AS SSD Benchmark benchmarks!



I have not had my undies blown off by my new Vertex, but it's not slow in Windows by any stretch. We shall see how a few weeks/months of being the C:\ drive affects it.



My ol' Intel still has it pegged though... A pretty useful game drive. :D
 
It is starting to look like Vertex 2/le drives don't keep their speed that well and the one area they still can't beat the intels in is small writes(as long term performance seems to degrade in real life), the Crucial however is fantastic, when properly aligned at 4kb threaded reads, and 4k random non threaded also.

I'm wondering if a Vertex 2 with great writes, or the better reads throughout on the Crucial is more important long term, after installing windows and most of your apps, theres less writing to be done, game loading is decently sequential but almost all writes, pagefile is mostly reads, real world tests the Crucial does come out above and it seems to keep its performance better.... when they aren't dying.

I won't post my AS scores as right this second I'm using 2x dm9+ 120gb's in raid 0 :p , my Crucial M225's both died. Though I'll say this, the 4kb random writes are surprisingly high on the maxtors, 3.5MB/s and ok reads aswell, 100mb sequential read/writes between them, only tested in Crystalmark.

I can't decide between a 128GB Crucial, which has noticably slower writes than a 256gb model, but great reads, or a 50GB vertex 2/80gb Intel at half the cost. Problem being, will a great trim/performance returning firmware, or a firmware updating it to a 60gb drive happen, who knows, pain in the ass not knowing.
 
AMDPower said:
why are seq read and write not as advertised or near to?

there seem to be large discrepancy, any reason for this?
Click to expand...

cliffy said:
I have no idea at all at the moment, i'm hoping someone on here who knows more about these things might be able to answer this question????
Click to expand...

Its due to the benchmark and how drives/windows treats data.

Atto uses the best type of data for compression, other benchmarks do not, at all. Intel and Vertex's at least and I assume the Crucial C300 all compress data so the better it compresses, the quicker it can write data as it has to write less to the actual memory.

Atto gives you best case scenario numbers, I believe it uses all 0's as data, so its just stupidly obviously compressible. Some benchmarks, Crystal Disk Mark for certain and maybe AS have the option to use either random data, or 0 fill data, check the differences in performance on a drive that uses compression, theoretically the same benchmark should be markedly faster using 0's or 1's only, and slower with random data. Random data though, is basically how real life usage is, certain situations change but you never ever get all 0's or all 1's in real life usage.

Atto also obviously shows massively lower performance the smaller the files are, because every file writen has overhead, so 100k 512b files have 100k chunks of overhead data to work with, and a single 5mb file has only one overhead with it, thats why iops sucks on 4kb random, but 4kb threaded basically lets it write a whole bunch together in sequence and so it removes all the talking between controller and drive till the file is writen rather than stopping between each file.

I haven't got an SSD in at the moment so can't check, if someone, preferably with A vertex 2, Crucial C300 and intel could run Crystal Disk Mark with the 0 fill option and the random fill option and show any difference that might show a large difference.

Thats the reason that OCZ give on their forum and several reviewers have mentioned for Atto results being no where near real life performance or that of other benchmarks.
 
drunkenmaster said:
Its due to the benchmark and how drives/windows treats data.
Thats the reason that OCZ give on their forum and several reviewers have mentioned for Atto results being no where near real life performance or that of other benchmarks.
Click to expand...

Yet OCZ use Atto for their marketing numbers of drive performance.
 
Broken Hope said:
Yet OCZ use Atto for their marketing numbers of drive performance.
Click to expand...

Of course, firstly it gives the best numbers, the numbers aren't inaccurate, just not really reflective of what you'd get, in any situation except filling your drive with 1's or 0's and not a mix.

Also, everyone uses Atto basically to "rate" their drives as it gives the best possible numbers, so if OCZ use random 4kb writes, and Intel use sequential reads, no one would ever buy an OCZ drive ever.

