Crucial m4 128GB vs OCZ Vertex 3 MaxIOPs 120GB

[Antioch]

I'm looking to buy a new SSD for use as my OS drive (Win7) with my Sandy Bridge CPU. I'm having a tough time deciding which of the two to pick.

If it were between the *plain* V3 and the m4 I'd grab the m4 as it's got better random 4k read/writes (but not as great sequential R/W) -- but I haven't been able to find any benchmarks that show how the V3 MaxIOPs version changes the picture.

If the V3 MaxIOPs delivers as good or better random 4k R/W then I suppose it'd be the drive to get (as it would also have better sequential R/W).

Does anyone have any thoughts on this issue?

Thanks.

[johnw]

I think the performance of the m4 and the V3 are sufficiently close that I would make my decision based on overall quality and company reputation, and there Crucial beats OCZ by a mile.

[Antioch]

Well m4 128 and vertex 3 120 *vanilla* are close enough in overall "realistic" system performance (not benchmarks), sure. But I'm curious how much the maxiops v3 drive will improve the random 4k RW.

[bluestang]

Also price...the M4 128 is much cheaper than the normal V3 120. I was holding off getting the M4 until the Agility 3 came out, but pricing for the 60 and 120 of them are more than the M4 64 and 128. I guess I'll be getting the M4...now it's just a matter of 2x64 in R0 or a single 128???

[Antioch]

Also price...the M4 128 is much cheaper than the normal V3 120. I was holding off getting the M4 until the Agility 3 came out, but pricing for the 60 and 120 of them are more than the M4 64 and 128. I guess I'll be getting the M4...now it's just a matter of 2x64 in R0 or a single 128???

I've read that the 64Gb aren't very good even in RAID0 (see this thread: http://www.xtremesystems.org/forums/...=268711&page=8 )

That and what Anand (from anandtech.com) said about the M4's garbage collection being a bit poor would make me shy away from using M4s in RAID as they really do need the TRIM command it seems.

As such I'm leaning towards an m4 128Gb right now but if I see any V3 MI 120GB reviews this weekend I may change my mind.

Anyone care to chime in on this?

[Nizzen]

2x Vertex 3 MI 120gb Raid-0 VS 2x Crucial m4 128gb Raid-0

4k Random read 16GB testfile, 1worker, QD=1, runtime 30s

[Nizzen]

2x Vertex 3 MI 120gb Raid-0 VS 2x Crucial m4 128gb Raid-0

[Lu(ky]

I would not mind seeing some 2 x 128gb C300 with these benches...

[Antioch]

Thanks for the benches. Although I am curious about a single drive, non-raid. I'm not sure you could simply "divide the bench results by 2" and arrive at a true comparison -- though the RAID0 is likely a hint.

[johnw]

That and what Anand (from anandtech.com) said about the M4's garbage collection being a bit poor would make me shy away from using M4s in RAID as they really do need the TRIM command it seems.

Anand's test of GC recovery is not very realistic. He uses HD Tach to do a 100+ GB sequential write to the SSD and looks at how the write speed degrades during that humongous write. Not many people do that to their SSDs. Also, he does the same thing to Sandforce SSDs, which is absurd, since HD Tach writes zeros, and Sandforce ends up writing almost nothing to the flash when you give it 100+ GB of zeros.

If you give the m4 a more realistic workload (for example, IOMeter with a reasonable mix), the GC works well.

[Antioch]

If you give the m4 a more realistic workload (for example, IOMeter with a reasonable mix), the GC works well.

Do you have an m4 you've been using? If so which one, if not what do you base the GC claims on? Just curious.

[mbreslin]

Anand got a defective drive as his first experience with the C300 and has been pretty much down on crucial ever since. His reviews put a very bad taste in many people's mouth and it took many benchmarks (away from Anand's test bench) for people to catch on that the drives really are excellent. He says the M4's GC is poor like the C300 but I've ran 8 in raid 0 for a year and individually iometer still spits out nearly exact numbers as when they were new. I haven't touched an M4 but if I had to bet money I'd bet they were excellent. Not the clear winner in sequentials (not an issue for me but if your use case requires large file transfers to be fast you may want to consider another drive) but otherwise excellent drives.

[Antioch]

Yes, that's the picture I've come up - that the m4 is likely a solid drive. I'm leaning heavily towards it at the moment but I'm still curious to see the V3 MI 120GB performance notes. Honestly, I don't know if it's worth the hassle with SandForce at the moment - still needs more testing and tweaking and I'm a bit sick of that.

That being said, I'm curious if there is an "SSD Toolbox" for the m4s of any sort? Something akin to Intel's toolbox that offers you the ability to force TRIM on the entire drive? I'm in Win7 so it wouldn't be necessary but I'd like to plan for the worst just in case!

Thanks

[johnw]

Do you have an m4 you've been using? If so which one, if not what do you base the GC claims on? Just curious.

