SSD for X4x sum up.

[bgx]

Hi all,

thanks to all for all the good posts first (i am a lurker for quite some time now).

I post here different choices of SSD for X4x, and their pro and cons.
I need your help as the disk are usually to tested (or tested badly) by websites!

----------
I Kingspec (available from ebay and memoryc.com)
a priori, the one would be the KSD-PA18.1-0xxMJ
M meaning MultiLevelCell and J meaning JMICRON (bad) controler.
PA if for Parallel ATA (no ZIF, standard 44 pin used by X4x).

pro: not so exp, easy to find, no need for modding, should just fit, performance in *reading* very good (13 MB/s at worse - 4KB random).

con: performance in *writing* 4K random abysmal (what could/should provoke stuttering of your computer). It could do only 7 writing operation per seconds (that is writing 28 KB/s - compared with 111 ops/ 430KB/s for my 4200tpm HDD). That's just plain awful. Flashpoint, a cache program should help.
Consumption does not seem to be better than a HDD (at idle at least), reporting 0.6W@idle and 1.25W@writing.
Bips as unrecognized HDD from bios (X41 only- fixable?).

perf/consumption from datasheet
http://www.kingspec.com/downloads/KingS ... t_V1.0.pdf

----------

IBM/Lenovo 60GB SSD.

It seems to be a refurbish Samsung SSD, with SLC (singleLevelCell -faster, more robust and twice as exp) edit: i doubt it is SLC, i guess it is MLC, at least the one i get. It seems to be the same as in the macbook air, the MCCOE64GEMPP-01A. A review shows that read is very good but write is worse than on even a 4200 HDD
http://www.anandtech.com/mac/showdoc.aspx?i=3226&p=14.
I dont have the exact numbers though, if someone can run iometers (crystaldisk and atto does not seem to do a good job benchmarking 4K random write accurately)...

edit: I got this disk, and run test:
crystal mark gives good figures:
sequential R/W: 41/21
512K R/W: 41/12.1
4K R/W: 14/1.5

But running IOmeter reveals it is 3times worst in writing wrt the HDD:
4K 100% random
R/W: 2500IOPS (10MB/s) /32 IOPS (140KB/s)

conclusion: Crystal mark is not a good tool to check performance!

consumption of the WHOLE computer:
idle: 6.4W (was 6.7W with HDD, gain 5% of powerconsumption).
max@write sequential: 8.3W (same as the HDD).

pro: no need for modding, no bips, should just fit, performance in *reading* very good (exact figures?- no samsung controller do less than 16MB/s at worse), seems to consume less than a 4200 TPM HDD (from Anandtech), in particular at idle. works on both X40 and X41/t.

edit: yes, it is checked by my experiments, but the powerconsumption is not much better. The only thing is that it would go to low power immediatly, while a disk would rarely spin down.

neutral: *writing* 4K random should be around 50 IOP/s, that is 200 KB/s, much better than kingspec, but still slightly worse than even a 4200 HDD. Stutterring should be gone though. (SLC robustess ?).

con: quite expansive, though not that bad from ebay, not so easy to find.
write speed worse than HDD, for instance, hybernate is slow.
Also, copying file from my ftp server seems slower (2.4 MB/s vs 3.1 MB/s in wifi G)

perf and consumption taken from PS410 which should be close here:
http://www.hardware.fr/articles/765-3/i ... ville.html

--------

Runcore Pro 4:

MLC, indilinx barefoot controller.

I got this disk but dont have it anymore.
Here are the data i noted:

Running IOmeter reveals it is very fast:
4K 100% random
Write: 1117 IOPS (4.4 MB/s)

pro: no need for modding, works well in X41/t, blasingly fast in read and write, no stuttering. Win 7 works very good on it. Should get a trim support in the to be realeashed firmware.

con: the most expansive of all, customer support from Runcore is [censored] or non existent. power consumption is higher than the HDD @Write. It seems it takes 3W for the SSD itself! It results that IT DOES NOT WORK IN X40(with the current hardware revision, confirmed by many) - bluescreen. but still work in X41.

---------

Wanna help? i need runs of iometer of
-random 4K writing and
-random 4K reading
of your SSD!
Power consumption figures may help too!
I ll make a post to explain how to use, as it is not as trivial as with other benchmark (but is also very accurate).

[bgx]

My post has 2 goals:

first, gather on the first post (i will edit as much as i can) all the SSD available for X40, and their performance (that is, performance which matter, read below).
For that, as the X40 is a niche, we need to do the testing ourselves, hope i can have the help of ppl of the forum for that!

