- Joined
- Jan 27, 2011
- Location
- Beautiful Sunny Winfield
I've been fooling around with Android development and part of the tool chain is an ARM emulator that runs on the host development system and supports testing and debugging of the Android apps. It's pretty CPU intensive since it reads and interprets the ARM binaries (which themselves are reading and interpreting Java bytecode.) I have the dev tools installed on three systems and noticed that the performance between the three differs widely. (I'm inclined to say "wildly" 
)
First is my now ancient Thinkpad T500 with Intel Core 2 Duo T9400 processor and 8GB RAM. Second is my signature system with an AMD Phenom II X4 820 and 12GB RAM. The last is the wife's Lenovo G570 with Intel Core i3 2350M and 4GB RAM. The first two systems run x64 Linux and the last x64 Win7. The emulator is single threaded and pegs that core. All of the systems seemed to have additional free RAM so it does not seem like RAM availability was an issue and I do not think that disk performance factored in past the point where the emulator was read from disk. Below is a table I compiled of the comparison. I also listed the PassMark single thread score since I expected the performance to approximately match that.
Obviously the CPU Mark was an extremely poor predictor of emulator startup time. The systems' PassMark scores are within 10% of each other yet the performance varied by by a factor of three (and ranked opposite of CPU Mark.) The disparity surprised me and leaves me wondering why this is so.
Some other interesting things... With SWMBO logged on on her laptop, it took over 15 minutes to start the emulator. I cut that nearly in half by restarting the PC to log her off. (Sorry honey!
) Also I was surprised when it took about 15 minutes to unzip the Android tool chain on her PC. It only took about 15 seconds on either of mine. Her laptop uses a relatively slow spinning rust drive and I've upgraded to SSD, but it still seems like a HUGE difference.
Anyway... I was curious if anyone here had more insight into the CPU disparity.
And since I completed the class, I have explored other ways to run the emulator that gain better performance (such as using virtualization.) But during the course I was reluctant to mess with the setup when homework was due.
)First is my now ancient Thinkpad T500 with Intel Core 2 Duo T9400 processor and 8GB RAM. Second is my signature system with an AMD Phenom II X4 820 and 12GB RAM. The last is the wife's Lenovo G570 with Intel Core i3 2350M and 4GB RAM. The first two systems run x64 Linux and the last x64 Win7. The emulator is single threaded and pegs that core. All of the systems seemed to have additional free RAM so it does not seem like RAM availability was an issue and I do not think that disk performance factored in past the point where the emulator was read from disk. Below is a table I compiled of the comparison. I also listed the PassMark single thread score since I expected the performance to approximately match that.
Code:
processor single thread CPU Mark freq - GHz cache - MB time to start emulator (mm:ss)
Intel Core i3 2350M 1090 2.3 3 8:22
Intel Core 2 Duo T9400 988 2.53 6 2:54
AMD Phenom II X4 820 992 2.8 4 (128K L1, 512K L2) 4:20Obviously the CPU Mark was an extremely poor predictor of emulator startup time. The systems' PassMark scores are within 10% of each other yet the performance varied by by a factor of three (and ranked opposite of CPU Mark.) The disparity surprised me and leaves me wondering why this is so.
Some other interesting things... With SWMBO logged on on her laptop, it took over 15 minutes to start the emulator. I cut that nearly in half by restarting the PC to log her off. (Sorry honey!
) Also I was surprised when it took about 15 minutes to unzip the Android tool chain on her PC. It only took about 15 seconds on either of mine. Her laptop uses a relatively slow spinning rust drive and I've upgraded to SSD, but it still seems like a HUGE difference.Anyway... I was curious if anyone here had more insight into the CPU disparity.
And since I completed the class, I have explored other ways to run the emulator that gain better performance (such as using virtualization.) But during the course I was reluctant to mess with the setup when homework was due.
![]-[itman](/forums/data/avatars/s/9/9309.jpg?1650325391){kind=avatar}
