Parallels vs. Boot Camp vs. VMware Fusion Benchmarks

Parallels vs. Boot Camp

edit: I snipped this post to comply with a competitor's EULA. Maybe the admins should just delete this thread.

--------------------------------------

Hello fellow Parallels users!

I regularly use Parallels and Boot Camp<snip>

<snip>I decided to run some computation tests.

The test platform was my stock MacBook Pro 17" Core 2 Duo 2.33 GHz with 2GB RAM and 160 GB HD. I used two pi calculation programs on Windows XP: QuickPi 4 and PiFast 4.3. I believe they're both multi-cpu aware. While the results of PiFast didn't vary much, QuickPi shows some differences in computation speed.

<snip>

Boot Camp
QuickPi 21.78s
PiFast 32.13s

Parallels
QuickPi 28.07s
PiFast 32.97s

<snip>

These programs only test processor speed. The calculation of pi to 10 million decimal places fits in the memory allocated to the VMs (512 MB). So what do these results say?

Boot Camp
Run Boot Camp if speed is paramount.

<snip>

Parallels
Parallels is still pretty fast. It works great for me for what I use it for. (It even works great with Microsoft Word's built-in speech recognition, which itself isn't that good.) <snip>

What does this not mean?
These results do not mean that ... <snip> ... you should ditch virtualization and just use Boot Camp. Virtual machines are a lot more convenient than booting into a different OS via Boot Camp. I didn't test graphics responsiveness, memory speed, hard drive speed or an other factors that affect speed.

I invite anyone interested to test for themselves and post the results here!

 

Was your PI test-app SMP ready?

If not then the 2 CPU test was rather futile. The results suggest as much.

There are better ways to test performance... sometimes referred to as benchmarking software

 

RE: Benchmarks and Multi CPU

edit: I snipped this post to comply with a competitor's EULA. Maybe the admins should just delete this thread.

--------------------------------------

I'm pretty sure these apps are multi CPU. (See Note Below)

Also, I tried using a benchmark program called Performance Test and it gave strange results... <snip>, and the test times were wrong. I'll try it again and geekbench to see how they do and post the results here. Any other benchmark software ideas?

Multi CPU Note
On Mac OS X CPU utilization is shown as a percentage up to 100 * Num CPUs (Dual Cores max at 200%, quad at 400%, etc...) Windows shows CPU usage as a percentage up to 100% no matter how many CPUs you have. If using a CPU intensive, non-threaded program on a dual core machine, 50% is your max for that program.

Since Windows shows nearly 100% CPU, and the QuickPi specifically spits out the number of cores and other CPU info, I'm pretty sure it uses both cores. I don't understand virtualization 100% but since <snip> Parallels <snip> [is] more than half as fast as Boot Camp, I figure [it is] <snip> good about directing code to both CPUs efficiently, but maybe I just don't get this. Also, I think I read somewhere that they are multi cpu aware.

 

Benchmarking Fun!

edit: I snipped this post to comply with a competitor's EULA. Maybe the admins should just delete this thread.

--------------------------------------

So here's why I used pi calculation. I used Geek Bench to test Bootcamp, OS X, <snip> and Parallels. Geek Bench does calculations, compression tests, image manipulations and general memory tests. It provides an overall score. Consecutive tests on the same machine produce similar results (usually).

OS X: 3045
Boot Camp: 2494
Parallels: -26841186
<snip>

So the results are in! <snip> Parallels actually is so slow at geekbench that it somehow manages to undo time and reverse the laws and mathematics and physics as we know it. That or it's so fast that the score wouldn't fit into the ~2 Billion range of a signed 32 bit number and converted to a 32 bit unsigned number should really have a score of 4,268,126,110! Dang that's fast. 1,401,683 times faster than OS X.

Final conclusion? Run all of your software in Parallels and go warp speed! Or realize that benchmarking software usually depends on low level timings that don't translate well because of virtualization, or use software like a pi calculator that uses the system clock for timing.

I couldn't run PassMark's Performance Test 6.1 because my trial period is over, and I don't have other benchmarking software. I will keep digging for a more general program that works correctly under virtualization.

Cheers!

 

I have used Performance Test (passmark) and gotten good results. A man who knows how to travel to the past can usually get around the expiration.

 

Performance Test Benchmarks

edit: I snipped this post to comply with a competitor's EULA. Maybe the admins should just delete this thread.

--------------------------------------

I ran Performance Test and it corroborates my results with the Pi software.

Here's a summary of the results:
CPU:
Boot Camp 1344.3
Parallels 451.0
<snip>

2D Graphics:
Boot Camp 525.1
Parallels 476.5
<snip>

Memory:
Boot Camp 562.8
Parallels 461.4
<snip>

Hard Disk:
Boot Camp 170.1
Parallels 839.9
<snip>

I don't know why Boot Camp has such dismal disk performance. It doesn't feel 5 times slower than Parallels. As a note I use Omega's Drivers in Boot Camp. It shouldn't have affected the results for 2D though.

Well benchmarks aren't always indicative of real world performance, but it shows that <snip> [Parallels] could improve with respect to Boot Camp <snip>. <snip>

Note that the scores above are made up of subtests that each might perform differently, some better, some worse with respect to the other systems than the final score. Your mileage may vary.

Enjoy!

 

@SanJoseBen:

Just a heads-up. The Fusion EULA prohibits posting or discussing benchmark results outside of VMWare's forums. Personally, I always enjoy seeing comparisons like this and am appreciative of anyone who spends their time compiling them. I just don't want to see you get into hot water w/the VMWare folks.

Disk I/O results are impressive. VMWare's always excelled in that department.

 

Thanks, I missed that.