Plextor M5 Pro: an other step forward

[felix_w]

yes

...do you have a review already???


Edit : so tomorrow the review will be available at tweaktown?

[johnw]

IThe Marvell controllers have always ran a low slower than SF when the drive is half full...not so much of an issue now.

I've always been skeptical of that data since the only tests tweaktown runs on the SSDs when they have some data is PCMark Vantage, and the compressibility of the data that PCMV writes is not well-specified.

It would be better if instead of (or in addition to) PCMV, tweaktown ran ASU with 100% incompressible data, or AS-SSD, on the drives when they are 25%, 50%, and 75% full, etc.

It would be interesting to see how AS-SSD and ASU benchmarks vary as the SSDs are filled up.

[Callsign_Vega]

Could you guys say if the 128GB will be similar in speed to the 256GB or will it be much slower like on certain series of drives?

[johnw]

Could you guys say if the 128GB will be similar in speed to the 256GB or will it be much slower like on certain series of drives?

Just look at the specs posted earlier in this thread. The sequential write speed is 340MB/s for 128GB and 450MB/s for 256GB. That is the main difference. The 128GB also has a smidge less 4K random IOPS at high QD than the 256GB, but that is a totally irrelevant spec for most usage patterns.

[johnw]

The Plextor M5Pro just showed up on Plextor's website:

http://www.goplextor.com/asia/index.php/ssd/m5-pro

I love the comparison charts. "E Brand, E series (SF)" I'm not sure which of the Sandforce SSD makers that refers to (possibly SanDisk Extreme?). Then "S Brand, 8 Series (SS)" is obviously Samsung 830 and "O Brand, V series (ID)" must be the Vertex 4.

[Highendtoys]

I've always been skeptical of that data since the only tests tweaktown runs on the SSDs when they have some data is PCMark Vantage, and the compressibility of the data that PCMV writes is not well-specified.

It would be better if instead of (or in addition to) PCMV, tweaktown ran ASU with 100% incompressible data, or AS-SSD, on the drives when they are 25%, 50%, and 75% full, etc.

It would be interesting to see how AS-SSD and ASU benchmarks vary as the SSDs are filled up.

The problem with those tests are they show peak performance and not average performance. Peak numbers don't tell you anything other than the best possible outcome. The next PC Mark will be a lot different and will aid in what I'm doing. Then there is the my own trace based tests, as soon as I can convince Bus Trace to send me the software I'll start thinking about building my own PC Mark HDD tests and call it ChrisMark

[Callsign_Vega]

Just look at the specs posted earlier in this thread. The sequential write speed is 340MB/s for 128GB and 450MB/s for 256GB. That is the main difference. The 128GB also has a smidge less 4K random IOPS at high QD than the 256GB, but that is a totally irrelevant spec for most usage patterns.

From what I've seen, real world testing and specs and what something should do don't match up that well all the time.

[johnw]

The problem with those tests are they show peak performance and not average performance. Peak numbers don't tell you anything other than the best possible outcome.

Wrong. The ASU and AS-SSD numbers do tell you something. They tell you what will be the sequential read and write speeds, 4K read and write speeds, etc. when they are measured with a 1GB test file of random data (AS-SSD) or a 1-32GB test file of 100% incompressible, 0% incompressible, etc. (ASU, depending) under the conditions that they are run for the SSD. They tell you EXACTLY what you will find under those conditions.

Unlike PCMV, which is a black box. Under some unspecified conditions, writing some unspecified type of data, out will pop some number that has no meaning in itself.

I'll take AS-SSD or ASU results, which I can understand, over PCMV results, which are totally opaque, every time.

[johnw]

From what I've seen, real world testing and specs and what something should do don't match up that well all the time.

If you look at Plextor, Samsung, or Crucial specs, they tend to match up pretty well to what you can measure yourself with AS-SSD or ASU. If you look at the specs for Sandforce SSDs, you have to be careful because many of the sellers will tout specs for unrealistic, highly compressible data.

[Highendtoys]

So take Anvil and AS SSD and test a drive that is 50% full and tell me what you see. I don't take many things seriously but testing is one of them. Been there and done that for weeks at a time.

What give the advantage to SF is the type of data stored on the drive. The same flash at 50% full is the same speed, the SF controller manipulates the data on the flash so less of it is stored and that is why SF gets the nod in that test. No one keeps just compressible or incompressible data on their boot drive. When I built the test I used both compressible and incompressible and use the same ratio of what I had on my notebook at the time.

As for what the manufactures spec their drive at, we all know that is hocus pocus because those numbers are based on an empty drive. No one has an empty boot drive. Some SF based drives are marketed with both compressible and incompressible speeds.

[johnw]

So take Anvil and AS SSD and test a drive that is 50% full and tell me what you see. I don't take many things seriously but testing is one of them. Been there and done that for weeks at a time.

What give the advantage to SF is the type of data stored on the drive. The same flash at 50% full is the same speed, the SF controller manipulates the data on the flash so less of it is stored and that is why SF gets the nod in that test. No one keeps just compressible or incompressible data on their boot drive. When I built the test I used both compressible and incompressible and use the same ratio of what I had on my notebook at the time.

