Mmm.. math error there.

The count is now an average of 34 erase cycles across all of the flash, after 76 power-on hours over 10 weeks or so. It's a notebook, and it's only powered on when I'm actively using it.

So, let's try that math again now:

10 weeks / 34 cycles = .29 weeks/cycle.
The flash chips are rated for 10000 cycles minimum,
so that's 10000 * .29 = 2900 weeks of use before it dies.

Which still works out to more than 50 years.

Another way is to do it by hours of use:
76 hours / 34 cycles = 2.2 hours/cycle,
so that's 10000 * 2.2 = 22000 hours of use before it dies.

Figure worst case of perhaps 40 hours of use per week,
that's 40 * 52weeks = 2080 hours per year.
At that rate, the drive would wear out
in 22000 / 2080 = 10.5 years of use.

I think I might have *one* other drive in the house that's survived 10 years, but it hasn't had nearly that amount of use.

Cheers

Which SMART attribute tells you the number of cycles? Some of the ones being reported by mine are pretty weird like 0xCE - Flying Height.

I tried looking for information on SMART attributes in the OCZ forum and all I could find was a handful of threads where a moderator keeps saying that SMART doesn't work and won't ever work for SSDs. The numbers being returned by the SMART commands are nothing you should be looking at. Go away. Thread closed.

Helpful...

These two smart data fields have been documented by OCZ:

http://www.ocztechnologyforum.com/forum/showthread.php?t=57898

Field 208 (0xD0) is "average block erase count".
Not sure what the scaling factor is, though, since it's
only an 8-bit field and the cells are good for 10000 cycles minimum.

208 Spin_Buzz 0x0000 031 000 000 Old_age Offline In_the_past 0

Field 209 (0xD1) is "Remaining drive life in %% by erase count".
So probably (field_208 / 10000) * 100%, except we know field_208 hasn't enough bits.

209 Offline_Seek_Performnce 0x0000 100 000 000 Old_age Offline In_the_past 0

Fields 001 (read error rate) and 009 (power on hours) are also valid.

Ahh. Thanks for the link. It would be helpful if Indilinx actually published more details on the various attributes.

0xD0 is 13 and 0xD1 is 100 for mine. Bit strange that its done 13 erases though since I've only installed Windows on it the other day. Reading the forum also shows some odd values. Its at 100% anyway for 0xD1 so I guess its all good still.

ooo, 320GB Intel SSDs. Hopefully the rumor is true of better capacities and noticeably discounted prices. 320gb would be enough to let me swap out the Velociraptor and gain a little bit of space.

That 320GB won't be cheap though! The prices of the 250GB should be more reasonable however.

In the traditions of companies aiming for overclockers, OCZ are releasing a special matched chip version of the Vertex which is a whole 10% faster.

Ok, I read it, but it's not making explicit sense.

Are SSD's sized in real (^2) or fake (decimal) GBs? Before or after formatting?

It actually has real powers of 2 GB in there but the flash controller uses some for wear leveling and you also lose a bit from formatting etc... Basically what OCZ sell it at is what usable space there is after formatting.

So comparing a 60GB SSD and a 60GB HDD, is it apples to apples?

Do SSD makers subscribe to the same ^10 size ratings that HDD makers use?

The idea is that a "60GB" OCZ Vertex is equivalent to a "64GB" model from any other manufacturer.

And they'll probably stop trying to be honest about it in the future, since so many people seem to have difficulty with the concept.

It's mostly a point of curiosity, for me. As long as all SSD's are rated the same way.

Though I am of the camp where HDD's should have been rated in GB^2s.

Based on your recommendation and the Anandtech article, I got a 120GB Vertex for a new install. So far, I love it. I was worried about space, but that hasn't been a problem yet (still have about 25GB free). It is stupidly fast. I hate having to go back to the old disk setup (RAID0 7200RPM drives with a VelociRaptor data drive - 1.3TB in space) because the speed difference is very noticable. Personally, I think it is worth the size limitations and cost for the performance boost.

Are you chaps talking about having purchased the SLC or MLC version of the 120GB Vertex drive.

Dunno about you guys, but I'm not that wealthy.

Definitely the original MLC drive!

Cheers

I assume this is important for Linux partitions too.
Can you give me any pointers how to optimally fdisk an SSD?

Thanks,
Pim

Just put the first partition at a sector offset of 1024, rather than the default of 63. In practice, even an offset of 64 is good enough (any even multiple of 4KBytes).

