Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

samsung evo 850 ssd uasp USB 3 speed question

2

2bcool2

macrumors regular
Original poster
Jan 31, 2008
186
7
hi
i have replaced my normal usb3 enclosure with an inatek hasp one.

my non uasp enclosure gave me blackmagic speed test results of write 250 and read 266.

the new uasp enclosure tested gave me write 423 and read 432 which is awesome, but the blackmagic speed test app only gives me 432 write speed on its first pass, then following passes give me 289 ...

can anyone tell me why that is ? is it a buffer? is the first or the second reading the true speed ?

is it anything to do with trim? i use with iMac 2012, would i get faster write speed by using trim and if so , how do i do that ??

many thanks in advance, hope someone can put me straight :)
 
JTToft said:
The speeds vary throughout such a test. It's normal.
Let it run for a few minutes to get a feel for the average.
Click to expand...

yh , i have let it run a few cycles, first pass is always in 400's write, then after always in 280's

was hoping to hear how to get the 400's i get on first pass lol

a non uasp enclosure gives me 250, a uasp one gives me 289 ,
thought there would be more difference from non uasp to uasp

any ideas ?
 
I think thats about what to expect for a good SSD with USB3 and black magic, seems normal to me. I average maybe 250MBps, peaks at ~300. Same disk in a TB enclosure averages about 375, peaks at ~450.
 
JTToft said:
- It isn't. USB 3 with UASP support should give around 400 MB/s. Non-UASP would be around what you're citing for yours.

My money is on a faulty enclosure.
Click to expand...

hopefully its not a faulty enclosure, i already have to send the non uasp one back lol.

i tested it on my 2012 macbook air also and get same results as my 2012 iMac.

is its anything to do with trim ? I'm gonna use it for video editing in premiere, so a fast read speed is probably the most important right ? and I'm getting 433 read. the fact i get 289 write speed shouldn't have much impact on video editing right ?
cheers
 
2bcool2 said:
hopefully its not a faulty enclosure, i already have to send the non uasp one back lol.

i tested it on my 2012 macbook air also and get same results as my 2012 iMac.

is its anything to do with trim ? I'm gonna use it for video editing in premiere, so a fast read speed is probably the most important right ? and I'm getting 433 read. the fact i get 289 write speed shouldn't have much impact on video editing right ?
cheers
Click to expand...

- Just to confirm that it is actually utilizing UASP on your iMac, follow these steps.

Perhaps also take a look at this thread.

As for TRIM, I'm not sure that's even possible over a USB interface.
 
dyt1983 said:
Nope.

Samsung state (the fine print, so to speak):

"Sequential Write performance measurements based on TurboWrite technology, The sequential write performances after TurboWrite region are 150MB/s(120GB), 300MB/s(250GB), 500MB/s(500GB) and 520MB/s(1TB)."

So it's buffered, and after the buffer is exceeded, one can expect 300MB/s (which I think is optimistic). As with many SSDs, smaller capacity is slower than larger capacity.
Click to expand...

Hmm... Reference? Anandtech has the 850 EVO at consistent speeds across all capacities for incompressible sequential read and write (which is what BlackMagic Speed Test does).

[Apologies about the double post. Didn't realise until after clicking submit.]
 
Last edited:
Read this the other day - makes sense really :cool:


Don’t run benchmarks on your new SSD
Some people, right after they buy a new SSD, want to enjoy the speed and are eager to find out how much faster exactly their new SSD is than the old HDD. So they get this cool idea of running some extensive benchmarks to see the amazing performance numbers. Benchmarks usually write a lot of data to the disk (to test the write speed), wearing it out. So it is the best way how you can ruin your SSD even before you start using it. Don’t do it.
 
dyt1983 said:
Samsung web site, 850 Evo SSD data sheet PDF. (I mentioned that Samsung stated it.)

Anandtech's specs are a repeat of Samsung's, without the footnote. If you look at page 2 of the Anandtech review you quote, it shows the difference.
Click to expand...

- Yes, but it's a bit of a chore for me to hunt around Samsung's website to locate your quote. ;)

- I wasn't referring to the listed specs, but to the actual incompressible sequential read and write (which is what the BM test does) test they performed (on page 8), which has the drive at very similar speeds across all capacities.
I added this clarification to my post above a few minutes after you quoted it.

ricede said:
Benchmarks usually write a lot of data to the disk (to test the write speed), wearing it out. So it is the best way how you can ruin your SSD even before you start using it. Don’t do it.
Click to expand...
- The write tolerance of moderns SSDs is in the petabyte range. The amount of writing performed by a few benchmarks will never make a difference.
 
dyt1983 said:
This isn't meant to be abrasive, so please don't take offense: I don't treat posting here like editing Wikipedia, so I don't footnote every statement I make (nor to 98% of the people who post here... sorry, no reference for that number!). I think it is reasonable for someone who questions something to research it if the results are easily found. To make sure, I just did a search "samsung 850 evo ssd", thanks to Samsung's enormous advertising budget it is the first page that comes up on Bing/Yahoo!. The spec sheet is in a collection of links on that page.
Click to expand...