I have been quite interested in SSD for my X40 and desktop, and read a lot about it. As it is a new technology, a lot of people get confused - in particular many websites testing hardware material, making many here confused
I base all my knowledge on 3 different website, anandtech.com, hardware.fr and
pcworld.fr (good comparator at http://www.pcworld.fr/comparatif/materi ... 09/434611/). What you have to look at in SSD is very different to what matters for HDD. Basically, read/write for HDD is similar, while very disimetric for SSD. The weakest point for SSD seems to be random write of 4K blocks. The reason is that to write 4KB on a non empty block (that is if there is a 0 in the zone of 4KB where it wants to write), the SSD needs to erase the whole 512KB block: it thus needs to read the remaining 508KB, erase the block, and write 508KB from memory + 4 new KB. That is, writing 4KB= reading 508KB + writing 508KB + erasing. that just destroy the bandwidth -
Intel and indilinx fight it by gathering several logically far away blocks to be written on the same physical block. Writting 16 times 4KB blocks is thus written in the same free 512KB block. Perf for these 2 controllers is degraded with time (unlike with other) when there is no more free blocks since the controllers cannot gather the writes on the same 512KB block. That is where trim comes into play to tell the SSD that some blocks are actually free (filling them with 1's, as explained by sjthinkpader below)!



To actually test 4KB random write, you need to use iometer (numbers reported using by crystal or ATTO seems way of the reality, probably being accurate for HDD, but they were not made with SSD in mind).

Here is an how to:
1 download from:
http://www.iometer.org/doc/downloads.html

2 unzip, go to src/release/ and lauch iometer.

3 select your computer under topology/all manager and check one harddisk (it just need to have free space).

4 go to access specification, new (twice, once for read, once for write). What to change:
- name (4Kread and then 4Kwrite)
- Transfer request size 4KB (from 2),
-percent of read/write, move to 100%write (when testing write) and 100% read (when testing read).
ok

then add these new access spec. to the assigned access spec.
(do not use the preconfigure 4K as they are not random!)


5 press the green flag, cancel the file to write the result (no need).
you can follow what is happening in the console.
first, it will grab all free space on your hard drive (preparing disk), and then it will start the test. After the test has started and before you are too fed up, you can just stop the test, and go to result display to see the results (the important are average IO/S and average MB/s).

post here with the SSD spec.

6 dont forget to delete the file iometer.tst after use of iometer as it takes up space on your harddisk for nothing!

For powerconsumption, using the battery gauge of ibm may be useful...

[sjthinkpader]

Welcome to the forum and thanks for the evaluation comments.

SLC=single level cell and MLC=multi-level cell. Originally, all flash were SLC. They stored 1 bit per cell. Later 2 different voltages were stored and detected to give two bits per cell.

You may know these comments below already:

Blank flash is a field of "11111....". Write to flash can only change "1" to "0" and never a "0" to "1". Erase is used to change "0" to "1. When free space is "dirty" or field of random old data "01001001..." etc, then erase must be done before write. TRIM in Win7 must be kind of operation done in the background to make dirty free space all "11111..." again. Originally this was called erase/reclaim.

TrueFFS and other file systems made for flash memory technology used read/modify/write instead of just write to free space in order to avoid erase as much as possible, until the whole block is full. Then valid data is collected and written to clean space, FAT updated then dirty block erased. Frequently a spare block is set aside specifically to do reclaim.

I have made a manual erase operation with a freeware called Eraser. Since it was made for secure erase of magnetic data, I made a single pass '1111..." data pattern and use it to clean dirty free space in flash.

[bgx]

Welcome to the forum and thanks for the evaluation comments.

SLC=single level cell and MLC=multi-level cell. Originally, all flash were SLC. They stored 1 bit per cell. Later 2 different voltages were stored and detected to give two bits per cell.

thanks! and thx for the MLC/SLC acronyms! i edited first post. Now if you want to give a little of your time you would be very helpful with all your SSD :p!

Here is the kind of thing i am looking for
(i just perform one test on my HDD):

Hitachi HDD 60GB, 4200 tpm.

4K 100% read random
49 I/O per s
0.19 MB/s

4K 100% write random
99 I/O per s
0.39 MB/s

power consumption:
Idle:
bare minimum (downvolted CPU, lowest luminosity, no pagefile on HDD, no internet, wifi off).
HDD vs disk under test: 6.7W - 6.7W
(of course, it is the same disk- i dont have SSD... yet).

same with IOmeters random write running:
HDD vs disk under test: 8.3W - 8.3W

W@write - W@idle: 1.6W.