The default of 63 is horrid, since it means every 4KB write to the disk can potentially require an internal READ/MOVE/MODIFY/WRITE cycle. Really bad for performance, that!

How would I do that with fdisk? It asks me for a start cylinder which usually is a unit of 255*63*512 bytes.

Does it mean changing the number of sectors per track?

Thanks,
Pim

I think you need fdisk's "expert mode" to make partitions on non-cylinder boundaries. (But I don't have Linux in front of me to check exactly how.)

Peter

The "u" command will switch fdisk to sector units.

Would changing the number of heads to 128 and the number of sectors per track to 32 do the same trick? Cylinders would then be multiples of 4KB.

Given that OS installers usually pick cylinder boundaries, I could then have the OS installer do the partitioning.

Thanks,
Pim

Has anyone come across 3.5 inch SATA/SAS SSD's they could recommend? Got some poweredges I'd like to convert.

No. But you could modify a drive carrier with different mounting holes to put the connections in the right place. I've seen this done on a lab system.

What is your budget? I've used STEC ZEUS-IOPS (stupid superscript name) 3.5" FC SSDs before and they worked well. They do a SATA and a SAS version as well.

This might work for you.

They mostly seem to be 2.5" form factor right now. But some companies (eg. OCZ) also market conversion brackets for 3.5" bays, which will work with any 2.5" drive.

Cheers

Just received the OCZ Vertex 32GB drive I ordered.

First thing to do was to see whether it is as fast as advertised.

What I'm seeing is that the read speed greatly depends on the SATA controller. I tested on three different PC's using this command:

dd if=/dev/sdX of=/dev/null bs=1024k count=8192

(I made sure to flush the cache by first dd'ing another drive)

Results:
ICH7, ahci: 78 MB/s
ICH7, ata_piix: 125 MB/s
ICH8, ahci: 78 MB/s
ICH8, ata_piix: 132 MB/s
ICH10, ahci: 78 MB/s
ICH10, ata_piix: 187 MB/s
sil3132, ahci: 72 MB,s
Marvell 88SE6121, pata_marvell: 138 MB/s

Interesting to see is how bad AHCI performs. Wasn't AHCI supposed to be "superior"?

For regular SATA drives, AHCI versus IDE mode does not seem to matter, at least for sequential read speeds, but choosing IDE mode would disable NCQ. Would that slow down regular SATA drives?

Thanks,
Pim

A simple dd won't saturate the drive, though it might have a decent shot if drive "read-ahead" is enabled.

Things for you to do:

1) upgrade to OCZ firmware 1.3 immediately, if not already on the unit.

2) ensure the drive is set for highest performance:
hdparm -W1 -A1 /dev/sdx

3) stop benchmarking, and just enjoy the speed!

As for AHCI, its pluses are (1) it's now a multi-vendor standard, which means new devices that implement it won't need to write drivers from scratch, and (2) it fully supports NCQ (multiple overlapping R/W commands at once to the drive). Point (2) helps servers much more than desktops/notebooks, and benchmarks generally show that the OCZ Vertex doesn't really take much advantage of it.

Cheers

Cheers

Just for everybody's information:

Intel has released their new "Postville" SSD's. These SSD's use the smaller 34 nanometer chips, compared to the previous models which used 50 nanometer chips. This has made the Intel SSD's even faster and... considerably cheaper as well! The 80 GB model now costs less than the 60 GB OCZ Vertex model does! Or should I say did, because OCZ has already responded by lowering their prices as well.

On this (Dutch) site you can find a table which shows an overview of both the Intel and OCZ SSD's at the moment, with the capacity, price per GB and recommended retail price.

Mark, that OCZ Agility range (also in the table), is that one comparable to the Vertex series?

Yes, the Agility has the same Indilinx Barefoot controller chip, but uses cheaper (usually slower) flash chips than the Vertex.

I also seem to recall that it has only half the channels (bandwidth) to/from the controller, but I cannot find where I thought I once read that. EDIT: Oh, perhaps something in this rather confusing thread.

In practice, The Agility normally performs so close to the Vertex that there's just no point in paying more for a Vertex now. Or at at least that's true for a SATA1 interface like the one in my notebook here.

And gotta love the pricing on the Intel G2 SSDs, too. Things are excellent in this space, and still getting even better very quickly!

Cheers