Minor brand SSDs

[mackerel]

This kicked off on another forum, and I've decided to buy a bunch of them to test for myself. Note my uses are niche, so the value proposition wont apply to everyone. Basically I have lots of crunchers which have their own disk for OS. Previously I just bought whatever was cheapest at the time from a known brand, usually Sandisk or Kingston at the lower price end. They don't need much more space than that needed for Windows and a bit of breathing space, so 60GB+ is decent and plentiful, at a comparable price-per-GB compared to the bigger size branded drives, but the branded ones go up a lot for smaller capacities.

After much hunting on ebay I've decided to order 4 low cost 60-64GB drives to try out.

Hectron X1 60GB, claimed up to 500MB/s read, unknown write
Kingfast K6 64GB, claimed 400 read 140 write
Fastdisk 60GB, claimed 500 read 98 write
Hypertec Firestorm 60GB, claimed 285 read 275 write

There were some others with lower stated write speeds so I decided against them. Recognising these are claimed sustained rates, and iops might be very different. As said, I have a bunch of recent Kingston and Sandisk SSDs I could compare against, as well as older OCZ and Intel models. At least 3 of these are expected to arrive next week.

I'm gathering software to run on them now, and have assembled Atto, AS SSD, CrystalDiskMark. Any others I should look at?

 

[mackerel]

I downloaded Atto, AS SSD, CrystalDiskMark, iometer... and done a practice run on my laptop SSD. Now I'm left debating if I really need all those? To test a SSD, who tests the test?

Let's start with CrystalDiskMark as a baseline. If I understand correctly it allows you to use random data or zeros, the latter might give inflated results for SSDs using compression. If memory serves correctly, sandforce controllers used to do that. Are they still around? Does anyone else do that too? Still random data shouldn't be very compressible.

AS SSD states to use uncompressible data. The results for sequential are very close to CDM, as is the 4k result. Then there is the other 4k test, CDM has Q32T1 and AS SSD has 64Thrd. Whatever they're doing, even if not the same thing, the results between those are within about 10% of each other. I assume from the names, CDM uses a queue depth of 32, and AS SSD uses 64... threads?

The remaining data point varies between the two, with CDM doing a Seq Q32T1 reaching even higher transfer rates than ordinary sequential, and AS SSD returning an access time. I guess I would need to run both for coverage even with overlap.

Moving onto Atto, if I'm reading it right, by default it just tests transfer rates at various transfer sizes. I'm not sure if I can extract value from this over the other software so I'm not sure I'll bother using it.

iometer... I gave up on loading it. I could RTFM and work out what to do with it if I was motivated enough.

Between CDM and AS SSD is that enough of an indicator of disk performance?

 

[EarthDog]

Look at SSD reviews around the web and see. Some are higher level (like ours, SSD review) and others go a bit more in depth (Anand).

 

[mackerel]

I had looked at the latest SSD review on this site, and was happy to see the same old software pop up. But having the data is one thing, making sense of it is another...

I'm going through my existing systems now, and benching every SSD I have in CDM and AS SSD only. One thing for AS SSD's - it is faster to run! But I still need CDM for max transfer rates. I also note Aida64 extract some interesting info on some drives. I don't know if they have to manually look it up, or if the drive self reports it in some way. Things like controller chip, NAND type used, and even some performance figures (from where? Not measured.)

 

[EarthDog]

All this testing..are you thinking the ssd is really holding things back?

 

[mackerel]

The question I'm trying to answer is how do the unknown brand SSDs compare to the more well known ones. It isn't exactly fair as all but one of my SSDs are 120GB or greater, and the ones I just ordered are 60GB class. There is usually some benefit to bigger capacity drives.

They'll have a job beating a 40GB Intel X25-V I'm testing right now, with it not appearing to do much with CDM 4k reads. Aida64 reports it has 34nm MLC, how old is it?... Intel says 2010, a year before Sandy Bridge. It was my 1st ever SSD.

 

[EarthDog]

Yeah, you can't compare them unless they are the same size. They spec differently.

Off brands generally perform pretty similar to higher end ones. Where you see a difference is generally in IOPS and how quickly 4k reads and writes are. That along with warranty and support are mainly the other differences.

 

[mackerel]

With this testing I found a "problem" I didn't know I had on one system... it has a Samsung PM951 256GB NVMe drive. I picked it as a balance between cost, capacity and giving a bit more performance than SATA drives. It runs my VR system and I've not had any problems with it, until I ran AS SSD. All was fine apart from the 4k random write. It was stupidly slow estimating 1 MB/s and around 20 minutes to run. I aborted that, finished the other tests (write latency was 4ms!!!) and hit Google to try and work out what was up. Found a comment saying that disabling the Windows write cache flush resolves it, and sure enough, doing that made write speeds go up a lot, and latency right down. The only caution here is Windows says that there is a risk of data loss in the event of power loss if I use that setting. What was reporting 1 MB/s in AS SSD is now 150MB/s. I didn't do a CDM before the setting change, but writes there are in 300MB/s ball park apart from 4k which is 200MB/s.

