Slow SSD read performance in CrystalDiskMark

[IctusBrucks]

Hey guys,

I have been benchmarking my new Corsair Neutron GTX 240gb SSD in a few different configurations, mostly in external enclosures since I am trying to get a system for fast transfers. I seem to be getting really strangely slow read numbers for USB3, while write numbers are great.

I connected the drive to the mobo's Sata3 port as well to get a control point.

System info:
Intel i7 3930k @ 4.5ghz
Asus p9x79 deluxe mobo
GTX 680 video card (Asus OC model)
Samsung 830 256 SSD (Operating System drive)
32gb Gskill 2400 memory running X.M.P @2400


Tested here:
Corsair Neutron GTX 240gb SSD and 3 enclosure configs + internal


First up I tested the Neutron GTX using the Oyen Digital Minipro USB 3.0 Enclosure.

As mentioned, read speeds are strangely low... write speeds about what I expected.

Next I tried the same drive in the Silverstone RVS02 USB3 enclosure:

I also have an older BlacX eSata/USB2.0 dock I tested in eSATA mode as well:

Interesting that for all 3 cases, read is lower than write (though a bit less for eSata which is slower in general probably because of older chipset in the blacX usb2/esata model).

As a sanity check, I put the Neutron GTX into the Motherboard's Sata3 port and ran the same test again:

So the drive itself is definitely performing great, about what I expect. Well its not quite as fast as my 830 for reads but beats in a few other categories (I forgot to take a screenshot of the 830 benchmark and I am trying to not over-benchmark that one as its my OS drive).

So I also ran the SiSoftware Physical Disk benchmark. I ran it only using the GTX and Silverstone RVS02 enclosure.

Interestingly, that test gave me a read result of 280 MB/s. Around double the number CrystalDisk is giving me, and much more in line with what I expected out of USB3 and an SSD.


So is there some factor that explains why crystal disk is performing so low in these USB3 benches, while SiSoft is showing good numbers? Could it be some artifact with the USB3 on the x79 chipset? The only thing I have found which supports this theory is this x79 mobo review where the USB3 transfer speeds were really low, about what mine are:
http://www.legitreviews.com/article/1841/12/


I actually haven't seen alot of x79 motherboards with USB3 speed tested. For some reason those benchmarks seem much more common on Macintosh forums lately.

Thanks to anybody who has any ideas... even if its "ignore the benches"

 

[blastingcap]

I guess I'm going to be in the "ignore the benches" crowd for the most part.

SATA to USB conversions (and vice versa) presumably involves overhead and thus would reduce performance compared to native SATA connection.

The BlacX SATA to USB converter may be a SATA-II internally. That's what I would expect given those numbers, anyway.

Now, a word about speed ratings:

Are you sloshing tons of data around all the time from one SSD to another? Fine, then sequential matters, I guess. But for most people Random 4K is what's important, because the difference between 300 MB/s sequential transfer and 400 MB/s or even 500 MB/s isn't that in real-life use. Your 4k randoms are very good (30 MB/s via mobo SATA), so I wouldn't worry about it. FWIW, I get only ~25 MB/s random 4K reads on my random 4Ks on my not-so-new Crucial M4 256GB in CrystalDM, same ~25 MB/s with my Samsung 830 256GB, and the M4 and 830 are both a couple of orders of magnitude faster than a HDD at random 4Ks. That's what matters.

And if you're sloshing data between a SSD and HDD you are limited by the read/write speed of the HDD anyway.

Similarly, you are limited by gigabit ethernet if you are trying to transfer data from one SSD to another.

 

[IctusBrucks]

Yes, I do have to copy over 100gb folders pretty often. Half the time I am just updating a portion of the files(then the random really matters like you say), but other times I have to copy the whole folder over because it's easier than pinpointing a specific error... Or because a new branch has been created or a few other possible reasons... It it comes up often enough to cause a big wait using usb2.0 speeds.

I am trying to establish a baseline before trying some software encryption options to see how much they affect performance as well, until I can get my hands on an 840 pro which supports HW encryption, supposedly with no perf cost.

Edit: I am looking to have a dedicated SSD on each PC that would be copied to/from as well as a portable/secure SSD option so hoping that Usb3 will be the bottleneck.

 

[blastingcap]

100GB folder @ 154 MB/s = 11 minutes
100GB folder @ 459 MB/s = 3.72 minutes

To put things into perspective.

Now that I look at the numbers again, I wonder if the USB3 Silverstone and the other one are also at SATA-II internally. Perhaps that's the issue.

Is rsync an option?

 

[IctusBrucks]

You are right, both those numbers are great compared to usb2 which is what I've had for the last few years... Benched at around 30 MB/s taking around an hour to copy... I just want to make sure my USB drivers are installed correctly etc and I am not hitting some unexpected wall.

And yea, we do use an internal sync tool that only copies what is needed.. But I am kind of thinking worst case scenario here.... sometimes almost every file really did change, say because a code change forces all content to be reserved which basically kills any merging advantage. I work from home while still have to deal with source control, and sometimes I need to leave the office while keeping the in-use drive connected and take a copy on the spot... Weird cases hard to explain.... In spite of the best planning I often have to start copying files late at night and still get things done that night so every minute counts.


The minipro is the only one I know for sure is sata3 as media 1051e, since I you can read the chip when installing the SSD... The silverstone says sata3 but no way to tell...

Thanks for the input.

 

[IctusBrucks]

Well I forgot about the Asus USB3 Turbo feature. It's enabled via the AI Suite software.

For comparison, here is with Turbo mode "off" (in AI suite you just click "normal" to disable):

After enabling, my Sequential writes went way up to what I was expecting from USB3.

I also ran the tests in AS SSD for some comparisons.

First, with Turbo off:

And with Turbo enabled:

I was hoping that the 4k-64thrd results would jump up in Turbo, but I am at least glad to see the sequential numbers getting there.

And just for comparison sake, the same Neutron GTX 240gb drive installed on the mobo Sata3 port (only took Crystal marks unfortunately)

 

[IctusBrucks]

I'm kinda curious to get my hands on a new BlacX 5g USB3 since they apparently support UASP. Asus AI suite has another turbo mode for USB3 UASP that some have found to be even faster than the Turbo mode I tested.

But the new BlacX is a bit steep for an enclosure. Wonder if any cheaper ones have UASP?


Also minor weird note: SiSoft Sandra does not detect the Asus USB Turbo driver as an actual physical drive, so I was unable to run the test to compare to the 280mb/s reading I got before.

Part of me thinks that rather than Turbo mode being really fast, something must be slightly broken or unoptimized with the standard USB3 drivers at least as they are installed on my PC.