Posted by Computurd
I can get similar numbers for my HDD Raid 5 array on an Adaptec 5805, still doesn't mean anything.
OT: if it can't boot, it fails. At least in my book.
But why? CN is a big state :confused:Quote:
Originally Posted by lowfat
Ok, in germany the ioXtreme isnīt available too, but the ioDrive for ~ 18 month.
My humble, horribly degraded, old firmware 2x x25-es on onboard raid do 29k vs your 20k on that test just FYI. The ioxtreme bogs down very visibly when written to so it doesn't seem so quick here. Stuff like game loading is 99.9% reads.Quote:
Originally Posted by Computurd
2 Es do 28k here...Quote:
Originally Posted by Computurd
Because 64k is so large it takes away the random aspect of the equation.Quote:
Originally Posted by Computurd
They did not use very good iometer measures; I should have thought about that before posting my last post. The card is all about reads and only reads. It should do south of 100k 4kb random reads. Don't forget that its access time is also 0.08ms. Yours is what, 3 times longer? These things is why it dominates so much in game loading. A card like your LSI will never be able to beat those specs because of the SAS overhead and will never be able to load games faster. The 1231 might be up to the job with acards or x25s in raid.
@jcool the card only has 512mb of cache so its not much to bench, if in fact that is what we are benching LOL. no matter what anyone posts here there will always be the "what if..." or the "but still..." so what is the point i wonder sometimes?:shrug:
but its fun!!
test is iops, not cache, its ssd's man, they do operations in the ms range bro. also couldnt you then say the same thing about the i/o extreme? that it is just benching ITS cache? so then does that make the fact that either of them are fast totally worthless and pointless cause it is only CACHE!!!!!!!!
dude gimme a break.
@ one hertz....dude its cool, but you are the one who stated that the lsi is crap, etc. i proved you wrong. get over it. we could argue for years about this.
I was only talking about that Crystal Disk mark being useless, the IOmeter results are fine, and of course unattainable with conventional hard drives. I use SSDs too, ya know. ;)
If you want to bench Crystal Disk mark you gotta chose a test size greater than your raidcon's cache.. or it will perform the same regardless of the drives connected (since 100MB test size will ALWAYS be ran purely in the controller's Dram cache). Even with my 4 drive Raid 5 (!) I get 90MB/s 4k :ROTF:
How much IOps (4k - 100% read/write - 100% random - 32 outstanding IOs) i can get with ioXtreme?
I want to see some proof that it's all about random 4k read and linear reads are totally useless. Until that time I think that people should tone down the insistence that small random read is king and all else is just for big benchmark numbers.
Just look at the review. A single x25-m (250Mb/s seq read) is faster at loading l4d than the zdrive (~700mb/s seq read). How could that possibly happen? An even better comparison is the new 40gb cheapo Kingston (Intel) drive. It loads games exactly as fast as the x25-m while having 30% less seq read speed and the same small file speed. What does that tell you? The statement that sequential numbers don't matter is fact and has been for a long time. This is also why you can stack as many drives in R0 as you want, but your games will generally load just as fast as a single drive (many members here including myself have confirmed this).Quote:
Originally Posted by saint-francis
It isn't exactly all about random 4kb reads. It is a mix of random/sequential 4kb-64kb reads. The mix really depends on the games you play.
Anand suggested this:
Kind of makes sense. There will be a few writes when saving the game, but it doesn't write at all during loading.Quote:
94% reads
20% 4KB
20% 32KB
40% 64KB.
69% sequential
queue depth 7.76 IOs.
@Computurd - there is no argument. The LSI is indeed crap just like adaptec. Your whole array (6 or 8 vertexes?) is a lot slower than 2x x25-es. It is not that the vertexes are that slow, it is the LSI. I know for a fact that my x25-es are exactly the same as a single x25-m at loading games. A single x25-m will load games faster than your whole array and the ioxtreme is leaps ahead of even that.
Yepp - the LSI ist a sequential-monster, but a 4k-baby
Arecaīs with IOP341/348 does a better 4k-job
How about for anything aside from games? Like boot times and other app load times? Personally I don't play video games and I think that the majority of PC users don't.
And maybe these SCSI controllers will shine when there are SCSI SSD's. The jury is out on that one.
I think thereīs no big difference with small apps.
We need more detailed reviews with ioXtreme
Sorry I know this is a bit off topic but is the Non-Volatile Memory Host Controller Interface (NVMHCI) the way things will go for ssd? I vaguely recall that it has less raw throughput than sata 2.0 but in practice was faster as it preserves parallelism and improves throughput efficiency:shrug:Quote:
Originally Posted by saint-francis
Depends on the app. Small apps will be like games. Apps like photoshop are completely different and are more into large sequential reads/writes. The biggest gains in boot times are done through software changes (nlite/vlite); other than that, it is also 4kb-64kb, but with a bit more writes. I personally don't really care about anything but game loading times... It is absolutely hilarious in L4D versus to load into the game a lot faster than everyone else and rush through half the map as survivors before the other team even loads.Quote:
Originally Posted by saint-francis
@ FEAR - even better we need someone here to get one. Reviews will always be limited. I'll take a look at my paycheck at the end of the month and I'll see if I can sell a 5870. There is a good chance I'll grab one in 2 weeks or so (I don't have a problem getting one in Canada).
Ok, grab one :)
I look forward to a review here in XS
thats exactly what i did on my 1231/512mb cache and 4x slc jmicron ssd (c:\)Quote:
Here we simply installed Left 4 Dead on each of the drives we've tested and then proceeded to launch single player games on either the No Mercy or Dead Air levels.
no mercy initial loadup: 5 seconds
dead air initial loadup: 4.5 seconds
post initial loadup: 2-3 seconds
freakin slow! time to upgrade to faster ssd! :D
Time to dump my 4 Super Talents to get one of these...
FEAR you have no idea what you are talking about apparently. i have a areca 1680-ix setting right here, and i can prove that it has #1 a iop 348 processor and #2 that it is nowhere NEAR as fast as lsi in 4k reads. i can begin to post screenies if you desire. that is a ludicrous statement.Quote:
Yepp - the LSI ist a sequential-monster, but a 4k-baby
Arecaīs with IOP341/348 does a better 4k-job
@one hertz....it is amazing to me how much of an expert you are at both of these controllers, considering you don't own or have in your possession or have even tested EITHER. just because you have FEAR coming behind you repeating the same thing you say does not make it a fact. do either of you have the aforementioned controllers? have you even used them? what makes you the experts then?
yoiu arrogance and insistence that your single x-25 is faster than my array, or any array, is ridiculous. are you now going to tell us how your single ssd is better than napalms setup as well???
it only shows how your apparently trying to bend reality into supporting the fact that you personally have the end all storage solution. dude get over it.
the fact that you keep comparing the z-drive to the lsi only shows continued ignorance, can you not read???/ that drive has a 3gb/s controller, with an iop that is massively underpowered. do you understand the difference in speed between these two iopsQuote:
single x25-m (250Mb/s seq read) is faster at loading l4d than the zdrive (~700mb/s seq read)
?
if you guys would stop and look over that review a little bit and actually read what it says you would realize that this is an overpriced solution. it is fast but not THAT fast. you say that they did not run the right tests? is it that, or is it just that it is not that good? you and the little cheering section can go buy one, then we can run these same tests over again,. and argue over it all for the foreseeable future it will be great fun.
