Hardware Crucial M4 256GB Review

Crucial's new SATA 6Gbps SSD features 25nm NAND. Is it the fastest thing on a SATA port?

http://www.bit-tech.net/hardware/storage/2011/05/16/crucial-m4-256gb-review/1

 

My 512GB M4 is very nice indeed. That would seem to me to be one of the few reasons to get one over the older C300s.
As a suggestion, it would be nice to see some real-world ssd benchmarks thrown in (stuff like compressing a big archive, dealing with raw video etc), since big MB/sec numbers are kiiinda helpful, but don't tell me that much about improvements. For example, I appreciate that my new SSD has slightly slower random reads than my old one - but how does that affect me in actual real world usage?

 

Unless I'm missing something, the C300 looks better than this drive, not obsolete; random reads are the key reason to get an SSD, and this drive is worse at them.

 

That's what my suggestion was about. Random reads WERE the key reason to get an SSD, simply because hard drives were so crap at them - a normal 7200rpm hard drive might give you 0.5MB/sec because of the horrible seek times.

But what is the difference between 320x faster than a hard drive and 400x faster than a hard drive? You're getting a titanic advantage either way.

 

From the article.

I'm assuming this is just a cut n paste error; otherwise, how do you fill a 256GB drive with 100GB of data? Wouldn't half-filling it change the test results, albeit slightly?

Regards

Spreadie

 

I agree with this. Benchmarks are great at showing how products compare but seeing how it translates into real world performance is more useful imo.

 

In real world tests it does better - e.g. time to load windows, time to load games - it's very much tuned to light work loads - in those the C300 can't compete, in fact it often beats the vertex 3, e.g. http://www.tomshardware.com/reviews/intel-ssd-320-crucial-m4-realssd-c400,2908-11.html

 

I am glad to see that the prices are moving in the downwards direction. ~$1.95/Gb is much easier to swallow than last year which was around $2.50-3.00/Gb.

 

Hi all

We've played with many different real-world benchmarks - virus scans of 50GB folders, game load times, Windows boot times, compressing data and so on - but found them all to be very inconsistent. Even disabling all the Windows caching tech we can doesn't reduce the variability, and the more Windows features we disable, the less real-world the test becomes anyway. After all, what does it matter if one storage device is faster than another if Windows caches the type of data you're testing in memory anyway?

As ever, we're open to suggestions, but we're still committed to only posting reliable, consistent and relevant data to form our analysis.

As to the TRIM question: correct!

 

The irony of receiving my CustomPC today to find this review is in it on the same day ^_^

 

Well, that depends on precisely what you're doing. The caching issue would only come in if you were repeatedly opening the test file - so if you were looping, you'd end up with issues - it wouldn't happen in regular windows usage.

I'm not sure I'm completely happy with your argument - none of the examples you gave are particularly demonstrative of something that would be helped by a fast ssd - Windows is limited by stuff like loading drivers (try booting before and after nVidia drivers are installed for an example), whilst game load times are limited for similar reasons - and are generally fast enough that a five percent improvement would fall under experimental error.
Compressing data is cpu-intensive - and throughput is generally far lower than the actual read-write speeds because of this. Only if you're compressing at 'store data' speed will you achieve speeds near SSD saturation - and that's not really compression.

If we've got to a situation where SSDs can't be differentiated from each other in real-world use by bit-tech (note that they can by other sites), then surely bit-tech should be placing a much greater emphasis on price than on artificial performance benchmarks, which don't reflect real world performance accurately? Maybe even the differences in how trim and garbage collection are performed on different SSDs, which can lead to significant performance differences?

 

Sadly, on a benchmark rig, we do tend to run tests a number of times. Thus, caching becomes an issue. We've experimented with file transfer tests too, transferring sets of data from a steam folder, windows folder and media folder, but found that results were not repeatable enough to be reliable.

We used to test windows boot times using Vista, where there was a noticeable difference between drives, at least in the first gen, although it was only 2-3 seconds. However, since switching to Windows 7 testing boot times has gotten much more difficult; in the end we found all the SATA 6Gbps were booting Windows in roughly the same amount of time. Even then there was a degree of variability in our results, by as much a 3 seconds. I'd love to include these tests, but I simply wouldn't be confident of the accuracy of the results.

I've also experimented with PC Mark Vantage, which seems very popular with other sites, but again, results were very variable, by as much as a few hundred points. I'll be checking out PC Mark 07 soon to see if that's more reliable, but wasn't confident of PC Mark's results to include those numbers.

A few of the real world benchmarks used by other sites (Anand, Storage Review) use a technique of bus skimming; they basically record all the data moving across the SATA bus over a set period (in which they play games, unzip files, etc), and are then able to reproduce it exactly to the drive in question. Sadly the software used for bus-skimming is super expensive, and isn't something we can afford, especially as the benchmarks we have used show the differences in performance, albeit at a theoretical level.

It's been frustrating, as bit-tech typically prides itself on real world benchmarks, and despite my extensive testing, we've been unable to pin down a reliable and reproducible test for them. In the end we opted for AS-SSD and ATTO, as they produce repeatable numbers that are easy to understand at a glance, and are a good indication of overall performance at many levels and with differing workloads. They also have the benefit of being easily performed at home by end users to see how their drives measure up.

Hope this helps as an explanation to our testing methods.

 

We filled the drive with data; the 100GB is the size of the install on our SSD test rig; it's just consta-copied until the drive is full.

 

Ah, thanks for the explanation.

 

I'm surprised the issue of price hasnt been mentioned. I'm in the market for 256GB SSD, and the Crucial M4 is actually the cheapest drive on Scan for near Vertex 3 speeds!

 

I think Crucial has decided that it can't compete this generation on stated speeds (which sadly, are what do sell a lot of drives), but is instead pushing the M4 as the better value proposition. Certainly very tempting.

 

i have to agree with the above, all you should be interested in, does TRIM and GC work Correctly, does it do good random read and Write numbers at 4KB and 1MB and is it got an good price and is it reliable

1. is it reliable (Trim and GC correctly working)
2. price
3. random read and Write speeds are more then an HDD (around 20MB/s or better, HDDs can do at best 0.2MB/s to 1MB/s)

as long as its not an First gen samsung (or second gen samsung in some cases) or an JMmicron SSD or Kingston (as they made so many random names and bad ssd makes inside you really do not always know what your buying so best to just ingore them unless you do more research on every same names SSD that kingston do)

 

Why use (3 x 2GB DIMMs) here when in all your P67 builds you use either 2...(or maybe 4) sticks?
Triple set doesn't really make sense on a P67, does it?

 

Also, you have a P8P67 test on the site, but none of the P8P67-M...
Is there a difference in performance or is it "just" size?

 

Happy with my M4 512 so far, after my c300256 failed after 8 months.

The ud9s marvel controller in ahci is limiting mine to 380 read and 235 write though.

In some respects the c300 is better, but i went for size also.