"Green" vs. "Red" mSATA adapter speeds

[ssp3]

I've made these tests some time ago. Before I forget which is which, I decided to post them.

The "Red" adapter is based on Marvell chip. Available in Europe under Delock name. Cheapest price in my neck of woods is EUR 32.00 + 5.00 shipping.
The "Green" aka "White box" adapter is based on JMicron JM20330 chip. Available everywhere. Prices do vary. I bought mine on AliExpress. Paid EUR 34.00 for 4 pieces, including shipping .

My test machine was (unsupported) 12" Al PowerBook G4 876 Mhz running 9.2.2 (with my own version of ROM).
I had top cover removed and was able to simply swap the adapters. I used the same 128GB mSATA drive in both tests. Drive formatted by FWB Hard Disk Toolkit 4.5.2 while booted from external FireWire drive.

Marvell




JMicron




A bit later, I installed another drive of the same type/brand in the same machine using JMicron adapter and then re-run the test. It gave me a slightly better read/write numbers.
EDIT. The drive might have been empty and test was probably run while booted from external FireWire drive.

JMicron, drive Nr.2




Now, decide for yourself, whether the extra ~5MB/s speed is worth the extra 20+ Euro per drive 

And yes, I know about the 2 drive problems in G4 desktops.

More later ..

[ssp3]

Part 2.

This time without direct comparisions. Different machines, different drives, different adapters.

Same machine as in my previous post, JMicron JM20330 based adapter, Samsung SM841 128 GB.




TiBook 1GHz, Marvell, Samsung PM830 32 GB.



TiBook 667MHz, JMicron, Samsung PM851 128GB



TiBook 667MHz, JMicron, Sandisk U110 16GB



TiBook 667MHz, original mechanical HDD

[FBz]

Well done. And pleasantly surprised that you weren’t all “Pro-Marvell” as many often seem to be. To me, throughout all of my testing the JMicron JM20330 based adapters are most definitely the better value. Of course it also depends upon the actual JMicron based adapters themselves (as not all are created equal). BUT once you find one that works (like the 40-pin Bribges - in so many situations) the cost /performance ratio simply can’t be ignored… especially if you’ve more than one machine to convert.

However if intentions are only to convert one primary machine and cost is of no concern, opting for the more expensive (but only sometimes faster) Marvells might be the choice for some.

I plan to go back and 3-panel, comparison re-test (with QuickBench 2.0 & 4.0) the Marvell based mSATA sled and the JMicron based 44-pin AS331 V1.5 adapter again, as those tests have become misplaced here in the archives that I call the dungeon. Eventually I also hope to test the “white case” mSATA AND possibly an M.2 based adapter in a Mac mini as well.

Just for grins… here’s a rare “red group” oddity.
“One of these things is not like the other…” 



Once again however, great job!

[ssp3]

I plan to go back and 3-panel, comparison re-test (with QuickBench 2.0 & 4.0) the Marvell based mSATA sled and the JMicron based 44-pin AS331 V1.5 adapter again, as those tests have become misplaced here in the archives that I call the dungeon. Eventually I also hope to test the “white case” mSATA AND possibly an M.2 based adapter in a Mac mini as well.

I'm already doing all those + one other, not that well known, JM20330 based, mSATA->IDE adpter. In various PB G4s and in Mac Mini 1.42.
Also, I don't have v1.5, only v1.3. (I played chinese lottery, ordered AS331 v1.3 but received MCA004 v1.3 instead).

As one TV host often says - 'watch this space' 

[FBz]

What is this that you're using for your benchmarks?

"SCSI Benchmark"?


Lately (in the last year) I have been running either QuickBench 2.0 (preferred) or 4.0... in full 3-panel comparison tests with Borgmac and V.Yakob.
Perhaps I've an aversion to graphs and prefer good ol' numbers for viewing "full spectrum" read/write performance (from 1 KByte up to 100 MByte).

Of course this requires more effort when posting results but I believe it yields a better overall performance "picture".

But by all means, do please provide the source for your benchmark program/app and I'll attempt to run same here, for comparison.

Here's an example reporting: Numeric View, Extended and Extended Plus... via all three QuickBench 2.0 panels.

[ssp3]

What is this that you're using for your benchmarks?

Wow! I thought you knew what it is.