@Computurd
Your wasting your breath, some folks can never see past the wood to see the tree's :)
All this BS about 4K is all that matters. :D It fits if you wanna be convinced that one drive is king. ;)
4k is for sure a very important factor, but it's only "the" most important factor if you can use all that 4k bandwidth, and most home users will never even come close to using that bandwidth.
I have proved this, and so has Anand in latest article with real world tests.
NapalmV5 already proved it on this forum too.
ioXtreme is for sure fast, but without being able to boot from it, that's a major drawback, IMO.
Irony :confused:Quote:
Originally Posted by NapalmV5
Agreed with pretty much everything you said except the not-bootable part. That doesn't bother me too much.Quote:
Originally Posted by Wendy
napalm is not being ironic, he beat the i/o extreme's load times. he is being sarcastic.
whats the problem fear? you got acards/sandforce ssd/1261.. you should be faster than i am :)Quote:
Originally Posted by F.E.A.R.
and there is always napalm to bring people back to earth with his goddamned array LOL!!!!
We are talking about 1231/1261/1281. The 1680 also sucks for the same reason as the 9260 sucks. They are SAS cards. The 1231 outperforms the 9260 from 2.5 to 6 times @ 4k as shown to us by SteveRO here who tested both cards: http://www.xtremesystems.org/forums/...d.php?t=234374Quote:
Originally Posted by Computurd
I fully trust Steve on his testing. There are other sources that I have looked at as well that show how slow the 9260 is. You showed in this very thread how slow it is. FEAR is not coming in behind me. We are both stating facts.Quote:
@one hertz....it is amazing to me how much of an expert you are at both of these controllers, considering you don't own or have in your possession or have even tested EITHER. just because you have FEAR coming behind you repeating the same thing you say does not make it a fact. do either of you have the aforementioned controllers? have you even used them? what makes you the experts then?
I dont have an x25-m right now. An x25-m is not faster than any array. It is faster than any array based on a non-sata raid card.Quote:
yoiu arrogance and insistence that your single x-25 is faster than my array, or any array, is ridiculous. are you now going to tell us how your single ssd is better than napalms setup as well???
Nobody is bending anything but you. A few ACARDs + 1231 is probably the fastest storage solution. That or the SLC iodrive.Quote:
it only shows how your apparently trying to bend reality into supporting the fact that you personally have the end all storage solution. dude get over it.
Oh the irony. If you read carefully (can you not read???) that whole part of the post was not directed at you. I was responding to saint-francis and talking in general terms about the impact of sequential read speed on the loading times of games.Quote:
the fact that you keep comparing the z-drive to the lsi only shows continued ignorance, can you not read???/ that drive has a 3gb/s controller, with an iop that is massively underpowered. do you understand the difference in speed between these two iops?
Nobody is talking about price here. That is always secondary at XS, especially when comparing high end solutions involving raid cards that all cost a lot to begin with.Quote:
if you guys would stop and look over that review a little bit and actually read what it says you would realize that this is an overpriced solution. it is fast but not THAT fast. you say that they did not run the right tests? is it that, or is it just that it is not that good? you and the little cheering section can go buy one, then we can run these same tests over again,. and argue over it all for the foreseeable future it will be great fun.
They ran a single valuable test, which is measuring the loading time of l4d. I have personally never seen the x25-m get outperformed this much in any review I've ever read. Even ACARDs don't outperform it this much in reviews.
@ Napalm - thanks for the benchmark results; I was actually going to ask you to bench your setup a bit later. We'll see what the ioxtreme can do when paired with proper cpu speed :)
Already asked you to LOAN me your array for 1 day so I can beat Q9650 PCMark05 WR :up:Quote:
Originally Posted by NapalmV5
Thanks, One_Hertz ;)Quote:
Originally Posted by One_Hertz
@ Computurd
SAS-Controllers only simulate the SATA-Protocoll by software.
We get best perf. with SAS-Drives. (hardware to hardware)
The best controllers (small files) for SATA-Drives are controllers with IOP341
i think that you guys are totally underestimating, or overestimating, the amount of latency hit for the lsi. upon looking back at the settings that stevero used for his lsi when he tested them and the fact that he was on the older firmware i believe this card is being misrepresented. he had the settings for cached i/o, and that is about the absolute worst setting for the latency on the lsi, it absolutely destroys it. i know several others who run this card and they agree to a man that "direct i/o" is the ONLY way to run it. Also he has no read ahead selected. adaptive read ahead is by far the best setting as well. i understand that stevero was not familiar with this card, and was also not testing with: first off, a fast ssd, and: second off, acards, these types of things require different settings. so he is not testing it with MY setup is what i am stating. his setup is far from the definitive end-all on testing with this card. the latest firmware for this card was 23 of october, after his testing was "complete". dont get me wrong, im not knocking his testing, he did a good job with what he had. but they have updated card firmware since then, and released windows 7 drivers, which they didn't even have at the time of his testing. this latest firmware was what really took this card over the top tbh. i am not saying that it is the fastest card on the planet, etc. but i AM saying that it is faster, or damn close to, that i/o extreme. napalm has already shown he can beat it, too. steve ro testing is without an OS on the drives, and also was with a 64k stripe on the lsi, which is far from ideal. ideal is 1mb, again that is agreed upon by many others who use this card.
Have you not thought to ponder for a moment that areca is working on the 18xx series?? which is 6GB/s SAS!!! they are not working on a next gen sata3 card that i am aware of. as a matter of fact i have heard that mentioned several times, in several places. that should give you pause. i think that the 6gb/s sas is being underestimated.
now conjecture aside, here are some results, i post the 8k i/o meters i just ran and compare them to the top pic here, which is the listing fo rthe i/o extreme. note IOPS and bandwidth vs mine beneath it.
http://i517.photobucket.com/albums/u...full-blank.png
http://i517.photobucket.com/albums/u.../8kiopsetc.png
also here are two screenies to prove im not just pulling this outta my ass.
http://i517.photobucket.com/albums/u...1/8ksecond.png
http://i517.photobucket.com/albums/u...8kfirstrun.png
I really really wish there were more stats and tests from this i/o extreme to compare to, and there will be soon, i am going to let my array run some GC for a while, then i will run some new tests with the lsi. this is hardly the results of a peice of crap raid card, i guarantee you that. and also i would challenge you to show me where else, other than this forum, that you have seen any testing "proving" how sorry the lsi 9260-8i is. i would love to see some links, because i am failing to find it with the aid of google.