Onto more testing!

- - - Updated - - -

Yeah, you can't compare them unless they are the same size. They spec differently.

Off brands generally perform pretty similar to higher end ones. Where you see a difference is generally in IOPS and how quickly 4k reads and writes are. That along with warranty and support are mainly the other differences.

I'm not doing a like for like test, as I can't. So I guess what I'm looking at is, is it good enough? I'm not sure what level that is exactly which I'm trying to work out by looking at what I do have. One of the ones I've ordered has low claimed sequential reads, but relatively high writes, so I'm wondering if they might have tuned it differently for example. Some of them spec very low on sequential writes so I'm skipping those totally.

 

[mackerel]

Well, I got my baseline data. Following drives tested:

Samsung SM951 512MB AHCI
Samsung PM951 256GB NVMe
Sandisk Ultra II 960GB
Sandisk Plus 480GB
Sandisk Plus 240GB
Sandisk Z400s 128GB
Crucial MX200 1TB
Toshiba HG6 128GB
Kingston V300 240GB
OCZ Vertex 3 120GB
Intel X25-V 40GB

The two Samsungs are generally faster, being the only M.2 drives breaking the SATA barrier. Things get complicated in the middle with differing strengths and weaknesses. The ancient Intel takes up last position.

It was curious, all the Sandisk SSDs resulted in an error when the AS SSD read latency test tried to run. I was also amused to see the Sandisk SSD software recommend turning off write cache flushing for performance testing, as I had to do to get the Samsung PM951 working well on random writes.

Got dispatch note on 3 of the no name SSDs now, so should be quick to add them to the results and do something with it.

 

[mackerel]

Two of the drives arrived today, the Hectron and KingFast. The Hectron is nicely packed, with an outer sleeve and inner foam holder and drive in anti-static bag. Haven't taken the drive out for a look yet. The KingFast was much more basic, a simple cardboard box and bare drive inside. The drive also looks cheaply made, looks like an insert slid into an aluminium frame. I think if I remove the one screw I can see I could take it out... will save that for after I've tested it. Will get some preliminary results tonight.

Elsewhere: I've got a bid on a SAS controller. I salvage two SAS drives from an old server, mistakenly thinking it was SATA. Been wondering what their performance is like... but I obviously don't want to spend big money on the controller. To cut the story short, I had to find a low price, in the right interface, with connecting cable, and drivers for a new enough OS for me to use it. In the end I chose a HP part since that includes drivers to 2008R2 (Win7), whereas the cheap Dells ran out of steam around 2003. Just hope I "win" it now... my suspicion is, even with their high rpms they'll still fall to SSDs, so I don't want to pay more for the controller than it is for one of the SSDs I'm testing here.

 

[EarthDog]

HDDs (not in a RAID stripe) will ALWAYS be slower than modern SSDs... even 15K RPM models (which, SAS tells us nothing about the speed, just the connection). What 'SAS' drives do you have? I am certain there is already published data on their speeds if you look.

 

[mackerel]

I'll have to look up the model when I get home, but reading about performance and experiencing it are quite different. I do know one was 10k the other 15k. Both I think were <100GB capacity.

 

[Culbrelai]

So doesn't every SSD basically OEM their nand from like 3 different people? Micron, Toshiba, and Samsung or something like that? So the difference between brands will be minimal?

Also, other brands you should test

Silicon Power
KingSpec
Patriot
Mushkin (Not really a "minor" brand but you certainly don't see much of them)

 

[EarthDog]

The NAND is one thing, the controller is a completely different story.

 

[mackerel]

Ok, some testing done and I've also opened them up for a look.

The KingFast K6 64GB first... sequential reads in the >300MB/s ball park, writes around 85MB/s. Randoms... gonna take a lot more analysis to put in perspective. Opening it up it uses Silicon Motion SM2246XT controller, and there's a single NAND chip taking one out of 4 possible positions, suggesting it could go to 128 and 256GB capacities also. This is on a smaller board on a 2.5" bracket, which might explain why it reports to AIDA64 as a 1.8" drive. The NAND device appeared to have a covered Micron logo on it, and the part numbers didn't turn up in a quick search.

