Inside CNET Labs: Windows virtual machine performance on the Mac
CNET Labs compares the Windows application performance on an Intel Mac of three different virtualization applications.
Note: This post was updated on August 17, 2007, to make a correction.
With VMware's official release of Fusion 1.0 less than two weeks ago, there are now no less than four different ways to run Windows applications on Intel-based Macs. Fusion, as well as SWsoft's Parallels Desktop 3.0 for Mac use virtualization technology to allow you to simultaneously run the Windows operating system as a virtual machine alongside the Mac OS. CrossOver Mac 6.0 from CodeWeavers uses a different virtualization approach by building on the open-source API, Wine, which allows you to run individual Windows applications in the Mac OS without needing to install or run the Windows operating system.
Perhaps the easiest and most common way to run Windows on a Mac today is with Apple's own Boot Camp Public Beta. While Boot Camp is currently the only way to run Windows natively on a Mac, it requires booting directly into the Windows OS, during which the Mac OS is not accessible (until you reboot the computer back into the Mac OS, leaving the Windows OS).
With all these different options for running Windows applications on Intel Macs, we wanted to see how they compare against each other in terms of performance; so CNET Labs applied its systems testing methodology to an eight-core, 2.66GHz Mac Pro running the Mac OS X 10.4.10. We used Boot Camp 1.3 Beta, CrossOver Mac 6.1.0, Parallels Desktop 3.0 for Mac (4560), and Fusion 1.0 (51348). Windows Vista Ultimate was used as the operating system for Boot Camp, Fusion, and Parallels. Fusion and Parallels were both set to 1,024MB of system memory, a 32GB hard disk, and the maximum amount of graphics memory they each supported (Fusion: 128MB, Parallels: 64MB).
(Shorter bars indicate better performance)
The Multimedia multitasking test is the most taxing of our tests. It encodes QuickTime video in the foreground, while iTunes simultaneously encodes audio in the background. While Fusion's performance on this test was significantly slower than we saw with Windows running natively under Boot Camp, Fusion was still more than 3.5-times faster than Parallels. One of the primary reasons why Fusion is faster here is that it supports up to two CPU cores, while Parallels supports only one CPU core. This test benefits greatly from multiple CPU cores, as is evidenced by the Mac OS and Boot Camp performance, each with eight cores available.
(Shorter bars indicate better performance)
Our Photoshop CS3 test results repeat what we saw with the Multimedia multitasking test: Fusion is faster than Parallels, but neither is as speedy as Windows running in Boot Camp. While the Photoshop CS3, Multimedia multitasking, and Cinebench (see below) tests alone are not enough to make an absolute conclusion of how Mac performance compares to Windows, the performance here does at least show that Intel Macs can be very competitive with Windows systems, and at times even faster.
(Longer bars indicate better performance)
 | Rendering Multiple CPUs | | Rendering Single CPU |
N/A
N/A
Maxon's Cinebench R10 test is a CPU-intensive test that renders a 3D photo-realistic scene. The same scene is rendered twice--once using only a single CPU core, the second time using all available processor cores. As Fusion is the only virtualization application here with support for more than processing core, it was the only virtualization application able to complete the multiple CPU-based render. As for single-core performance, Fusion and Parallels put in nearly identical performance. Interestingly, this was the only one of our tests that CrossOver Mac could run. Its single-core performance on this test was marginally better than the other two virtualization applications.
(Longer bars indicate better performance)
The only virtualization software we could get to run Quake 4 was Parallels. We used somewhat conservative settings by today' standards (1,024x768, High Quality, 4x antialiasing, 8x anisotropic filtering) and saw frame rates that were barely playable. Lower quality settings should show higher, more playable frame rates, but wouldn't look as good. However, we were pleasantly surprised to find that we could get a Windows-based game to run at all in virtualization software! The developers of the virtualization apps promise improved compatibility with Windows games in the near future.
From a pure performance perspective, Fusion proves to be a faster performing platform than Parallels. Additionally, in our anecdotal hands-on testing we found Fusion to be somewhat more stable than Parallels--which is not necessarily a fair comparison, as we've spent more time using the shipping version of Parallels than we have Fusion. Usability between Fusion and Parallels also differs slightly, and personal preference will likely depend on individual tastes--I find Parallels a bit more intuitive than Fusion, but this could also be a function of my having spent more time with Parallels. CrossOver still has a lot of catching up to do, but the folks over at CodeWeavers appear to be making progress all the time. To date, mainstream gaming on the Mac still has only one significant outlet: using a native Windows install via Boot Camp. All that said, if you all you need to do is run more pedestrian Windows applications on your Mac, such as Outlook or Excel, you are not likely to notice significant performance differences between any of the virtualization options mentioned here--for less resource-intensive applications, the performance of any of these virtulization applications is probably fast enough.
Find out more about how we test systems.
System configuration:Apple Mac Pro
2x 2.66GHz Intel Xeon 5355 CPUs (8 total cores); 2.048MB DDR2 FB-SDRAM 667MHz memory; 512MB ATI Radeon X1900 graphics card; 500GB Seagate 7,200rpm SATA/150 hard drive