P.s. i have managed to best stevero's results with the the lsi-9260-8i and acards with my array as it is, no gc yet or any tweaking at all this is a fully functional os.
Top result is from stevero's testing with acards on a lsi 9260-8i bottom result is mine. on the top result you are looking at the 4k random reads, i ran the same file he posted in his review, same blend.
http://i517.photobucket.com/albums/u...d4kresults.png
http://i517.photobucket.com/albums/u...tsbeatat4k.png
Ok so you are at 35k now. Not bad. That is a single X25 ssd. Your biggest problem is latency, which is 0.2ish? The 1231 does 5Xk with a very small latency hit. What makes your card a crappy card is that it is only $200 less than the 1231, while being a lot slower. Do you have l4d? Can't really compare loading times directly because they are using a slow cpu (x25-M should be around 8 seconds for those maps with 4.5ghz+ C2Q; ioxtreme around 5 seconds I guess), but can definitely make some inferences based on that.
In regards to your 'other reviews' question. I think toms did a test with two of those cards raided together with 16x x25-es plugged in. From my memory the results were pretty pathetic considering the setup, except sequential numbers. There were other reviews but I would have to search for them. Maybe later tonight.
Edit: I really want one, but I am a bit weary of basing my whole purchase off a single l4d loading benchmark from a site that may or may not know what they are doing. If I knew for sure that it would drop loading times by 30%+ from a single x25-m then I would get it in a heartbeat. Sigh, I will probably get it anyway.
We need pioneers like you One Hertz. Don't fail us! :D:D
In want to be a pioneer too, but no shipping to GER :(
fixed :p:Quote:
Originally Posted by F.E.A.R.
Is ioXtreme now bootable?
There is a shop in 10km distance :p:
But only ioDrive - i think
http://www.youtube.com/watch?v=D03av...layer_embedded
They shows only 100.000 iops by seq. - not random!!! :confused:
no.Quote:
Originally Posted by F.E.A.R.
Windows has to be installed on a separate drive.
Something is def. wrong there, very low IOPS.
I'm sure it can do waay more. At 4K random IIRC a single IoDrive should do 100,000 IOPS. )around 400MB/s)
BTW, recently released IoDrive Octal :D 6GB/s speeds, 800,000 IOPS :D :D :D Soon to replace RAM? :D
Not for me :( :down:Quote:
Originally Posted by lowfat
I better wait for a driver to boot from this thing
@ alfaunits
Nice comparison - data center footprint before and after :)
But what to do with the free-up place??? Yes, more Octals! :D
http://www.fusionio.com/load/media-d...ctal-Study.pdf
Do you not have any friends in the US you trust enough to just have them forward you stuff?Quote:
Originally Posted by lowfat
Currently trying to find one :shrug:Quote:
Originally Posted by One_Hertz
If all goes well I'll order it tonight.
Awesome, so you will be the ginny pig. :up:Quote:
Originally Posted by lowfat
2x Intel 25-m G2 160GB + Areca 1231ML
60 sek for 4KB 100% random 100% read, queue set to 64
http://img94.imageshack.us/img94/443...mread60sek.jpg
Yea this is OS drive
But give me bootable drive and buy it :P
I would love to try out the ARC-1231ML, but in Canada the card itself is more than the ioXtreme. :down:Quote:
Originally Posted by Agrophel
What size was the file you made with iometer?Quote:
Originally Posted by Agrophel
8k is size of file.
But got 2GB cache
Do you seriously expect us to believe that that config can do 100.000 random IOps?:) Give me a break...
Only in cache - skiiiip.
Quote:
Originally Posted by Agrophel
Good evening all, I really want the ioxtreme formfactor (single pcie card as opposed to a raid card with all the data and power cables and drives all separate) to be the better solution. For me, Ioxtreme needs to be bootable and needs to come down in price a bit. I would think the sweet spot (if the hot hardware performance assessment is real) would be something like $499.
So anyway - about performance – again if hot hardware has an accurate report - ioxtreme should stack up pretty well against many high end solutions out there right now. See the data below, the areca 1231ML/acard and the ich10r/acard combo’s are faster in some areas and slower in others (caveat: @ 8KB, 80% read, 80% random). I clocked down my i7 to 20x133 to match the stock 920 clocks used by Hothardware – to make it a more apples to apples comparison.
edit - Also note - I can't tell for sure if Hothardware used one or eight workers for this portion of the bench. My acard testing was with iometer set to one worker (not eight). I will rerun some of the acard tests (below) with eight workers and post the results. When lowfat gets his unit we can construct a more accurate assessment.
edit - looks like moving from one worker to eight decreases the throughput 1-2% based on limited testing of ich01r/6xR0 64, 256 and 1025 outstanding I/Os.
So Ioxtreme looks very promising, I would like to see some 4KB 100% random numbers before I’m totally convinced.:D
http://img682.imageshack.us/img682/8...xtremerevi.jpg
http://img233.imageshack.us/img233/5...card2xr04x.jpg
Ordered. Will be a few weeks before I receive it likely though.
http://smiliesftw.com/x/BermudaBowned.gif
post your link you bot from!Quote:
Originally Posted by lowfat
http://www.amazon.com/Fusion-io-ioXt...8678675&sr=8-1Quote:
Originally Posted by SteveRo
AFAIK this is the only place that they are being sold currently.
@one hertz ..so then you are saying that because of both the latency and the IOPS that the lsi is crap correct? well then you are saying the exact same thing about the i/o extreme as i am posting the SAME LATENCY and HIGHER IOPS than that card in all of the benchmarks . 55687 is the iops on my card, in 8k, which is higher than the i/o extreme, at 41k, and my average response time is around .18, while its is .19...so then the x-25 is faster than the io extreme is exactly what you are saying. please inform me as to how you are reasoning this, i would like to see. also please include the links promised, as i am eager to learn. ia m not understanding the hype on this thing, even what the linked video shows me is not impressive AT ALL.