It is ATTO SCSI utility that, surprisingly, when it comes to benchmarking, works with other interfaces too.
It is called ATTO ExpressPro-Tools and I've been using it since Michael Breeden's articles about SCSI at xlr8yourmac turned my attention to it.
The year was 1998. 
http://web.archive.org/web/20000520002145/http://www.xlr8yourmac.com/SCSI/9600/index.html

This is the only tool that has enough test options and meaningful graphical representation of the results. When you use it extensively and learn what shape of graphs represent what (or what change in SCSI drive's mode pages), it becomes almost like an analyzer.

Intech's QuickBench is good, but, IMO, those numbers are hard to navigate when you have many results to compare. I was always wondering why are you sticking to it.


* Garden probably has it + there's my "installed, but not yet personalized" version floating around that I uploaded to some HL server many moons ago. ATTO used strange "drag and drop install" interface, that did not work on all, especially upgraded machines, back in the day, hence the naked app.
The last version was 2.8.2, but I remember using older version (2.3?) for some tasks that 2.8.2 did not allow anymore. I have 2.5, 2.6 and 2.7 too.

[IIO]

does any of these tools output the results in whatever textformat so  that we can make us a proper toosl to create graphical representations of them?

[ssp3]

QuickBench 2 does, but it does not have enough data points to create the graph like this.

[ssp3]

To be honest, I personally don't sweat much about it. 
I know that, given the decent SSD, I can get around 60MB/s on machines with ATA-66 bus and around 90MB/s on machines with ATA-100 bus. Plus or minus. I'm fine with that.

Factors that affect drive speeds / speed tests in OS9 are:
1. Drives themselves, obviously.
2. Whether the drives are empty or have some files on them.
3. Drivers, i.e. formatting tools used.
4. SATA to ATA controllers (Marvell, JMicron or some other)
5. Controller implementation by manufacturer.
6. Interface type - SATA, mSATA or M.2 SATA (in case of JM20330 chip based adapters).
7. ATA controller on Mac motherboard.
8. Bus speed.
9. Processor speed. To some extent.
10. Drive size. Not always, but some earlier models had obvious write speed relation to their size. If it's over 100MB/s, it's not important for OS9 machines anymore.

SIDE NOTE.
My observation so far is that modern SSD drives with huge caches and Turbo Write functions are not much faster (if at all) than drives without them on native ATA buses in OS9.

[IIO]

the main subject is that there are cheapo adapters which give you only 50-70 on a 100 bus.

[ssp3]

the main subject is that there are cheapo adapters which give you only 50-70 on a 100 bus.

Such as? Got any evidence?  I only have met JMicron from the cheap department. Are there any others out there?

Btw, I got slow speeds with red board on 100 bus in one of my experiments when using Drive Setup 1.9.2. As soon as I switched to HTD 4.5.2 (and later to DS 2.1 to verify), the numbers went up.

[ssp3]

Mac Mini time!

Mac Mini 1.42 GHz, 1GB RAM, OS 9.2.2. ROM v.9.6.1 and other relevant parts from Ross' v.9 installer. Drive initialized by HTD 4.5.2, unless noted otherwise.

The same Samsung drive moved between 3 adapters - Delock Marvell based, "White box" JMicron 20330 based and small form factor, "naked", also JMicron 20330 based. The latter seems to be identical to "White box", except everything is squeezed closer together. Sold by everyone and his brother in China and by Delock (!) in Europe. Surprise, surprise 

Marvell




Hmm.. what's that drop in read speeds? Let's check what happens after 2MB.




As can be seen, there is a drop in performance between ~1MB and 5MB file transfer size, but after 5MB speed returns to normal.


JMicron, "White box", 2MB file transfer size




JMicron, "White box", 8MB file transfer size




JMicron, "Naked", 2MB file transfer size




JMicron, "Naked", 8MB file transfer size




The drop is still there.    Let's try Drive Setup 2.1 and another, same model drive.

JMicron, "White box", Drive Setup 2.1. 2MB file transfer size




JMicron, "White box", Drive Setup 2.1. 8MB file transfer size




Same picture. No change.
I don't remember ever seeing this kind of drop in reads with this drive model, when testing it with 66 and 100 bus G4 portables (Ti and Al), I would have spotted it immediately. See the same drive tested in my post Nr.2.

The only difference between those portables and this Mini is that portables had 10.2.1 ROM, but Mini has 9.6.1. Could that be the culprit? Any ideas?


EDIT. Here are the test results by Quick Bench 2.0. IMHO, the numbers produced by it do not represent the whole picture of what's going on in the region between 512kB and 6MB.






More testing to follow..

