Android 2.2 - as fast as you wished!

[Hugh Jass]

I get 9.915mflops in linpack with BB1.0 at 1.2ghz lv. The JIT is doing something, that's for sure be it synthetic or not.

Umm... from what I see here LinPack scales nearly linearly with clockspeed.

Then what explains the variation between the results for Droid and Nex? Differences in on board memory or cache?

 

[doublejack]

I'm very skeptical that JIT could actually have this big of an impact in real world applications. I guess we'll see when it arrives.

I wonder how fast it is on 3g vs. the wifi they are running, haven'tweI seen high 25's on the droid on wifi

If you look at the records, people have scored into the 30's already on a Droid.

 

[Hugh Jass]

Sorry what I'm saying is isn't there a big speed difference doing a benchmark test on WiFi vs doing it on 3G

The linpack benchmarks are assessing the performance of the processor inside your phone, where as other benchmarks might assess the speed of your internet connection (such as speedtest.net) therefor the data connection is irrelevant in these tests.

 

[doublejack]

I get 9.915mflops in linpack with BB1.0 at 1.2ghz lv. The JIT is doing something, that's for sure be it synthetic or not.

Umm... from what I see here LinPack scales nearly linearly with clockspeed.

Then what explains the variation between the results for Droid and Nex? Differences in on board memory or cache?

LinPack scales linearly on a given platform. The Nexus One has a different CPU than the Droid, thus 1GHz on the Nex does not equal 1GHz on the Droid.

 

[Darkseider]

I'm very skeptical that JIT could actually have this big of an impact in real world applications. I guess we'll see when it arrives.

I wonder how fast it is on 3g vs. the wifi they are running, haven'tweI seen high 25's on the droid on wifi

If you look at the records, people have scored into the 30's already on a Droid.

THIS. I don't see JIT having a huge impact on real world applications. There may be some performance boost but nothing like the 4.5 - 5x increase that the benchmark is seeing.

 

[gatewayad]

Sorry what I'm saying is isn't there a big speed difference doing a benchmark test on WiFi vs doing it on 3G

The linpack benchmarks are assessing the performance of the processor inside your phone, where as other benchmarks might assess the speed of your internet connection (such as speedtest.net) therefor the data connection is irrelevant in these tests.

Thanks, didn't know there was a difference, now I do

 

[Hugh Jass]

Interesting writeup right here:

The HTC Nexus One with Android 2.1 received a score of 6.5 to 7 MFLOPS, still impressive compared to the HTC Hero's lesser score of 2 MFLOPS. The Nexus One with Android 2.2 blows both of them away, though, posting a score of 37.593. That's a 450 percent performance gain over Android 2.1, at least. To put that further in perspective, an Eee PC scores about 66 MFLOPS, at max.

The Linpack results appear to have come thanks to the new just-in-time (JIT) compiler in Froyo (Android 2.2). JIT compilers boost performance of interpreted codes like Java. While this won't help native apps directly, it means that many of your apps will get at least a 2 to 3-fold speed increase. And native apps should indirectly benefit, as faster non-native apps means more CPU freed for native ones.

Link

So it seems real world increases will be seen. dancedroid

From my post in another thread.

 

[ConqSoft]

My "stock" moto droid gets 4.5

 

[adrynalyne]

I'm very skeptical that JIT could actually have this big of an impact in real world applications. I guess we'll see when it arrives.

I wonder how fast it is on 3g vs. the wifi they are running, haven'tweI seen high 25's on the droid on wifi

If you look at the records, people have scored into the 30's already on a Droid.

JIT will have NO impact on applications unless they are compiled to use it.

 

[Darkseider]

My "stock" moto droid gets 4.5

Again backing up the linear scaling of Linpack

.8 * 5.5 = 4.4

 

[aminaked]

That's a huge speed increase but believe me, JIT makes a huge difference when it comes to java. I remember working on some fancy java applets several years ago. As soon as JIT-enabled VMs came out for browsers, the speed of these applets went from annoyingly clunky to blazing fast. JIT makes a very big difference. I don't know about 5x faster, but it is noticeable whatever it is. Very exciting!

 

[Darkseider]

I'm very skeptical that JIT could actually have this big of an impact in real world applications. I guess we'll see when it arrives.

I wonder how fast it is on 3g vs. the wifi they are running, haven'tweI seen high 25's on the droid on wifi

If you look at the records, people have scored into the 30's already on a Droid.

JIT will have NO impact on applications unless they are compiled to use it.

Then a question. If an application has to be compiled to use it then can we assume that the Linpack benchmark is? Otherwise that statement holds no water. Next question is IF that statement is indeed true and the app. has to be compiled to take advantage of JIT. If JIT is not present on a device will the app. still run?

 

[aminaked]

JIT will have NO impact on applications unless they are compiled to use it.

That's not true. The virtual machine handles JIT. The exact same apps will be faster.

 

[adrynalyne]

Yes, Linpack has been compiled to use it. Even Google devs have admitted that there are no real world applications outside of benchmark apps that can test it.

I'm very noob when it comes to Java, but I'm pretty sure if JIT is not present for a JIT enabled app, it will fallback to other compilers.

 

[adrynalyne]

JIT will have NO impact on applications unless they are compiled to use it.

That's not true. The virtual machine handles JIT. The exact same apps will be faster.

No...

If a java application does not use JIT, it cannot take advantage of it.