Harddisk speed test

Posted on 19 June 2013


As manufacturers always claim very positive numbers it’s best to test the performance yourself.

To test a harddrive for speed there are a few options. The first thing to decide is what kind of performance you want. Do you want fast copying of big video files or do you need fast OS performance? All depends on queue depth, packet size and in case of HDD: physical position on the drive.

A list of programs to test it:

  • Windows Device Manager
  • Roadkil´s Disk Speed (doesn’t recognise any disks)
  • ParkDale
  • PCmark 7 (not free). Their total storage benchmark is the best for your OS disk decisions.
  • IOmeter
  • HD tach / HD tune. The blue line on the graph shows the sequential read performance across the drive (data on the outside of the drive is much faster than data recorded on the inside, because modern hard drives use a zone bit recording technique).
  • USB flashspeed.com
  • AJA system test (mac and windows)
  • Black Magic test (mac only). For video files,  Blackmagic Disk Speed Test uses long sustained block transfers just like capturing and playing video.
  • ATTO disk benchmark: graph for different block sizes.
  • Crystal Disk mark. This blog explaines reading of crystal Disk mark results.
  • AS SSD (same as crystal, but with uncompressable data)

A great source of benchmarks is hardware.info/reviews.

I found that the same test can give different results. I guess that’s why you can select 5x or 9x with Crystal Disk Mark.

IOPS * block size = Random speed.

Toshiba 7k320

This disk is 99% full at time of testing. Internal SATA. Win7/64bit. Toshiba Harddisk Security enabled.

  • AS SSD: seq 62/58. 4K 0.5 MB/s, after that it got really slow.
  • Parkdale: seq 12/15. 4K QD32 1/0.3 random  (write = 260 IOPS, 4 ms)
  • Crystal: 3.0.2: seq 77/77.  4K 0.8/0.3 (both QD1 and QD32)
  • HDtune: average seq: 72 MB/s (100 MB/s at beginning, 50 MB/s from 80%. , burst rate 86 MB/s. Access time: 17.8 ms.

Same test without Toshiba HDD security and Sophos on-access scanning.

  • AS SSD: seq 64/58. 4K test is extremely slow!
  • Parkdale:  seq 84/79.
  • Crystal: seq 82/82. 4K QD32 0.8/0.3.
  • HDtune: 71 (100>50), 18.6 ms
  • AJA: 82/79 up to 94/83 (not very consistent).

Without the scanning and/or gravity sensor it might be 5 MB/s faster (Crystal) but it might also not be (AS SSD). There should be several rounds of testing efore this is true.

As you can only run one program at the time, a good test round takes quite a bit of time.

WD my Passport 1TB

Empty new disk, NTFS. Sophos off:

  • Parkdale: 107/114, 1.6/0.5 (2.62/8.10 ms)
  • Crystal: 120/120, 1.4/0.6
  • HDtune: 92 (120-56), 16.4 ms
  • AJA: 115/115 (both at 1 and 16 GB. at smaller sizes (128 MB) it is faster).

Then with the password protection (smartware enabled?) on it’s about the same:

  • Parkdale: 111/115, 1.6/0.5 (2.50/7,97 ms)
  • Crystal: 120/120, 1.5/0.6
  • HDtune:  90 (115-56), 16.5 ms
  • AJA: 115/116
  • ATTO (set to 256 MB, direct I/O met QD=4):


After exFAT (256K cluster size) reformat it’s a bit faster for seq but might be slower for 4K random:

  • Parkdale:  117/117, 1.5/0.5 (2.75/8.19 ms)
  • Crystal: 122/123, 1.4/0.6
  • HDtune: 91 (120-56), 16.5 ms
  • AJA: 117/117
  • Atto: 6083-123702. So with it is a bit faster for all cluster sizes, including 4K.

Filling the disk with 8.59GB nex5N video files from anold disk: 83 MB/s. With 274 GB it’s 82 MB/s. This is the speed limit of the Seagate FreePlay 1500GB over USB3 (crystal measures 80 MB/s). When the disk is 78% full it’s much slower for seq but (i’m surprised:) faster for random 4K:

  • Parkdale: 75/75, 1.7/0.7 (2.30/6.00 ms)
  • Crystal:  78/78, 1.7/0.6
  • HDtune: 92 (120-56), 16.5 ms
  • AJA: 74/75
  • Atto: 5941-78033 (16k). From 16K it’s full speed.

I have to say that HDtune is the only one with consistent results. Now with 20% free space, AJA (at 1 GB 1920×1080 8bit) is much more reliable then at 99% used space: it’s generally a nice avarage between Parkdale and Crystal. I think it will be HDtune + Parkdale/Crystal/AJA.

Seagate 1TB Backup-Plus

New with 4K NTFS (when I select format I can choose from 512 byte up to 64 KB):

  • Crystal: 121/122 | 122/122 | 122/122 (9x1GB) | 122/124 (9x100MB), 0.8/0.5
  • AJA:  106/117 | 118/117 | 115/114 | 115/115 (4 GB) | 114/115 (16 GB) | 129/119 (128 MB)
  • HDtune: 84 (115-47), 17.2 ms

I think Crystal has much more constant results as it uses multiple runs. When I set it to 1 run it gets less consistent too. But it always measures about 5 MB/s higher then AJA does.


Then after a exFAT (256k) format Crystal results are: 124/124, 0.85/0.56. Atto: 9424-126026 (32k)
After reformatting it to 64k (exFat) the results are:  124/125, 0.86/0.55. Atto: 9320-126026 (64k)
conclusion: read is a little bit faster with 64k cluster size, while write is a bit faster with 256k. But it doesn’t really seem to matter. As the above exFAT takes 18/700= 2.6% more disk space then 4k NTFS, I’d like to see some improvement, so I chose 64k for this one. Now it only takes 4/700=0.5% extra space.