I'm basing it on the C300. Anand made the same claims about the C300, but GC works fine for me on the C300. Also, there are plenty of people running the C300 in RAID 0, and I have not seen many reports of problems attributable to GC.

[Computurd]

curious that the M4 beats the vertex 3 in the "2010 storage bench" , yet anand switches to a new bench the "2011 storage bench" and the V3 wins handily. hard to tell these days. we need more of these drives in the hands of us enthusiasts to come to our own conclusions

[Antioch]

Well, take any intro to computer architecture course and the *first* thing they teach you (same with any good texts on the subject) is that any benchmark can be twisted to show whatever you want and that you shouldn't really rely on them completely.

Lies, damn lies, and statistics, right?

[Computurd]

Lies, damn lies,

i wouldn't go that far! just saying we need to conduct our own real world testing.
could be a coincidence, or maybe his new bench is just seriously a better representation of real life usage than the prior one. or he could be wrong and it could be worse than the previous version.


so that would leave it up to us

[Ao1]

The difficulty in benchmarking SF drives is you need to be in a used state, but then you have to be careful you don't trigger throttling.

Assuming this diagram is correct it seems a bit strange that light/ modest use is plotted above the life curve. That seems to imply that over time even light and modest use will result in throttling.

Why SF drives need to throttle at all is also strange considering they compress data and supposedly have such as great wear levelling capability.

[Ao1]

Thanks for the benches. Although I am curious about a single drive, non-raid. I'm not sure you could simply "divide the bench results by 2" and arrive at a true comparison -- though the RAID0 is likely a hint.

Not at queue depth one with a 4k xfer. I doubt the 4K figures would be much different with a single drive.

[bluestang]

@ Nizzen,
What's the possibility of you providing some single drive benches???
Thanks!!!

[johnw]

Assuming this diagram is correct it seems a bit strange that light/ modest use is plotted above the life curve. That seems to imply that over time even light and modest use will result in throttling.

The diagram makes sense to me. It will allow you to do anything you like as long as you do not go below the straight line. If you do get near the straight line, then it will start throttling the write speed. At least, that is what the diagram is meant to show. I'm not sure if the actual Sandforce firmware implementation does exactly what the diagram portrays.

[Ao1]

Not sure I follow, but I don't understand why the drives need to be throttled in the 1st place considering that SF claim their controllers have the best wear levelling in the industry. My take on that graph is that write activity needs to be ether below or on that life curve line - not above.

I also don't understand how throttling works. For argument/ simplicities sake is it like this?

You are allowed to write 20GB a day. The next day you write 30GB. You have exceeded your daily write allowance by 10GB. Throttling then kicks in over the next few days to make sure writes are slow enough that it is impossible generate 20GB of writes. After a few days you are back on an average of 20GB. Throttling is then turned off.

How much do you need to write for throttling to kick in? How long does the drive remain throttled? Is the extent of throttling directly correlated to amount of writes that have been exceeded?

Does a secure erase clear the throttling?

[Anvil]

That diagram could have been presented in another way but...

You aren't allowed to go below the green "Life curve" line, excessive (accumulated) writing would be throttled if it was to "break the line".
Secure Erasing/Cleaning looks to work but there is a finite number of "clean-cycles" as well I'd presume.

I agree to some extent that the "roof" for daily writes should be well above the limits that e.g. Intel sets for the daily average.
Unless the SF drives are capable of writing the same amount of data it would mean that there is something fishy going on, compression/write amplification should easily make the drive last "years" longer than other controllers *if* using NAND of the same quality.

[Ao1]

I excluded poorer quality NAND from the equation as everyone using SF controllers claim to use only top grade NAND. That is where I am confused. I remember Anandtech saying that the high cost of the controllers could be offset with lower quality NAND due to its enhanced error correction and reduced write rate to the NAND via compression, yet top grade NAND is used, so why do the drives need to be throttled?

[Anvil]

That is a good question
Assuming that quality was on par with other drives (and I do believe it is) and the daily limit was somewhere between 20-30GB one shouldn't notice the throttling anyways and so the only reason to having that limiter was if the drive was used in the "wrong" environment, like e.g. servers.

I'd say that write throttling is OK as long as I don't get hit , I don't think I have, yet.
I've been hit by "steady-state" though on the 1XXX SF controller and depending on the number of drives the only option is cleaning.
(not really an issue with 3-4 drives but for 1-2 small drives like the 60GB the impact is huge, imho)

I'd like to have the option to select my own "life-curve" on the SF drives, I'd rather use the drives for 2-3 years with "unlimited" writes than being hindered by "life- preserving" limits that go beyond the useful life of the drive.
I'd say 2-3 years is sufficient for me, by that time there would be better alternatives with better performance, larger capacity,..., well most things improve over a 2-3 year period.
It does look like the SF-2XXX controller is way better than the 1XXX series and so those options may be obsolete, I'd still like that option for my ageing SF drives.

If they last 2x the Intels I'd just don't know what to use them for, if I can't use them while they are in their prime of their lives, whats the use?