A SINGLE X-25 gives you 135 mbps in 4k 100 percent random??? i would like to see that screenie.
http://i517.photobucket.com/albums/u...ingstevero.png
just ran that at 8k 80percent read 80 percent random, besting both the i/oextreme and the results stevero posted with the "1231ML acard 6xRO" with the same queue depth of 256.
http://i517.photobucket.com/albums/u.../lookythar.png
This might be vaporware (?) but take a look at the fusionio Octal -
http://www.fusionio.com/load/media-d...ctal-Study.pdf
I already mentioned this before, but the ioxtreme (like the x25-m) is not good at writing. Those iometer setups combine reading with writing. I know the M becomes very very slow when those are done together and I am assuming the ioxtreme is similar. If it is just reading, the numbers will be very different, as we will probably (hopefully) see when lowfat and I get one.Quote:
Originally Posted by Computurd
I should have said access time not latency. Access time of ioxtreme is 0.08. Acards + 1231 are 0.05 or so. Yours is 0.22?
Yes, a single x25-m gives 135mbps at 4kb 100% reads que @ 32. This is the official spec for the drive as well, as you can see on the Intel site. Realistically on a used drive I believe it is a bit less though. Like 125. I dont have an x25-m to give you a screenie, but this is a pretty basic and well known spec of the drive...
You know if you just ran l4d and measured your loading time you would see things for yourself...
Look very carefully at the average response time on the graph here. .19
http://i517.photobucket.com/albums/u...full-blank.png
Now look at this run, case in point..
http://i517.photobucket.com/albums/u...8kfirstrun.png
also on the video it shows access time over .1 and hdtach rounds on access time measurement, hdtach is very inaccurate with access times so we should probably disregard that, but i believe in the video they are testing pure read performance. I understand it is fast in certain scenarios, i just personally do not see the need for these restrictions, it can be fast, and write. Mine does. maybe that is why in why the anandtech article speaking of real world usage patterns, where they actually used the drives in input/output loads of an os, the vertex beats the x-25 series under heavy load usage. The x-25 does best it in two other scenarios, granted.http://www.anandtech.com/storage/sho...spx?i=3667&p=8
I am not understanding these limitations with writing however, these are "next-gen" devices. anyone who has not read the above link should take the time to read just that page. it is very eye-opening and has tons of useful info. show me again where it says that a single x-25m gives you 135 MBPS throughput at 4k? i cant seem to find it all i have seen around these parts is crystal disk mark screenies saying 40 MBPS...perhaps i am looking in the wrong place again. like with the tons of reviews saying the lsi is crap, i still cant find any of those either.in the post your harddrive benchmarks thread here on the site, i am seeing tons of these drives with anywhere from 20 to 40!! 4K throughput, but not one with anything EVEN CLOSE to 125. i also checked intel's site, and couldnt find anything saying that. here is one for example http://www.xtremesystems.org/forums/...&postcount=684. those 4k results on there is what i am referring to.
http://www.xtremesystems.org/forums/...9&postcount=33
150mbps for a single x25m. The benchies you've seen are at lower ques.
Intel site:
http://download.intel.com/design/fla...eam/322296.pdf
Page 8.
I say for the third time, just run L4D and you will see for yourself.
I do not own l4d and dont intend to. I do not run games as benchmarks, too many variables. then you will complain about my tri-sli and overclock and ram speeds.
that is why i run benchmarks, it does the most that can be done to alleviate the BS. One benchmark i do not use is as ssd as it is unreliable when benching hardware raid solutions. it works fine on normal ssd's but gives borked, random unreproducale results in tandem with hardware raid. that is a known problem with many benchmark programs that is why professionals, intel included, run i/i meter.
on the page 8 of that intel pdf it says nothing of 4k random access MBPS. it says about iops. not what you are saying it says. i looked at that very white paper before i posted above.
on AS SSD i believe the 4k-64 means 64 queue depth, but i am not sure. that is not what i am speaking of. the 4k on my results is 150+ with crystal bench mark. there isnt a x-25 out there that can get close to that 4k throughput. I have posted every test imaginable to prove to you what i am saying.
Now you are down to saying..."run a game you will see?"
surely you are joking.
i believe that no matter what i post here, you will continue to say something else different. i have beat it at iops.; i have matched access/response times. i have matched it , or beat it at all throughputs. i am more than willing to run any i/o meter test that you wish to prove yourself correct. but you aren't. period.
i could post a video of this raid card growing legs, walking on water, curing cancer, then creating world peace. you will come back with...that card is crap. i was thinking of taking my card off of this slot that has the nf200 and pulling my other two video cards, because that adds about .05 latency to my array, as the nf200 is a bastage. but what would be the point? i have beat the thing without even disabling services or turning off my gadgets on my desktop for the love of christ.
by the very logic you proclaim then the i/o extreme is a piece of crap as well. this lsi has beat the 1231 with ACARDS IN RAID 0 at iops. still a piece of crap! LOL
you have to also understand, that 4k is not the end all, or even the begin all, of storage solutions. you have such a narrow view of performance that it is laughable.
here is annandtechs results for different computer usage patterns:
The light trace is composed of 37,501 reads and 20,268 writes. Over 30% of the IOs are 4KB, 11% are 16KB, 22% are 32KB and approximately 13% are 64KB in size. Less than 30% of the operations are absolutely sequential in nature. Average queue depth is 6.09 IOs
heavy trace:The benchmark is 22 minutes long and it consists of 128,895 read operations and 72,411 write operations. Roughly 44% of all IOs were sequential. Approximately 30% of all accesses were 4KB in size, 12% were 16KB in size, 14% were 32KB and 20% were 64KB. Average queue depth was 3.59.
he gaming trace is made up of 75,206 read operations and only 4,592 write operations. 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. The average queue depth is 7.76 IOs.
now these are people who know what they are talking about. professional. note that the gaming trace, which is what you are purportedly using your system for, has the LOWEST MIX of 4k reads. interesting. ALSO the highest percentage of sequential IO's. and what is that queue depth of?? 7.76!! that is extremelly lower than the x-25 64 queue depth results you keep mentioning. so then would you say that this lsi would absolutely KILL that l4d loading time, eh?? 1.75 gb/s seq should just about do it.:rofl:
your basic understanding of storage and usage within an operating system are flawed. somewhere along the lines you began to believe that 4k is the answer to everything. which would be nice seeing as how my card absolutely owns at it:p:
i implore you to read this link http://www.anandtech.com/storage/sho...spx?i=3667&p=8
There has been much much discussion about what the computer really uses as it loads and uses an OS., what is the mix of reads/writes, inputs/outputs, randoms/sequentials? Over at OCZ they took a program, much like anandtech did, and monitored the usage of the operating system throughout all phases of use. from this they derived a I/O meter profile, it mirrors real usage, that you can download, that has a realistic mix of computer inputs/outputs. I would love for more people to run it, as it is real world test. the profile is called BOOTUP and they need to include it with downloads of I/O meter!!
you can get it here: http://www.ocztechnologyforum.com/fo...05&postcount=2
i really wish more people would get their heads outta the sand on these things, as you can see from the articles at anandtech, and other very highly respected sites, real world usage actually entails a very small mix of random 4k file usage.