Then I deleted the partition and set truecrypt to encrypt the full drive. I got a warning about windows wanting to initiate it everytime I’ll plug it in, so we’ll see. Then I started the encryption benchmark which shows that parallelization and HA AES are supported, but takes 4×100% processor to encrypt a 1Gb benchmark file. AES wins at 1.5 GB/s while the second best are twofish (247) and AES-twofish (212 MB/s). Next is to choose the hash algorithm, I chose SHA 512. The next step is to format as either NTFS or FAT. There is no exFAT option here, so I selected ‘None’ and then clicked ok. After the partion was created I could mount it and format it as exFAT (64k).

  • Atto: 7552-125730 (32k). So only for files up to 16k it’s faster unencrypted.
  • Crystal: 125/124 (cpu at 24%), 1.0/0.46. So random Q32 write got a bit faster, read got slower.
  • AJA: can’t see the mounted truecrypt drive. So I tried Parkdale: 84/118, 1.22/0.54
  • HDtune: 84 (115-50), 17.2 ms (I doubt if it uses truecrypt, no processor usage)

Conclusion: truecrypt with AES HA doesn’t affect performance much, compared to the same USB3 drive without encryption, but it needs about 20% of CPU power.

Back to the comparison with the SED WD:

  • Parkdale:  seq write with truecrypt is a lot slower at 84 vs 117. The rest is the same.
  • Crystal: seq is a bit better with truecript, but random is about 25% slower.
  • HDtune: the SG disk is a bit slower then the WD (84 v.s. 91)
  • Atto: Faster on the truecrypt drive.

Conclusion: the SG is a bit slower then WD, but in total their performance will be about the same.

Now, when copying data from the WD to the SG drive (both connected by USB3) speed is limited to about 60 MB/s. I also got an error message about the source path being too long. It’s this folder:

F:\Dropbox\NaarLibrary\Apparatuur\SoftwareMacOS\FCS3\dvd\Installer\FinalCutStudio.mpkg\Contents\Resources\Requirements Checker.app\Contents\Resources\AELicensingAquaPro.bundle\Contents\Resources\Japanese.lproj\AELicensingMultiPrevSNAquaPro.nib\keyedobjects.nib

Total of characters is 260. Another file in the same folder has a path length of 255 characters and copies without problems. However: it makes my windows explorer crash. A file that’s in the Japanese.lproj map opens without problems. It’s strange as exFAT should (like NTFS) be: “32,760 Unicode characters with each path component no more than 255 characters“. Here I found the explanation: Windows Explorer (File Explorer in Windows 8 and Windows Server 2012) uses ANSI_API calls which is limited to 260 characters in the paths (MAX_PATH). So it’s not a problem with the filesystem, but with the Windows Explorer. Here is the solution: a microsoft hotfix. Unfortunately it doesn’t work for me. This might be a solution: \\?\F:\Etc.

After 700 GB of data on the disk:

  • Parkdale: 61/70, 1.12/0.53 (3.57/7.58 ms)
  • Crystal: 42/41, 0.62/0.25 (!!!??)
  • HDtune: 84 (115-50). 17.2 ms
  • AJA: 66/71
  • Atto: 7585-75297, full speed at 16 KB

At the end of the drive, it’s about 5 (read) -10 MB/s (write) slower then the WD. Also random is seriously slower. I’d like to compare that to the same drive but unencrypted.

So what are the speeds if the drive is unemcrypted? Now it’s 81 MB/s at 15% CPU usage. At 700 GB full:

  • Parkdale: 61/71, 0.99/0.45 (4.03/8.93 ms)
  • Crystal: 44/46, 0.62/0.24
  • HDtune: 84 (115-50), 17.3 ms
  • AJA: 67/72
  • Atto: 8849-76477, full speed at 16 KB)

There is not much of a speed test difference with all tests, but with real-life transfer it’s really a lot slower (60 vs 81 MB/s).

http://refit.sourceforge.net/ EFI boot menu for macs. http://gparted.sourceforge.net for resizing win partition. http://www.macworld.com/article/2018853/when-good-macs-go-bad-steps-to-take-when-your-mac-wont-start-up.html for mac startup problems. jMicron PCIe SD/MMC Host Controller. JMCR SD SCSI Disk Device: system32\DRIVERS\jmcr.sys. system32/jmcricon.dll

Installing win7/64 on SD card: it doesn’t show in the partitions. I guess I’ll have to install the SD cardreader first (can do that from USB). It does see a 200 MB partition 1 (primair), 325 GB osx partion  2, a  619 MB partion 3 and the 140 GB system partion 4 (bootcamp).

I did copy the old win7/32 installation with ??? but the SD card doesn’t show in the startup options (alt). So I installed win7/64 on the primary disk. Then I found that my ultimate serial doesn’t work for the professional installation, but it’s running anyhow (takes 16 [new installation] -20 GB [after updates] on the disk). Then I had to install drivers for the WLAN adapter. I used the free PCwizard to find out which devices are in the MBP, it’s the broadcom 802.11n network adapter, I’ve tried several but they didn’t work, so I ended up installing the bootcamp 5.0.5033 drivers (566 MB download, 820 MB on the USB stick). After that I updated the GeForce GT 330M drivers.

So next up was to download a win7/64 bit ultimate ISO and put in on a bootable USB with ISOtoUSB. Another way would be to mount the iso and make a bootable disk with command prompt.


Another thing that will speed up your drive is defragmentation and using caching programs like FancyCache.

When you test SSD, think about measuring steady state performance. See more about that here.

Posted in: Computers