Of course, the wattage reading is not precise, so i guess the best is to report the minimum power consumption observed several times during the test. Of course, it should be done unpluged from the AC :p.

---

Here is another one for the kingspec SSD, taken from kingspec itself:

Kingspec SSD 60GB.

4K 100% read random
3334 I/O per s
13 MB/s

4K 100% write random
7 I/O per s
0.028 MB/s

W@write = 1.25W
W@idle = 0.6W

W@write - W@idle: 0.625W.


---

I saw someone using a mobi MOBI3500 in his X4x.
perf and consumption data are taken from
http://www.hardware.fr/articles/765-3/i ... ville.html
http://www.hardware.fr/articles/765-7/i ... ville.html

4K 100% read random
7439 I/O per s
30 MB/s

4K 100% write random
59 I/O per s
0.23 MB/s

W@write = 2.8W
W@idle = 1.2W

W@write - W@idle: 1.6W.
clearly much too high, but actual power consumption may be lower (i dont know how it was mesured).

----

I d be very interested by the IBM/Lenovo 60GB SSD figures, as if it is close to the samsung PS410, it would be amazing (good perf of mobi + perfect power consumption). I am still not sure if the SSD is really SLC based...

perf and consumption data are taken from
http://www.hardware.fr/articles/765-3/i ... ville.html
http://www.hardware.fr/articles/765-7/i ... ville.html

4K 100% read random
5109 I/O per s
21 MB/s

4K 100% write random
114 I/O per s
0.4 MB/s

W@write = 1W
W@idle = 0.3W

W@write - W@idle: 0.7W.

[sjthinkpader]

I will look into the IOMETER but it will take me sometime.

[Frobe70]

Have anyone tried the new RunCore Pro IV?
1.8" size is available both as ZIF and IDE, both should fit X40/41 (ZIF need adapter).
16-128GB. MLC flash, Indilinx controller.
www.mydigitaldiscount.com list most of them as preorder except 32GB ZIF.

[bgx]

Have anyone tried the new RunCore Pro IV?
1.8" size is available both as ZIF and IDE, both should fit X40/41 (ZIF need adapter).
16-128GB. MLC flash, Indilinx controller.
http://www.mydigitaldiscount.com list most of them as preorder except 32GB ZIF.

seems very promising with indilinx, as it would be probably wonderful! great read, write, consumption perf. The only problem would be the eror/bips, but that s not so bad...

i checked the price and the figures, if true, that would be simply amazing!
4K write @ 18MB/s (very credible from indilinx), power consumption of 0.65W-1.2W. That is, the price of the kingspec with the perf of the ibm-lenovo/samsung!

[sjthinkpader]

I have the older SLC Sandisk SDU5B. I have to check its performance with iometer.

[greg.billings]

I used the flashpoint software (before the name change) on a Umid M1 (Koren ultra portable hand top) and what I would get is the system trying to do checkdisk but it would fail. I uninstalled and did the check disk and life was good for a while. Because flashpoint creates a cache out of system memory so that the writes can be made to appear to be quickly, the problem is if you want to hibernate or god forbid a bsodsay hellow to data corruption.

I think the driver support in XP is the problem for flash drives - it just wasn't designed for it. I don't have any background to support it but it seems to me the file system and OS are all designed around the premise of a rotating drive. I would be fine with flashpoint but the risk of data corruption concerns me. I am sure that flashpoint would solve the write issues with the kingspec.

[mgo]

Hi all,

thanks to all for all the good posts first (i am a lurker for quite some time now).

I post here different choices of SSD for X4x, and their pro and cons.
I need your help as the disk are usually to tested (or tested badly) by websites!to do a good job benchmarking 4K random write accurately)...
iometer of
-random 4K writing and
-random 4K reading
of your SSD!
Power consumption figures may help too!
I ll make a post to explain how to use, as it is not as trivial as with other benchmark (but is also very accurate).

I hope you will excuse my off topic question. In your experience, is it good to turn off Prefetch, Superfetch and Defrag when using a solid state drive? Answers on line are often very contradictory! My SSDs are Inel 80 gig.

[sjthinkpader]

On my Sandisk 32GB SSD SDU5B-032G, 4K random write came to between 14-16 IO/Sec.

The WD1600BEVE 160GB, 5400rpm, 8MB cache HDD in my T42p did 320 IO/Sec for the same 4KB random write.

[bgx]

I hope you will excuse my off topic question. In your experience, is it good to turn off Prefetch, Superfetch and Defrag when using a solid state drive? Answers on line are often very contradictory! My SSDs are Inel 80 gig.