Not necessarily. Iometer isn't a good representation of real world tasks. yHell I don't think any synthethic benchmark is. That is what all these 'respected sites' problems are. It takes much less time and effort for them to do a synthetic bench than running applications/games that people actually use.Quote:
Originally Posted by Computurd
The 'bootup' Iometer config scaled almost perfectly for me when adding another drive to the array. Yet photoshop and game load times didn't differ at all.
http://www.xtremesystems.org/forums/...d.php?t=232796
the fact that your photoshop and game load times didnt differ is probably because of your cpu and GPU . your input/output is far over what they can handle already, load wise. they are bottlenecks now!!!! you have turned the bottleneck from your storage system, into your cpu. run the programs at 2.6 on yer cpu, then run them at 4.6 . they will load faster of course, but note that you did not have to change your storage system to do it. that is the purpose of these benchmarks. to eliminate other bottlenecks as areas of contention.
you can feed them information faster at that point, but you cant make them process it faster.
Computurd- you seem stuck in the world of benchmark numbers. Benchmarking only works when you know how to interpret the results well, and know what exactly it is you need to benchmark. Why don't you list some applications or games that you actually use and you can do a comparison with the guys here. What's the point of having awesome benchmark numbers, but yet the applications and games you use don't run any better?
Looks like a billion dollars!Quote:
Originally Posted by SteveRo
and finally some room in the data center for that pool table:D
well griff805- i would be happy to oblige LOL. there is nothing that i run on my system that does not run, immediately. period. i post to desktop about 10 seconds after raid cad initialization, and everything runs on demand. i usually run at 4.4 on a i7 920, 12gb of ram blasting away at 2045 7-8-7-24. I also have tri-sli-260's on a Evga classified motherboard, audio by creative as well. umm thats about it for rig specs. watercooled with a honda radiator.
I play RTS gaming quite a bit, with multimonitor setup. supcom and fa and universe at war. the obligitory crysis as well. again, no wait for anything. I DO bench quite a bit for fun or boredom. really stopped playing with the array for a few months until here recently...heh. i have never set there with a stopwatch and timed anything, because there is nothing to time. instantaneous.
i do manipulate video quite a bit from home recordings, tons and tons of hd pictures as well. my games and manipulating of files, etc DO run faster LOL.
#13 on futuremark hall of fame for pcmark vantage, high ranking on the hwbots for pcmark05 as well...but i dont use dice or ln2 or anything like that, but my storage system allows me to beat computers with MUCH higher overclocks. it suits me. seriously though it is all about performance with me, and benchmarks are the only way to prove that unfortunately IS by benchmarks. i havent benched my computer at all for the last few months until recently.my system is well rounded, and performs well at everything. i fit all three categories in the anandtech profiles of usage. just depends on what im doing that day. its amazing that you get told repeatedly how inferior your equipment is, kinda riled me up. but oh well i feel like i have derailed the thread here,....sorry.
Running games/apps is the only way to really measure performance. You can run from it all you like, but I have a feeling you are already aware of how slow your PC is in comparison to others.Quote:
Originally Posted by Computurd
Yes, AS SSD doesn't work well with cache. Do not tell me about iometer. I have started getting people to use it about a year ago, back when hdtach was king. Then ATTO came around and I was still trying (failing) to get people to use iometer due to people like Tony (ignorant at that time) saying ATTO is a great measure simply because it showed overrated results and was quick and simple. Only very recently did people accept it. So please don't tell me about it.Quote:
that is why i run benchmarks, it does the most that can be done to alleviate the BS. One benchmark i do not use is as ssd as it is unreliable when benching hardware raid solutions. it works fine on normal ssd's but gives borked, random unreproducale results in tandem with hardware raid. that is a known problem with many benchmark programs that is why professionals, intel included, run i/i meter.
basic math fail... 35000 iops @ 4kb = 140,000 kb/s.Quote:
on the page 8 of that intel pdf it says nothing of 4k random access MBPS. it says about iops. not what you are saying it says. i looked at that very white paper before i posted above.
Yes it means 64 queue depth. It does the same speed at 32 (in fact I dont think Intel SSD support/do anything past 32). Are you saying your iometer run showing 35k iops was with less queue? Your CM run is 100% cache. Take your own advice and use iometer.Quote:
on AS SSD i believe the 4k-64 means 64 queue depth, but i am not sure. that is not what i am speaking of. the 4k on my results is 150+ with crystal bench mark. there isnt a x-25 out there that can get close to that 4k throughput. I have posted every test imaginable to prove to you what i am saying.
Not joking at all. Your L4D will take 9++ seconds to load and you will see. It doesn't have to be l4d. You must game a lot if you have tri sli, so you have to have a lot of games.Quote:
Now you are down to saying..."run a game you will see?"
surely you are joking.
You haven't posted anything but useless 8k runs that involve 20% writes. And in those useless 8k runs you get smoked by mere 2x x25-Es. Your access time is 0.23ms (looked at some of your benchmarks from before), which gets beat by 2x by almost any ssd. Your 4kb random iops is equal to a single x25-m (ioxtreme is a few times more than this). Max throughput is irrelevant.Quote:
i believe that no matter what i post here, you will continue to say something else different. i have beat it at iops.; i have matched access/response times. i have matched it , or beat it at all throughputs. i am more than willing to run any i/o meter test that you wish to prove yourself correct. but you aren't. period.
You haven't beat anything. And turning off services and such doesn't change iops by more than 1%.Quote:
i could post a video of this raid card growing legs, walking on water, curing cancer, then creating world peace. you will come back with...that card is crap. i was thinking of taking my card off of this slot that has the nf200 and pulling my other two video cards, because that adds about .05 latency to my array, as the nf200 is a bastage. but what would be the point? i have beat the thing without even disabling services or turning off my gadgets on my desktop for the love of christ.