- I actually quite agree with you on that. And I did look up Samsung's specifications, but I couldn't find your quote on the page, so I decided to just ask you for the location. Admittedly, though, I didn't look at the spec sheet.

dyt1983 said:
Thanks for the clarification. I noted it but didn't change my response as it would be the same:

The specs are on page 1. Test results are on page 2. If you look at page 2 of the Anandtech review quoted, it even has a table showing the difference, and there's a large graph with performance figures that don't correspond to your claim that they are "consistent". No, seriously. Click the link. Go to page 2 to see the table. And I will link the graph below.
Click to expand...
- I did see page 2, and I do see the difference there. But am I completely wrong to think that the more relevant data to our current discussion is the incompressible test on page 8, which indeed does show consistent speeds across all capacities?

70039.png


70040.png
 
Last edited:
dyt1983 said:
You don't know exactly what disk he has! Well, maybe you do. It's a 250GB.



Nope.

Samsung state (the fine print, so to speak):

"Sequential Write performance measurements based on TurboWrite technology, The sequential write performances after TurboWrite region are 150MB/s(120GB), 300MB/s(250GB), 500MB/s(500GB) and 520MB/s(1TB)."

So it's buffered, and after the buffer is exceeded, one can expect 300MB/s (which I think is optimistic). As with many SSDs, smaller capacity is slower than larger capacity.



OP was correct about this (but not about UASP previously).
Click to expand...

thanks everyone for your info, so now i see all that you guys have posted getting 10 under the 300MBs as stated in one of the posts for s 250GB SSD, sounds fine ...

i just wondered why it was reading in 400's for the first pass always in speed test. i guess it is the buffer. and trim isn't for usb3 i understand now, so thats good.

i think the uasp makes a difference, if only slightly for the write side of it.

cheers, for all the input. i bout the 850 as an impulse buy as i heard it was on sale an hour later on french amazon lol, seemed to be the make everyone likes, so i jumped on it.

i had never heard of uasp before buying this or trim haha.

i am waiting for an updated mac pro to buy now, i really hope its announced june! fingers crossed.
i bet they put the price up here in uk :-(
 
dyt1983 said:
I understand what you're saying, but would you agree most think that the IOPs are more pertinent to real world performance than just streaming out a bunch of data?
Click to expand...

- Yes, most definitely. But what we're discussing here isn't the way in which SSDs primarily achieve their performance improvement over conventional hard drives. What we're discussing is why the OP is seeing varying - and lower than expected (whether it's a justified expectation or not) - speeds from his 850 EVO in a UASP-enabled USB 3 enclosure.

dyt1983 said:
But back to that... the 128K incompressible data test doesn't indicate the parameters used for the test, so it is reasonable to assume based on the results that the writes are just staying in the buffer, and that the BM test is probably exceeding the 3GB buffer. The regular 128K sequential test is listed as being 1 minute, which should easily exceed the buffer and the results show the lower capacities as being lower performance.

Just speculation.
Click to expand...

- That could indeed be the case. But I would like to think that Anandtech's testing methodology is a bit more rigorous than that.

I don't believe I've seen anyone report this low performance using BM on their 850 EVO installed internally - which they should be getting if the low performance is because of the 'issue' you mention and not because of the enclosure.
But I must admit that I can't actually remember any forum posts with speed tests of a 250GB 850 EVO... Although one guy did mention averaging 520 MB/s read/write on his 256GB 850 Pro, which is specced very similarly to the EVO for read/write performance.

[EDIT: Reading a bit more about the TurboWrite implementation from different sources, I think it's likely that you are right about the 3 GB cache being saturated at high speeds, after which the speeds drop for the 120GB and 250GB versions. I agree that seems the likely reason for the OP's experienced performance drop. I am still a little puzzled by the Anandtech incompressible test, though, but it must be due to it not exhausting the TurboWrite cache, like you said. Perhaps they should have used a different methodology to achieve more stable write numbers...
Thanks for correcting and educating me on the matter.]
 
Last edited:
JTToft said:
- It isn't. USB 3 with UASP support should give around 400 MB/s. Non-UASP would be around what you're citing for yours.

My money is on a faulty enclosure.
Click to expand...

He's using the BlackMagic benchmark with defaults, and in the real world one gets much less than 400 MB/s with well working USAP enclosures and the better SSDs. You can perhaps hack to get better, but USB is just not that good running that benchmark (video POV). One can certainly find a benchmark that will demonstrate 400 MB/s. There are certainly a lot of things that can go wrong with USB, cables, enclosures, drives, firmware quirks being the most troubling....
 
Thanks for correcting and educating me on the matter.]
Click to expand...

You're welcome, and I too appreciate your input and informative posts.

edit: to remove personally identifying information not relevant to thread.
 
Last edited:
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back