1) it is not off topic, if we can gather all thing on first page.
2) i am not god or a SSD designer, i know only from many readings online.

3) defrag is clearly stupid for SSD (defrag is here to fight random read which is bad for HDD but wonderful for SSD).
TRIM would be a kind of defrag version for SSD (having blocks of 512KB free is better than having them fragmented - but the OS does not know the real physical layout of the data (specially when having free blocks is useful: intel/indilinx), it can be very different from the logical layout it knows). So if you have trim utilities (e.g. with OCZ/indilinx drive or the intel one...), then use it!

4) I dont see problem with fetch and superfetch. Though it may not bring much (of course, memory is faster than the SSD, but may be the SSD is fast enough).
you can try to bench software lauch time (the one you use often)
activating and disactivating it, and judge based on it. I never read data about that.

[sjthinkpader]

SSD doesn't need Superfetch or file Indexing due to the very fast read time. Frequent Write is bad for flash also. So you want to turn them off in the System/Services list. Superfetch is not applicable to WinXP, only Win7.

If you are using Win7 with HDD, you can still take advantage of flash/SSD. Just plug in a PCIExpress or USB flash device and turn on the Readyboost function in Device Properties.

[bgx]

On my Sandisk 32GB SSD SDU5B-032G, 4K random write came to between 14-16 IO/Sec.

The WD1600BEVE 160GB, 5400rpm, 8MB cache HDD in my T42p did 320 IO/Sec for the same 4KB random write.

Thx!

from the figures, i would say the sandisk is jmicron based + SLC? figures seem to high for MLC Jmicron and too low for samsung controller. If you can give us the 4K read, it would be great too!

For the WD HDD, the figures seems high (a velociraptor is supposed to do 218 here), but my own raptor (not veloci) scores around 400 (which is also way too high). Anyway, the numbers may depend on a lot of things (we dont test on totally empty HDD, it depends on the partition we use etc). Hence i would advise not to take this numbers too seriously, but it shows the ballpark we are in.

[bgx]

Have anyone tried the new RunCore Pro IV?
1.8" size is available both as ZIF and IDE, both should fit X40/41 (ZIF need adapter).
16-128GB. MLC flash, Indilinx controller.
http://www.mydigitaldiscount.com list most of them as preorder except 32GB ZIF.

You got me very tempted.
I emailed a german site who have it, and they told me it is bill to order, and should take a good week to got it, and they have no idea which controller it uses.

I emailed RunCore (chinese based) to know whether this version has really a indilinx inside. the 128/256 GB seems better have they have 64MB of SDRAM instead of 32MB for the 32/64GB version, and probably twice as many channels for better performance (and higher consumption too?), but quite expansive (500+ euros). I ll buy the 64 GB version, but i will try to get it from Singapore as i am here for a short time. Else, it will cost me 250 euros with shipping included...

Anyway, should got one in a month!

[sjthinkpader]

....i would say the sandisk is jmicron based + SLC? figures seem to high for MLC Jmicron and too low for samsung controller. ...

This may be M-System design still. The Sandisk board level 8GB SSD made for Acer netbooks is much slower and may be based on JMicron, Phison or other lower cost controller.

[syhead]

this thread deserves to be on the top of the section

not that old HDD clicking thread....

[bgx]

After a long silence, i updated the first post with runcore and lenovo SSD.
I got both of them, so could test on them...

lenovo is a good pick, cheap from ebay, and good power consumption, even if it is not a win/win vs a HDD. in particular, writing is slower than a HDD.

runcore is a great pick for performance on X41 (read/write speed + trim), but does not work on X40, consumes more than a HDD.

Overall, i was a bit disapointed by the speed of the SSD.
Truth is, i hate waiting after my comp, so all the program i used were lightweight already (no stupidly big WMP - i used VLC or MPC instead etc), so the SSD does not change much in that respect.

Fact is, these conputers are not the fastest on earth . Werent when they were realseased, and still are not, so it really feel the proc is too slow to get up to the speed of the SSD (opening a program needs 2 things: readin data from the disk and running the program).

----

For what i do - writing latex files, reading video, looking at my email/internet, only emails get a good boost from the SSD (opening my big folder in OO very fast).
I appreciate the 5% better autonomy also, even though it does not change much thing in my life.

I am almost always on or sleep mode.
Thus I rarely start windows (which would be boosted by a SSD), and i also rarely use the hibernate mode (which is slower with the lenovo SSD)...