Perhaps at an irrelevant IOPS setting at an irrelevant queue.Quote:
by the very logic you proclaim then the i/o extreme is a piece of crap as well. this lsi has beat the 1231 with ACARDS IN RAID 0 at iops. still a piece of crap! LOL
I have posted this EXACT ARTICLE BEFORE IN THIS VERY THREAD. Read post #32. I do NOT think 4kb is the only measure; it is one of the relevant ones. Try to read before posting.Quote:
you have to also understand, that 4k is not the end all, or even the begin all, of storage solutions. you have such a narrow view of performance that it is laughable.
here is annandtechs results for different computer usage patterns:
The light trace is composed of 37,501 reads and 20,268 writes. Over 30% of the IOs are 4KB, 11% are 16KB, 22% are 32KB and approximately 13% are 64KB in size. Less than 30% of the operations are absolutely sequential in nature. Average queue depth is 6.09 IOs
heavy trace:The benchmark is 22 minutes long and it consists of 128,895 read operations and 72,411 write operations. Roughly 44% of all IOs were sequential. Approximately 30% of all accesses were 4KB in size, 12% were 16KB in size, 14% were 32KB and 20% were 64KB. Average queue depth was 3.59.
he gaming trace is made up of 75,206 read operations and only 4,592 write operations. 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. The average queue depth is 7.76 IOs.
Funny. The 8k you keep running isn't even a part of the trace. And you are certainly not running queue of 8 with 1 worker like the article suggests. Try that out and see what happens. And that 8 is average. Meaning, it ranges with use from 1 to whatever the max is. Your 4kb bandwidth @ queue of 1 is like 17mbps (comparing to ioxtreme which is 40+).Quote:
now these are people who know what they are talking about. professional. note that the gaming trace, which is what you are purportedly using your system for, has the LOWEST MIX of 4k reads. interesting. ALSO the highest percentage of sequential IO's. and what is that queue depth of?? 7.76!! that is extremelly lower than the x-25 64 queue depth results you keep mentioning. so then would you say that this lsi would absolutely KILL that l4d loading time, eh?? 1.75 gb/s seq should just about do it.:rofl:
You can not mirror real world usage with iometer. It is impossible. The config they have made has absolutely no relation to the real world. It is just a read/write test with 4kb-64kb files @ queue 64. First of all the queue is all wrong as you have already mentioned.Quote:
There has been much much discussion about what the computer really uses as it loads and uses an OS., what is the mix of reads/writes, inputs/outputs, randoms/sequentials? Over at OCZ they took a program, much like anandtech did, and monitored the usage of the operating system throughout all phases of use. from this they derived a I/O meter profile, it mirrors real usage, that you can download, that has a realistic mix of computer inputs/outputs. I would love for more people to run it, as it is real world test. the profile is called BOOTUP and they need to include it with downloads of I/O meter!!
you can get it here: http://www.ocztechnologyforum.com/fo...05&postcount=2
i really wish more people would get their heads outta the sand on these things, as you can see from the articles at anandtech, and other very highly respected sites, real world usage actually entails a very small mix of random 4k file usage.
Second of all (and more importantly) this approach is inherently wrong. Let me dumb it down for you so you can hopefully understand. Say you have a program that does 100% reads for 5 seconds and 100% writes for another 5 seconds. Tracing it and converting it to iometer would run this "real world simulation" as 10 seconds of 50% reads/50% writes, which would produce a result that is COMPLETELY UNRELATED to the performance of the original application. Running a very complicated simulation, like bootup just makes it even more useless if that is even possible.
I love your point by point refutations..perhaps i should do that as well to make my posts look as long and important as yours!!:p:
simply not true. i wait for nothing, i was past that point with a rocket raid 3520 and eight vertexQuote:
Running games/apps is the only way to really measure performance. You can run from it all you like, but I have a feeling you are already aware of how slow your PC is in comparison to others.
your agreeing with me. wow that is nice.Quote:
Yes, AS SSD doesn't work well with cache. Do not tell me about iometer. I have started getting people to use it about a year ago, back when hdtach was king. Then ATTO came around and I was still trying (failing) to get people to use iometer due to people like Tony (ignorant at that time) saying ATTO is a great measure simply because it showed overrated results and was quick and simple. Only very recently did people accept it. So please don't tell me about it.
touche`Quote:
basic math fail... 35000 iops @ 4kb = 140,000 kb/s.
used for comparison only, as it was what i could find with a quick search. also believe me, only things i really trust is everest and i/o meter. agreeing with me..in a way...againQuote:
Yes it means 64 queue depth. It does the same speed at 32 (in fact I dont think Intel SSD support/do anything past 32). Are you saying your iometer run showing 35k iops was with less queue? Your CM run is 100% cache. Take your own advice and use iomete
i do have lots of games, however i have never had to wait for anything long enough to say...GEE this is taking so long i should time it. i dont wait. i can guarantee that it wont take that long. feel free to send me a copy of the game.Quote:
Not joking at all. Your L4D will take 9++ seconds to load and you will see. It doesn't have to be l4d. You must game a lot if you have tri sli, so you have to have a lot of games.
max throughput is relevant entirely. as you agree with the anandtech post, then you see that in a gaming mode it is 60 percent sequential...how in the hell does that not matter? it is the vast majority of the reads. 8k was only as frame of reference, working from the benchmarks provided to me by the review. how can i compare my settings to theirs if they dont have it listed??? when i show you better access times, you dont acknowledge them.Quote:
ou haven't posted anything but useless 8k runs that involve 20% writes. And in those useless 8k runs you get smoked by mere 2x x25-Es. Your access time is 0.23ms (looked at some of your benchmarks from before), which gets beat by 2x by almost any ssd. Your 4kb random iops is equal to a single x25-m (ioxtreme is a few times more than this). Max throughput is irrelevant.
i beg to differ.Quote:
You haven't beat anything. And turning off services and such doesn't change iops by more than 1%.
again anything that i post that is good is irrelevant. it does not support your statement. so therefore it means nothing, you are like the gestapo LOL.Quote:
Perhaps at an irrelevant IOPS setting at an irrelevant queue.
you refuse to acknowledge anything that even remotely differs with your pinhole view of computing.
it was so irrelevant that stevero, who you have said that you trust implicitly, posted them. do you need to revise your stance on his benchmarks??????
i thought after reading your other statements throughout the whole entirety of this thread that maybe you should read it again. you don't get it. you keep repeating the same mantra. 4k 4k 4k 4k 4k 4k 4kQuote:
I have posted this EXACT ARTICLE BEFORE IN THIS VERY THREAD. Read post #32. I do NOT think 4kb is the only measure; it is one of the relevant ones. Try to read before posting.
yes it isnt part of the trace. i have only ran it because that is one of many comparisons i have made to the benchmarks provided in the review. if they had one at some other size as well i would run it. i believe i have ran almost all of them...40+?? i missed that part!i have reached over 90 before with one worker? would you care to see the screenshot? i never tire of posting them.Quote:
Funny. The 8k you keep running isn't even a part of the trace. And you are certainly not running queue of 8 with 1 worker like the article suggests. Try that out and see what happens. And that 8 is average. Meaning, it ranges with use from 1 to whatever the max is. Your 4kb bandwidth @ queue of 1 is like 17mbps (comparing to ioxtreme which is 40+).