Its a standard, unfortunately its not hugely accurate and anyone that doesn't know how drive performance drops off and how random data is FAR slower to be writen/read, will almost certainly end up with a "wtf" expression on their face.

I'm about to order a 128GB C300, I don't really want to, I'd probably get a 64gb version if there was one but its being said that the Vertex 2 is 2 months at least away from a end user upgrade to a 60gb capacity, and it might be that long till a decent garbage collection or trim firmware is out aswell.

In the meantime, the Crucial seems to outperform it when the Vertex 2 is at its best, but it would seem the Vertex 2 is rarely at its best.

Again benchmarks in reviews are heavily misleading, often run as the non OS drive and empty which will also drastically alter the numbers, and in a used state certain drives seem to get no where near fresh performance. Intel does, the indilinx does a very good job these days of keeping performance levels close to new, the Sandforce seems to be heavily lacking.

EDIT:- £900 later, a 128gb ssd, another 5850, a monitor and a 890fx later, and I'm already having buyers remorse :p

I'll whack up some 890fx + C300 scores tomorrow, and will put up Crystal Mark with random and 0fill if there is a noticeable difference. Judging by a post on OCZ's forum, there most certainly will be(someone with a few drives in raid got 380mb/s writes in random and 680mb/s writes in 0 fill, with 600+ reads in both tests).
 
Last edited:
My new INTEL X25-M 80GB which actually got faster after i'd finished installing all my apps and programs! ..... but only by 6 points :)

ASSSDX25-M80GBAHCIMBsfullinstall.png
ASSSDX25-M80GBAHCIiopsfullinstall.png


Not sure why its not showing the full FW version although it is def on latest...

Can someone tell me if it's normal for my other platter drives to be showing as BAD :confused: HD Tune reports it as all OK...

ASSSDBAD.png
 
Last edited:
The 31k is the offset, which isn't perfect, but its really for ssd's as they write in 4kb clusters and theres overhead to be calculated when instead of trying to write to the first 4kbs in one sector, your OS is trying to write 4kb data to two separate 4kb blocks on the SSD, which is daft and means either the hdd compensating which leaves a lot of overhead, or a specific part of the controller to handle the offset.

Check Anandtech, actually http://www.anandtech.com/show/2944/10

Check the random write on an unaligned(first graph) and aligned drive. Sandforce controller still does quite well unaligned as the controller is designed to cope with it, but it still gets over 3 times faster when aligned. The crucial C300 controller clearly can't deal with it and gets about a 8 times bump in performance when aligned(in this specific situation).

So alignment is INCREDIBLY important for ssd's, but afaik it makes no difference to mechanical hdd's.

The reason being ssd's are set up in 4kb blocks, and can ONLY write a FULL 4kb every time, and if theres data in that block it needs to delete 4kb, and then rewrite that 4kb, which means worst case scenario you are writing only 2kb of data, but your ssd need to erase two full blocks, and then write one 1kb into each block(because the alignment has caused the 2kb to overlap two blocks. Aligned it should only write 2kb to a single block, so you eliminate 50% of the reads and writes as you only delete and rewrite a single block.

Mechanical hard drives, don't require you to delete a whole block before writing so none of this matters, on a mechanical drive its individual bits of data and you can change any single one independantly of the other.

Thats why in windows 7 it does that extra 100mb hidden partition for booting, its largely done to make sure alignment is always correct on ssd's. Which basically means alignment has to be a number divideable by 4kb. 31k no, 1024k, yes.
 
I would think there must be something else wrong too for it to be so low. Is it connected in a Sata 2 port? Also, have u tried running it again in case it was a one-off? Also make sure no other program is running that would be accessing the SSD at the same time.

Probably someone else who has the same drive might be able to offer some more help, make a new thread about it.
 
Is it normal to get quite varied results between different runs of this benchmark? This especially applies to the 4K-64Thrd score, which seems to a lot lower on some occasions.
 
Have you turned write-caching off? Those are the sort of write results I was getting on my Intel SSD when I did.
 
You must log in or register to reply here.
Back
Top Bottom