Since I don't have an M5P, I cannot run ASU on it. That is what tweaktown should do. I already said that. Also, you seem to have missed the point. Yes, running ASU under one set of conditions is not very representative of real conditions. So don't just run it under one set of non-real conditions. Run it under several different conditions. Run it with the drive 75% full of data. Run it after hitting the drive with a lot of sequential or random writes. Run it after TRIM. ASU is a tool that gives precise and well-understood results, but it is up to the person running the tool to pre-condition the SSD into various interesting states before running ASU.

As for your claim of Sandforce "advantage", as I already said, I am skeptical. I am well aware of how Sandforce SSDs operate. What we do not know is exactly how PCMark Vantage operates. What types of data does it write? How much? In what percentage sequential? What percentage random? How much 4K, 16K, 32K, 128K, 512K?

The problem is the PCMV is just a black box, and no one reading the reviews has any clue about what it is actually measuring. All we have is a number that pops out that is meaningless by itself.

[kaktus1907]

user review on a laptop hence crap 4Ks

http://translate.google.com/translat...B8%AC%E8%A9%A6

[felix_w]

Images are not shown, is there any chance you saved them ?

[kaktus1907]

sure

[Highendtoys]

First review is up. http://www.tweaktown.com/reviews/488...iew/index.html

[johnw]

First review is up. http://www.tweaktown.com/reviews/488...iew/index.html

The included desktop adapter bracket offsets the drive to the side so these plug in points are in line with 3.5" form factor drives.

tweaktown has been making erroneous claims like this for multiple SSD mounts for months or years now.

In reality, it is IMPOSSIBLE to make a purely mechanical 2.5" to 3.5" adapter bracket, including both sidewalls, that lines up the SATA connectors with a standard 3.5" form factor. This is self-evident since if you line up 2.5" and 3.5" drives with their edges flush, the SATA connectors line up. If you then put a gap, such as the width of a piece of metal (wide enough for a tapped mounting hole), on the side of the 2.5" drive, then the SATA connectors no longer line up.

[Callsign_Vega]

First review is up. http://www.tweaktown.com/reviews/488...iew/index.html

Hm. M5P not as fast as I thought it was going to be with that Marvell controller and toggle. It loses to quite a few drives in a few tests.

[Highendtoys]

Sorry, I guess you are correct. To the naked eye it appears they line up.

The point is still valid though, the drive is offset to the side so you can install your cables in a nicer and easier manner.

Thanks for your trolltastic post.

[johnw]

Sorry, I guess you are correct.

You "guess" I am correct? Now we see why you have been printing reviews for years with incorrect and misleading information. When presented with facts and clear evidence, you "guess". And then you accuse me of trolling for correcting your mistakes? Incredible. I did not think it was possible for my opinion of tweaktown to go lower, but I was obviously wrong. Mea culpa.

[Highendtoys]

You know what johnw. If you think you can do a better job then get off your ass and do it. I spend 18 hours a day 7 days a week doing this. Instead of trying to bust my balls over something pointless you should simply say "thank you". The issue has been noted and will be corrected.

It's people like you that make it difficult for me to come to forums and chat to readers. I thought it would be a little different here at XS since it is a community of real enthusiasts.

[Ao1]

I don't think there is anything wrong with TRIM on the M5Pro. If you monitor with hIOmon you can see TRIM commands being executed as files get deleted. Also I just created a test file with Iometer that filled the drive. I then ran 4K random writes for half an hour or so. I then deleted the test file, waited a couple of minutes and ran AS SSD.



4K reads seem to have taken a bit of a hit, but not bad considering what the drive had just been exposed too.

The drive also uses idle time GC btw.

[Callsign_Vega]

Ya, bracket alignment seems a little trivial to me. I thank you for your hard work Highend. I know doing tests like yours above take a huge amount of time. I look forward to your updates on the M5P. As of right now, like you said there are some unanswered questions and the drive doesn't seem to be much better if at all over the M3P.

[johnw]

If you think you can do a better job then get off your ass and do it.

I just did. I presented facts with clear supporting evidence to you so you could correct your mistake. Instead of accusing me of trolling and saying that the issue is pointless, any decent reviewer would have said, "Oops, I made a mistake. Thank you for correcting it. I will not make that mistake again."

It is not pointless at all. Anyone who wants to install a 2.5" drive in a 3.5" hot-swap bay or in certain Macs that behave like hot-swap bays would have been mislead by your mistake, and you have been making similar erroneous claims for more than a year (despite having been corrected at least once in the comments of the review).

It is people like you that give web reviews a bad name. Good reviewers place high priority on accuracy. But now we see that tweaktown considers accuracy to be a low priority. I used to think tweaktown just made a lot of mistakes, but now I can see that tweaktown actually does not care about accuracy.

[Ao1]

Hi Highendtoys. How did you create the files in your TRIM test?

[johnw]

4K reads seem to have taken a bit of a hit, but not bad considering what the drive had just been exposed too.

Hmm, did you have a CPU-intensive background task (like 1 thread of Prime95) running during that AS-SSD? The 4K results look like what I see when C-states are in effect (which can be counteracted by running a background task). Do you have a "before" AS-SSD to correspond with the "after" you posted?