[ssp3]

Here's another test that I've conducted, but this time in OSX 10.5.8. I think, it was 12" 1.5Ghz PowerBook G4, but might also have been either 1GHz 12" or 1.33GHz 15" . Same PM851 mSATA drive as in Mini tests above. "Red" Marvell based and "green" JM20330 "White box" adapters.

As far as I can see, there is no performance drop in writes between 1MB and 6MB file transfer sizes as it is the case with my Mac Mini.
(Something strange is going on with the Mini.  It needs to be investigated further. I already spent several hours yesterday troubleshooting it, but can't report any success.)

Marvell







JMicron

[FBz]

So now you’re using QuickBench 4.0?
And not running the Extended Tests / 20-100 MB?
AND opting to display the graphs as Bar instead of Line?
 

[ssp3]

So now you’re using QuickBench 4.0?

a) Not now. These are older tests.
b) There are no better tools available in OSX on PPC. AJA is only so-so.
c) Posted these especially for QuickBench fans. 

And not running the Extended Tests / 20-100 MB?

No. There's nothing meaningful going on past 8MB or so on ATA66/100 bus. Waste of time.

AND opting to display the graphs as Bar instead of Line?
 

So what?

[IIO]

No. There's nothing meaningful going on past 8MB or so on ATA66/100 bus. Waste of time

there are SSDs which get slower after 3 minutes and similar puzzles.

you already mentioned disk size above, then there is the buffer size, the question how many free space is left on the disk, or how many files are present (aka size of catalog)

so many factors... that also explains why each of us has different numbers in his biological memory, because experiences are so different.

[ssp3]

there are SSDs which get slower after 3 minutes and similar puzzles.

The fact that SSD drives get slower (on modern OSes) after their RAM buffer is full is nothing new.

Speaking of puzzles - do you guys realize that many of you are running drives that draw a lot of current on adapters, whose voltage regulators are spec'ed only for 1A @25C ambient and have internal overcurrent protection? When that kicks in, all kind of strange things can happen.

Also, with "white box" adapters the drive sits in confined space and gets constantly "pressure-cooked" by the heat, produced by 3,3V regulator, bridge chip and the drive itself. And there is no way for that heat to escape, because the box is plastic! PCB copper area on the bottom is not sufficient.. (Did they teach physics at school in china?)
Many SSDs have onboard thermal protection. If that kicks in, the results also might be unpredictable.

Ant then there's G4 PB and Mini motherboard capabilities to supply current to HDD or SSD. Did anyone ever noticed that original rotating platter drives on those draw less than 1A?

Puzzles, you say? Sure, sure.. 

[IIO]

some people suggested - or asked - in the past if it could be a good idea to install the "white case" adapters in a macmini without the white case.

similar to how fury suggests to use vertical PCB adapters and then connect a 2,5" SSD without its 2,5" enclosure.

i am not sure if that will help inside a macmini. i am all but an expert in these things, but i do not see that there would be much airflow happening around the HD slot anyway, so why bother about a plastic case?

moving a macmini board (without its case) into a 19" rack is on my to do list. of course i have to learn more about airflow and measuring temperature before i do.

btw. where does the heat come from in a SSD? keep in mind that we are only using 15% of its maximum read and write speed with OS9/IDE100

[ssp3]

The problem is not the SSD, it's the 3,3V regulator on mSATA and M.2 adapter boards that gets hot if one is not careful choosing his/her SSD.
P = U x I
5V - 3,3V = 1,7V drop across the regulator
1,7V x drive's current in A = P in Watts converted into pure heat.

btw. where does the heat come from in a SSD?

As far as I know, major part of it comes from the controller, operating at several hundred MHz, but flash chips also contribute.

https://www.youtube.com/watch?v=hq8DjQWYqOk

keep in mind that we are only using 15% of its maximum read and write speed with OS9/IDE100

It does not matter how fast the transfer speeds are on the G4 bus. The SSD drives operate internally at their own speeds (300-400MHz). If over-current condition occurs even for a couple of nanoseconds, the current limiting by voltage regulator will take place. No need for the SSD drive to be hot.

Re. choosing drives - if someone thinks that choosing drive of smaller size is safe bet, think again. Picture of current production Transcend mSATA 32GB drive is attached. The whole line is like that.
When it comes to everyone's darling "El Cheapo King of the Specs" aka KingSpec, they don't even publish the specs, but this video should give you a clue:

https://www.youtube.com/watch?v=ewZDgkmUCwE

[ssp3]

And here's another horror video (in russian). This time about Transcend M.2 SATA drive. I had someone who helped me with translation.
The guy in the video said he saw the temperature rising to 95 deg Celsius for a brief moment. He returned the drive because of that.

https://www.youtube.com/watch?v=uDwabiqW6AU

[ssp3]

Another anomaly on my Mini's native ATA bus.
This time it is Sandisk U110 16GB mSATA drive. The drive is spec'ed at 50MB/s write speeds, but in this case writes drop down to several hundred Kb! And that is with HDT 4.5.2 driver. When initialized with Drive Setup 2.1, machine freezes.