Onto the Hectron X1 60GB. This was able to reach sequential reads of 500MB/s with writes floating around the 70MB/s ball park. Inside I found the SM2246XT once again, but the board had 8 NAND positions available on each side, with 4 devices fitted on the same side as the controller. Again this suggests it could go up to 120 and 240GB capacities, assuming the same NAND was used. NAND had GLOWY on top and h27qdg8d2b8r which comes up as a Hynix part. Indirect info suggests it is 16nm MLC.

The SM2246XT is also reported by AIDA64 to be used in the Sandisk Z440s and PLUS drives I have.

Culbrela, it is not my intention to try everything. Only to try some of the ones I can buy easily, and see if there is much between them.

Elsewhere: I won the SAS controller... better pay for it now.

 

[mackerel]

The other two SSDs arrived today. The Fastdisk is in a low budget plastic retail pack. Hypertec came bare in a padded envelope. I believe it is an OEM part so they wont necessarily have individual packing for them.

The SAS controller also arrived. Sender obviously didn't have an anti-static bag as it was put inside two ordinary plastic bags so I hope it is still ok. I could also tell they're a smoker and will air out the cables before use.

So, lots more benching to do this weekend then.

 

[PanteraGSTK]

Good info so far. I've stuck with the brands I've used and respect, but there are still REALLY cheap models to be had by most of the big names too. OCZ makes some really cheap drives. Patriot as well. I've never had any issues with any.

One thing I'd like is a larger SSD for my gaming PC. My 7200RPM 2TB is quick, but it could be better. Those are still pretty expensive SSDs though.

 

[mackerel]

Just done benchmarking of the 2 new drives. The Hypertec was disappointing in its writes, being in the same class as the others whereas the sales listing claimed it was a few times faster. The only reason I picked that one was for the writes, as the reads were below average and they did come out as such. The Fastdisk was also unremarkable. Overall impression for the budget set is that they're slower than the (higher capacity) branded SSDs I have, but the numbers are plenty enough for general use, and still superior to rotating HDs.

I haven't opened either of these up yet. I'll do that tomorrow and will take photos along the way with the first two also. I'll see if I can get the results into some presentable format later.


On the SAS controller (HP P400), I made another mistake there. It did come with cables, but I didn't look at them carefully enough and they're not the right ones to connect to the drives. So I still need a cable if I want to use it. Also, the first Win7 system I plugged it in didn't recognise it at all. Not even an unknown device in Device Manager, and drivers refused to install without the device. I tried two slots to be sure. Firing it up in a Win10 system, it got detected and drivers installed right away. Can't do anything more at this point without a cable... a quick look on ebay suggests it'll cost as much as this card with wrong cables did... which combined is more than one of the budget SSDs...

 

[mackerel]

And the results are in! I've colour coded a bit to make it easier to see how classes of product compare.

In yellow, we have PCIe connected SSDs. These are the only ones capable of breaking the SATA interface barrier. In blue are assorted SATA branded SSDs. In green are the small capacity lesser brand SSDs. And in red are the spinning hard disks. On that note, the Toshiba 3TB is a 3.5" 7200 rpm drive. The Seagate Momentus XT is a SSHD with a flash cache of some GB but otherwise is just a hard disk. The two Hitachi hard disks are 2.5" laptop drives. You can guess which is 5400 or 7200 rpm from the model name.

I'm only presenting 3 results from CrystalDiskMark, as the AS SSD results were not that different but I didn't run that on the HDs.

I would caution I only ran the benchmark once for each unless there was something obviously wrong with the results. Run to run variation may move the positions around a bit but not in a significant way.



This is the best case read speed, and no surprise the PCIe drives blow the others away. The lesser brand drives are at the lower end of the SATA SSDs, but this may be more due to the small capacity. The hard disks take up the rear.




For sequential writes, things shake up a bit. The Samsung PM951 drops down the ranks, possibly a combination of it using TLC and relatively small capacity. The Toshiba 3TB HD holds strong, and the lesser brand SSDs even drop below the hard disks. So if you have a lot of sequential writes, hard disks can still be competitive especially if you need the capacity also.




The queued random reads separates the SSDs from the HDs quite clearly. While we have a big spread amongst the SSD models here, even the slowest one is significantly faster than any of the hard disks.




The situation is similar for queued random writes.




Random reads again perform similarly to before, just slower overall.




And similar again for random writes, with the interesting observation the lesser brand small SSDs manage to outpace bigger Samsung PM871 and Sandisk Plus SSDs here.


Overall, I never expected these to be miracle performers, and they are not. What they are, is good enough performance for general tasks, and still far faster than hard disks in most situations. Due to their small capacity, if you did have a large quantity of sequential operations a hard disk would be a more interesting option anyway.

 

[Vishera]

Anybody else think this could have been an article? Nice bit of info for the general population to know that cheaper doesn't always mean it'll blow up in the first 2 hours of use.