partially true, partially not. unfortunately we cannot get a perfect picture of system usage, however we can get close. you assign levels of access to the individual components, giving some more precedence than others, giving you an albeit imperfect, yet blurry view. you dont agree with it because it makes your solutions, such as i/o extreme and x-25, slow. they cannot do read and write at the same time and have decent numbers. how real world is THAT???? and with the way that things are going with you and your acceptance of things, it doesnt matter how close we get anyway tbh. its like the access times. i show you how i actually have faster access than the i/o extreme and you immediately start another thread in the forum about how the computer industry in general must be measuring access times incorrectly.Quote:
You can not mirror real world usage with iometer. It is impossible. The config they have made has absolutely no relation to the real world. It is just a read/write test with 4kb-64kb files @ queue 64. First of all the queue is all wrong as you have already mentioned.
Second of all (and more importantly) this approach is inherently wrong. Let me dumb it down for you so you can hopefully understand. Say you have a program that does 100% reads for 5 seconds and 100% writes for another 5 seconds. Tracing it and converting it to iometer would run this "real world simulation" as 10 seconds of 50% reads/50% writes, which would produce a result that is COMPLETELY UNRELATED to the performance of the original application. Running a very complicated simulation, like bootup just makes it even more useless if that is even possible.
dude that is awesome.
because someone has results you either A: dont agree with or B: dont believe, you begin to question the industry as a whole. you are awesome i love these long talks with you in circles. what is in that guys signature, particles rules of internet or something? i would look it up but god im sick of looking things up, i must save myself for refuting your next post. anyways it goes something like this:
i believe you are banking on that my friend, as your posts are becoming increasingly repetitive. but hey, im here all night!Quote:
Whoever repeats the same thing over and over for the longest period of time will win
has that ever occurred to you that is a limitation of that drive? a liimitation that could be overcome by what?? cmon humor me....thats right folks!!! RAID CARD!!!Quote:
Yes it means 64 queue depth. It does the same speed at 32 (in fact I dont think Intel SSD support/do anything past 32)
69% sequential throughput at block sizes of 4kb-64kb is not max throughput. Max throughput is 100% sequential, block size of like 512kb. Max throughput it irrelevant.Quote:
Originally Posted by Computurd
Because almost everything you posted so far is irrelevant to real world. Iometer is mostly irrelevant to the real world to begin with. Go ahead and post something as per the anand benchmark. i.e. 1 worker, 8 queue, 4kb-64kb size, reads, partially random.Quote:
again anything that i post that is good is irrelevant. it does not support your statement. so therefore it means nothing, you are like the gestapo LOL.
you refuse to acknowledge anything that even remotely differs with your pinhole view of computing.
I trust that his results are true. Different things.Quote:
it was so irrelevant that stevero, who you have said that you trust implicitly, posted them. do you need to revise your stance on his benchmarks??????
You can read 4kb as "small files in general". 4kb is just an example.Quote:
i thought after reading your other statements throughout the whole entirety of this thread that maybe you should read it again. you don't get it. you keep repeating the same mantra. 4k 4k 4k 4k 4k 4k 4k
CM works fine on devices that have no cache (ioxtreme). It is showing 40-odd mb/s. No, you did not reach 90mb/s with 4kb 100% random, 100% reads, 1 worker, 1 queue.Quote:
yes it isnt part of the trace. i have only ran it because that is one of many comparisons i have made to the benchmarks provided in the review. if they had one at some other size as well i would run it. i believe i have ran almost all of them...40+?? i missed that part!i have reached over 90 before with one worker? would you care to see the screenshot? i never tire of posting them.
I tried my best to explain it to you. I really did.Quote:
paritally true, partially not. unfortunately we cannot get a perfect picture of system usage, however we can get close. and with the way that things are going with you and your acceptance of things, it doesnt matter how close we get anyway tbh.
You seem to have read that thread, so go ahead and read what access time is. Hint: queue of anything but 1 is not the measure of access time. Stop being so self centered. That thread has nothing to do with you.Quote:
its like the access times. i show you how i actually have faster access than the i/o extreme and you immediately start another thread in the forum about how the computer industry in general must be measuring access times incorrectly.
dude that is awesome.
Because you do not understand the basics so I have to keep trying to get my point across, which unfortunately leads to me repeating myself.Quote:
i believe you are banking on that my friend, as your posts are becoming increasingly repetitive. but hey, im here all night!
nice of you to explain that for us. however, max throughput as i am referring to it is the reads not being entirely random. the less random the reads, the closer to max thouroughut is the thought i am trying to convey. my reads at 4k random, vs 4k entirely sequential are two different beasts entirely, that i assure you. non-random aka sequential does equal max throughput of that file size. MAX throughput of ANY file size is entirely relevant.Quote:
69% sequential throughput at block sizes of 4kb-64kb is not max throughput. Max throughput is 100% sequential, block size of like 512kb. Max throughput it irrelevant.
here is anandtechs direct A whopping 69% of the IOs are sequential, meaning this is predominantly a sequential read benchmark"
not 69 percent sequential throughput.
69 PERCENT OF THE INPUT/OUTPUTS are sequential. ( so 69percent of the reads are 100 percent sequential) i am trying hard to explain here.
the input/outputs are NOT 69 percent sequential. again failed logic. i am trying hard to explain here
i already have, and it was better. thus invoking cries of "it sucks!!" from you.Quote:
Because almost everything you posted so far is irrelevant to real world. Iometer is mostly irrelevant to the real world to begin with. Go ahead and post something as per the anand benchmark. i.e. 1 worker, 8 queue, 4kb-64kb size, reads, partially random.
explain how it is different when he runs it? are his 8k magically better than others?Quote:
I trust that his results are true. Different things(I.E. stevero results).
like 8k??? i thought that 8k is between 4-64k. but then you state 8k is irrelevant. strange. it is closest to your favorite number., how is it irrelevant?Quote:
You can read 4kb as "small files in general". 4kb is just an example.
i find it amazing that you make that post within five minutes of my screenshot showing better access times than the i/o extreme. but i will take the self centered comment with a grain of salt.Quote:
You seem to have read that thread, so go ahead and read what access time is. Hint: queue of anything but 1 is not the measure of access time. Stop being so self centered. That thread has nothing to do with you.
you are trying to define what the basics are, according to YOU. however you have flawed reasoning.Quote:
Because you do not understand the basics so I have to keep trying to get my point across, which unfortunately leads to me repeating myself.
