Snapdragon vs Hummingbird (CPU Benchmark)

partyatmygaff

We all know the PowerVR GPU on the Hummingbird chipset wipes the floor with the adreno GPU on the Snapdragon but there is a marked difference in CPU power.

Many people believe that the Hummingbird (Samsung Galaxy S for all intents and purposes) has better CPU performance than Snapdragon phones like the Desire or the Nexus One. The benchmarks speak otherwise

HTC Desire @ 998MHz (Snapdragon SoC)

Samsung Galaxy S @ 1.0GHz (Hummingbird SoC)



As you can see the Snapdragon is much more efficient with executing native code than the Hummingbird. It also is slightly more efficient under the NEON instruction set.

antodeco

Interesting. Im getting:

C:601.
NEON: 319

partyatmygaff

On what phone?

eoferrall

what app you using to test this? searched CPU benchmark but it doesn't seem to have the same app

partyatmygaff

Setcpu

eoferrall

partyatmygaff wrote: »

Setcpu

ah it's a paid app. explains why not on market so! :rolleyes:

osnola ibax

I've gotten:

c. 604.766846ms
NEON: 305.023193ms

HTC Desire - Android 2.2
Baseband 32.43.22.32U_5.09.00.20
Kernel 2.6.32.15-g1fc4045
Software NeoPhyte Mod 1.7.1 Sense-FroYo FRF91

antodeco

Just ran it there with no underclocking/overclocking:

C: 589.996338ms
NEON: 307.128905ms

HTC Desire
2.2 (LeeDrOID_v1.8)

Baseband: 32.42.00.32U_5.09.00.08
Kernel: 2.6.32.15-gf9c0527

JLarsen

I'm waiting for the 2.2 update in September before I make any assumptions on which CPU is better.

partyatmygaff

A native benchmark isn't affected by software like 2.2

leahcim

partyatmygaff wrote: »

A native benchmark isn't affected by software like 2.2

If its not affected by software how come your desire is getting different results to antodeco's desire.

antodeco wrote:

Just ran it there with no underclocking/overclocking:

C: 589.996338ms
NEON: 307.128905ms

Surely if this is an accurate benchmark of CPU Hardware it should not take into account firmware and android versions.

Something doesn't seem right.

Either the benchmark algorithm gives different results for each run (which would render it pretty useless) or its benchmarking more than just CPU.

evolutionqy7

SetCPU is a paid app only in the market to show your support...you can download it free on XDA for free...

you can get it here http://forum.xda-developers.com/showthread.php?t=505419

hmm i never check the benchmark with the HD2 ill report later

Why Should I

http://www.computerworld.com/s/article/9180475/Qualcomm_to_ship_fastest_chip_for_phones_tablets_in_Q4

partyatmygaff

leahcim wrote: »

If its not affected by software how come your desire is getting different results to antodeco's desire.



Surely if this is an accurate benchmark of CPU Hardware it should not take into account firmware and android versions.

Something doesn't seem right.

Either the benchmark algorithm gives different results for each run (which would render it pretty useless) or its benchmarking more than just CPU.

I could get those results as well, it takes a few runs and antodeco more than likely had run a task killer beforehand. Native code benchmarks aren't affected by the JIT compiler as do other benchmarks like Linpack or Quadrant. In any case, the Galaxy S being used for the test is running on Android 2.2

evolutionqy7

HTC HD2 @ 1GHz (running of the MircroSD)

JLarsen

I get the following results with SetCPU 2.0.2 on stock firmware (2.1):

C:843.199707ms
NEON: 334.843506ms

and that's the native bench.

EDIT: on subsequent native bench I'm unable to get above 830/333, avg. is around 824/333

partyatmygaff

JLarsen wrote: »

I get the following results with SetCPU 2.0.2 on stock firmware (2.1):

C:843.199707ms
NEON: 334.843506ms

and that's the native bench.

EDIT: on subsequent native bench I'm unable to get above 830/333, avg. is around 824/333

I get the same results on 2.2 . Software revisions will not influence a native code benchmark (usually).

JLarsen

partyatmygaff wrote: »

I get the same results on 2.2 . Software revisions will not influence a native code benchmark (usually).

Your screenshots on page 1 says otherwise

Also, the 2.2 leaked roms out there are unstable at best,
I'll wait for a better optimized rom before I make up my mind.

leahcim

Its hard to believe this benchmark is accurately comparing the CPU performance of these chips.

For the Snapdragon the results reported go from evolutionqy7's HD2 getting NEON:280 to partyatmygaff's Desire getting NEON:339 and for the Hummingbird from NEON:382 to NEON:334.

