Whilst performing some (frankly needlessly painful) IT support on our internal SCM server I embarked on a small exercise to determine how ‘fast’ my personal collection of drive solutions actually are in the real world (or as near as can be judged).

In this instance, I have a number of SSD’s floating about, and my development PC has a small set of internal SSD and HDD’s, so I decided to see how they performed with a view to redistributing them based on my actual usage of the drives. I have an SSD for the operating system (of course) and also an SSD dedicated to all the source code and assets for our projects to make a nod towards improving build times. I also use a RAMDrive as a build cache for Visual Studio, to also dramatically increase build times (and also save needless wear and tear on any SSD).

For fun I also ran the same test on my 2011 Mac Air, and the RAMDrives on both the dev PC and the Mac Air. The results surprised even me!

Now, I don’t sell this as a scientific analysis and frankly YMMV, but I think the results are worth posting here.

I looked at (and tried) a number of (free) tools, most were ‘ok’ but one stood out – the excellent CrystalMark for Windows. If you decide to try this out, I do encourage you to donate when you download – as it’s a tool you tend to use only once (or infrequently).

The tool is quite simple, it runs a pattern of sequential and random read/write tests on a chosen drive, and allows you to set how many repeat tests, the total file size to read/write and how many background threads are used to run the tests. This allows you to set a reasonably rigourous pattern of stress tests, to get the best average results. For my purposes, I just wanted a quick indication, so I chose two repeat tests, a 1Gb filesize and left all the other options as default. For the RAMDrives, I had to dropped to a 256Mb filesize, due to their native sizes and available free space. I may run all the tests with 256Mb filesizes to get a more consistent set of results if I get time.

The test subjects were;

  • 2TB (Make, RPM) HDD
  • 1TB (Make, RPM) HDD
  • 128Gb OCZ SSD
  • 240Gb Kingston UV300 SSD
  • 64Gb Kingston SSD
  • 256Gb 2011 Mac Air SSD
  • 1Gb RAMDrive (32Gb Corsair Xtreme, MB, 3Ghz 64b i5 CPU)
  • 512Mb RAMDrive (2011 Mac Air, 8Gb RAM, 2.1Ghz 64b i7 CPU)

As the tool is Windows based, the Mac Air test were ran within a Windows 10 Boot Camp installation, launched directly and not via Parallels. As CrystalMark runs four read and four write tests, I averaged the read/write tests. Now, you may say this is unfair given the range of values, but I think it gives a better and fairer indication of ‘real world’ usage and makes for a better comparitive value to judge against other drives.

The results

Read Speeds (MB/s) – sorted by best average read bandwidth

Drive Sequential 1 Random 1 Sequential 2 Random 2 Average
AMD RAMDrive (Mac Air) 3758.81 703.89 6,105.87 733.60 2,825.54
AMD RAMDrive (Dev PC) 2173.25 210.16 5,204.30 313.06 1,975.19
Mac Air 256Gb SSD 824.7 221.00 640.60 17.70 426.00
OCZ 128Gb SSD 400.76 233.68 373.22 21.87 257.38
Kingston 240Gb UV300 SSD 284.06 214.62 274.06 26.35 199.77
Kingston 64Gb SSD 254.34 13.63 248.95 13.30 132.56
1TB HDD 122.12 1.11 124.15 0.48 61.96
2TB HDD 80.46 0.91 80.97 0.36 40.67


Write Speeds (MB/s) – sorted by best average write bandwidth

Drive Sequential 1 Random 1 Sequential 2 Random 2 Average
AMD RAMDrive (Mac Air) 4909.431 641.91 7,571.61 665.38 3,447.08
AMD RAMDrive (Dev PC) 3446.044 330.56 6,523.09 150.63 2,612.58
Mac Air 256Gb SSD 670.4 150.10 560.00 35.88 354.10
OCZ 128Gb SSD 378.45 169.64 366.36 81.31 248.94
Kingston 240Gb UV300 SSD 270.244 191.70 260.71 76.43 199.77
1TB HDD 105.633 0.96 115.57 1.02 55.80
Kingston 64Gb SSD 98.564 17.94 50.34 13.24 45.02
2TB HDD 79.308 1.07 78.64 1.06 40.02


Naturally, the RAMDrives are blisteringly fast compared to SSD, but they help establish a single clear winner – the 2011 Mac Air!

This surprised me – although I enjoy using the machine, I’ve never considered it to be ‘that fast’ when dealing with files. Fast, yes, but that fast? Subjectively it feels no faster than the main dev PC I use, although thinking about it now it does feel ‘as fast’, which is interesting given the Mac Air has a slower CPU, less RAM, and so on. I’d imagine this is testimony to Apple’s very tight bus integration across the motherboard, but certainly the SSD speed alone is remarkable, and when combined with the RAM speeds which outperform high performance PC RAM, I’d say the Mac Air is definitely a nicer piece of kit than I’d previously thought!

0 Responses

  1. […] the recent article comparing HDD vs SSD vs RAM, I thought I’d expand a little on using a RAMDrive as a build cache with Visual Studio 2015. […]