You haven't posted anything at queue 8, 1 worker. I've attached an iometer test which is a conversion of the anand gaming trace. It doesn't really mean real world performance, but since you believe it does, you can knock yourself out.Quote:
Originally Posted by Computurd
1 worker
94% reads
20% 4KB - made this into 25% (amplified everything to make it total 100% access specification since anand didnt specify 100% of it)
20% 32KB - made into 25%
40% 64KB - made into 50%
69% sequential
queue depth 8 IOs (7.76 was anand's trace, but I am not sure it works in decimals).
3minutes
10,000,000 sectors
Anyhow, anyone that understands about storage has had their laughs by now.
hold on one minute here, i am wanting you to explain yourself here. i made an edit while you werent looking apparently.
nice of you to explain that for us. however, max throughput as i am referring to it is the reads not being entirely random. the less random the reads, the closer to max thouroughut is the thought i am trying to convey. my reads at 4k random, vs 4k entirely sequential are two different beasts entirely, that i assure you. non-random aka sequential does equal max throughput of that file size. MAX throughput of ANY file size is entirely relevant.Quote:
69% sequential throughput at block sizes of 4kb-64kb is not max throughput. Max throughput is 100% sequential, block size of like 512kb. Max throughput it irrelevant.
here is anandtechs direct A whopping 69% of the IOs are sequential, meaning this is predominantly a sequential read benchmark"
not 69 percent sequential throughput.
69 PERCENT OF THE INPUT/OUTPUTS are sequential. ( so 69percent of the reads are 100 percent sequential) i am trying hard to explain here.
the input/outputs are NOT 69 percent sequential. again failed logic. i am trying hard to explain here
who is laughing now? and at whom? your whole basis of your argument that it is for game loading. you dont get it.
The max throughput of the device is the maximum mbps it can spit out at you. This is achieved at 100% seq, 100% read, and large blocks.Quote:
Originally Posted by Computurd
Yes, they are entirely different. Yes if you go 100% sequential, that will be your maximum iops at your selected file size. Edit: not true for raid setups with no read ahead.Quote:
my reads at 4k random, vs 4k entirely sequential are two different beasts entirely, that i assure you. non-random aka sequential does equal max throughput of that file size.
I don't completely understand what you mean there.Quote:
MAX throughput of ANY file size is entirely relevant.
Yes, 69 percent of the accesses are sequential... This can not be interpreted in any other manner. A single IO can not be partially sequential. It is either sequential or it is not. This is how iometer does it too. i.e. if you want to do 100 4kb reads, and set it as 69% sequential, 69 of them will be sequential and 31 wont be. I do not understand what you thought I meant.Quote:
here is anandtechs direct A whopping 69% of the IOs are sequential, meaning this is predominantly a sequential read benchmark"
not 69 percent sequential throughput.
69 PERCENT OF THE INPUT/OUTPUTS are sequential. ( so 69percent of the reads are 100 percent sequential) i am trying hard to explain here.
the input/outputs are NOT 69 percent sequential. again failed logic. i am trying hard to explain here
exactly what you said. you said:
emphasis added of course. but i refuse to debate grammar. but you are revising your statement now i see. from what i understand this is what you are saying, which would mirror what the article says:Quote:
69% sequential throughput at block sizes of 4kb-64kb isnot max throughput. Max throughput is 100% sequential, block size of like 512kb. Max throughput it irrelevant.
A input/ouput is either sequential, or not.
69 percent of input/ouput commands are sequential under normal gaming scenarios.
so sequential throughput is relevant. because it is NOT random in 69 percent of the input/output.
can you agree to that much^^^ because i believe that max throughput of any block size is the maximum sequential access, non random access, of that block. i mean of course it is. that is the very definition of maximum. the most.
so how can you say that maximum sequential speed is irrelevant? that would mean that not only is max throughput of a small file block relevant, it is MORE RELEVANT than random. it is a 2 to 1 ratio. making it absolutely more important. this raid card will blow it away at 4k sequential. would you like to see some? that is where we are getting lines crossed. this thing is WAY faster at sequential than any last-gen card. positively torches them.
this has not been the crux of our debate of course,, but now young padawan...
....you see why i say, i dont wait. ever. for anything. this controller is fast a little bit at sequentials. this is 4k 100percent sequential. so IF SEVENTY PERCENT of I/O commands during gaming are sequential...guess what? that is one worker. that is the smallest block.
http://i517.photobucket.com/albums/u...sequential.png
now i brace myself for the onslaught of x-25 specs that are faster, single drive, for gaming than my array. and also others benchmarks lol :yepp: now i think were getting somewhere. i think the only thing for us to solve this is to wait until you get your drive. the 1231 can barely hit that with its largest block size in maximum sequential, right?
You can't just pick a single access measure you like and run that. Why haven't you ran my iometer test file I provided which combines everything anand has said?
To the other post: your definition of max throughput is different from normal. I stand by my definition of the term. Throughput is the rate at which data can be moved. Your max throughput is roughly 1.75gb/s I believe. That number is irrelevant. That is what I was saying.
well, you were the one who said it best, inaccurate picture:clap:
but i also agree that it is a hazy picture. what were your results, and have you imbedded a virus?:ROTF:
yes you can. if you are comparing two controllers, one that tops out at 2.85 GB/s and one that tops out around 800mb/s, would it not be suffice enough to say that the IOP on the latter is considerably faster? and i mean not by a little. A LOT. i have not even topped out this card. it is impossible at this point, it is designed for 6gb/s devices man. IN PARRALLEL. wait until ocz next round of sandforce ssd with the 6gb/s sas connections, jesus then it will be a real slobberknocker, of course then there wont be ANY question. the top of this card with 6 gb/s devices is 2.85 gbps. let that sink in bro. with a sas connection which allows parallelism. jesus lord the magnitude will be shocking. have you pondered the thought that areca is not making a sata3 controller???that the 18xx series is going to be SAS??? 6 GB/S??? are you seeing a pattern developing here? this is next gen, that is last, get over it.Quote:
To the other post: your definition of max throughput is different from normal. I stand by my definition of the term. Throughput is the rate at which data can be moved. Your max throughput is roughly 1.75gb/s I believe. That number is irrelevant. That is what I was saying.
The cpu on your raid card can be as fast as you want it to be. The problem that it is emulating SATA through SAS.Quote:
Originally Posted by Computurd
I get 7100 iops on the so called anand test.
UMMM.... dude, what were you running that bench on because you are not gonna believe this....this might be a little faster than a x-25...maybe just a tiny little bit....like maybe on the magnitude of...i am not sure if it is running correctly. it cant be.