EDIT. It happens with both, Marvell and JMicron adapters.




The same drive works flawlessly in TiBook 667 (see my 2nd post in this thread).
When put on a mSATA to SATA adapter and into Thunderbolt enclosure, I get these speeds in Yosemite.

[ssp3]

The same Samsung drive moved between 3 adapters - Delock Marvell based, "White box" JMicron 20330 based and small form factor, "naked", also JMicron 20330 based. The latter seems to be identical to "White box", except everything is squeezed closer together. Sold by everyone and his brother in China and by Delock (!) in Europe. Surprise, surprise 

In case you were wondering what is the "naked" adapter, I was talking about - it's this one. I forgot to attach the picture in the first place.
How do you like the symbolic 'China transition' on the sticker? 




The design of it is not ideal, if not flawed. The mSATA drive gets mounted right on top of 3,3 Volt regulator, which gets quite hot. JM20330 gets hot too, but not as hot as regulator. As a result, they both heat up the mSATA drive, and, under circumstances, that might trigger the drive's internal thermal protection or reduce its lifespan.

Still, it can be useful, if you pull out your old fart electronic engineer's "sewing pin and piece of cotton swab" trick 
The latter is to prevent accidental lateral movement of the pin.




Speeds of this combo

[ssp3]

I'm taking back what I said about Mac Mini and write speed drops a few posts back 
It appears that those drops occur only with one specific drive family (mSATA or 2.5" versions) and they occur on other ATA-100 bus G4s too.

Here are the results of the same drive in unsupported PowerBook G4 1.33 GHz.
I had it open while I was fitting Hi-Rez 1440x960 display to it and decided to run a couple of tests.
The adapter used was MCA004 v1.3, the 2.5" drive was removed from its case





After setting the PowerBook G4 1.33 GHz to full speed in OpenFirmware. This gave additional ~2MB/s speed bump.

[ssp3]

Just FYI, there is another JM20330 adapter available from 'over there'.
It is marked as HXSATA-IDE V1.3 11.18.
At a first glance, the implementation seems to be the same as with the other adapters, except that they've added three LEDs, a couple of resistors and one transistor.
It is relatively new and, at the moment, it is also the most expensive of any of JM20330 adapters available - around $8.00/pcs or so.

Speeds are the same as with other "green" adapters, so I won't post any graphs this time.

What makes it interesting and the reason I bought it, is, that, when installed, it has the PCB that's "stepping" downwards, as opposed to upwards as is the case with the popular, cheap v1.3 and v1.5 adapters.

Connector  ===
                       II
                         =========  PCB      <- HXSATA-IDE


                         =========  PCB      <- v1.3 and v1.5
                       II
Connector  ===

The dimensions are also different.

I wanted to try it inside the 12" 1.5 GHz PB G4 in combination with one specific low current, low profile "naked" SSD to try to tackle the "crazy fan" problem, that this model has, when mSATA or M.2 adapters are used.
Tested it briefly in my newly acquired 1.25GHz Mini, but, unfortunately the manufacturer used some sub-standard capacitors in the production and, as a result (of piezoelectric effect), that stupid thing is producing whining noise, that is synchronous to changing graphics on the display.
It drives me nuts 
Unless the manufacturer fix this problem, I would not recommend to buy it.

[DieHard]

I guess I got lucky the original msata I purchsed 4 years back was...

ASENNO mSATA Mini PCIe 240GB 256GB Series SSD Solid State Drive (30 50 mm) for Pos Machine and Game Machine

It was real cheap at the time on Amazon and a real "no-name", and seems to hold up well (knock on wood) with read/write speeds Between 86 & 89 MB/sec on all transfer sizes, no heat issues.  These tests were using the Ableconn

[ssp3]

Just by incident I discovered another test utility that was there all the time, right in front of my nose.

It is called TimeDrive. Last version is 3.0.3. Available on Sonnet CDs and on garden too.

Now, this is a monster tool!
If you set the step size small enough, the results will drive any noob, who doesn't know how to interpret them, absolutely crazy. 
The results can be saved as text files, but those will drive II0 crazy too. 

Take a look.