Thats a 20% differential, that is just two wide a gap.

Benchmark algorithms are ment to be some way deterministic and repeatable, you might expect a small differential due to atmosperic conditions but nothing like 20%.

I would guess the OS and Firmware versions are having more of an impact that we think.

conor.hogan.2

This is nice - but just looking at reviews and vids and anecdotal evidence shows that the samsungs run slower than the htc phones.

They just seem laggy at times - laggy like G1 laggy.

partyatmygaff

JLarsen wrote: »

Your screenshots on page 1 says otherwise

Also, the 2.2 leaked roms out there are unstable at best,
I'll wait for a better optimized rom before I make up my mind.

Both were running 2.2 in those photos so i'm afraid they don't say otherwise. A stable or unstable ROM won't affect a native code benchmark, if we were talking about pre-alpha ROMs then you may have a point but this is what we would call a fairly advanced beta therefore it's as good as it can get for executing native code.

leahcim wrote: »

Its hard to believe this benchmark is accurately comparing the CPU performance of these chips.

For the Snapdragon the results reported go from evolutionqy7's HD2 getting NEON:280 to partyatmygaff's Desire getting NEON:339 and for the Hummingbird from NEON:382 to NEON:334.

Thats a 20% differential, that is just two wide a gap.

Benchmark algorithms are ment to be some way deterministic and repeatable, you might expect a small differential due to atmosperic conditions but nothing like 20%.

I would guess the OS and Firmware versions are having more of an impact that we think.

I did only one run on the Desire ((Although I did multiple runs on the Galaxy S). I'm quite sure I could have managed NEON:280 with a few tries.

Software and Firmware revisions should not have any tangible affect on native code benchmarks.

JLarsen

partyatmygaff wrote: »

Both were running 2.2 in those photos so i'm afraid they don't say otherwise.

But they do say otherwise, you're telling me you get the same results as me on 2.2, but your screenshots show over 100ms diff. Saying you can rerun the native bench a few hundred times to get the same result as me is bollocks too. I've run quite a few and never once gotten over 843ms, and most of them were in the 824 area.

IMO leahcim is right in saying the OS and Firmware plays a role as the diff is not just marginal. Once again, I'll await the official FroYo so I can compare the Desire and the Galaxy S on equal footing.

partyatmygaff wrote: »

A stable or unstable ROM won't affect a native code benchmark, if we were talking about pre-alpha ROMs then you may have a point but this is what we would call a fairly advanced beta therefore it's as good as it can get for executing native code.

The 2.2 you're running is a month old leak, unfortunately Samsung has addressed the leaks, so I guess we won't see anymore before the official release. But I'm sure they've come a long way in optimizing the rom in a month and with another month or so before official release (if they're on time), there's plenty room for even more optimization.

partyatmygaff

JLarsen wrote: »

But they do say otherwise, you're telling me you get the same results as me on 2.2, but your screenshots show over 100ms diff. Saying you can rerun the native bench a few hundred times to get the same result as me is bollocks too. I've run quite a few and never once gotten over 843ms, and most of them were in the 824 area.

IMO leahcim is right in saying the OS and Firmware plays a role as the diff is not just marginal. Once again, I'll await the official FroYo so I can compare the Desire and the Galaxy S on equal footing.

Just out of curiousity, could someone still running a Desire on 2.1 run the native benchmark and let us know what the results are like?

evolutionqy7

came across this thought it would be relevant

http://www.phonearena.com/htmls/Snapdragon-vs.-Hummingbird-vs.-OMAP---the-mobile-CPU-war-beyond-1GHz-article-a_12546.html

cnocbui

A slightly disingenuous and factually incorrect article.

Apple is not a leading SoC player. Maybe at some point in the future they might be, but not at present.

the iPhone 4, is having a custom designed chipset called A4, which is running the iPad as well. Samsung developed the Hummingbird platform based on intellectual property from Intrinsity, a processing solutions company that Apple bought last year for $121 mln to lock in the A4 exclusivity for its own mobile gadgets. Therefore, when dissected, both Hummingbird and A4 share a lot of commonalities, and we’d assume they are fairly similar in basic capabilities too.

And that bit is just pure bull dust.

The A4 is not a custom chipset and Apple didn't commission it or have any part in it's design.

Samsung contracted Intrinsity to optimise the ARM design for the Hummingbird, which they did. It was only afterwards that Apple bought Intrinsity. The A4 is just re-badged Hummingbird, the design of which Apple played no part in, AFAIK.

I suppose you could say the A4 variant that is used in the iPad is custom tailored by Apple - they ordered a version that only had 256mb of RAM instead of the Hummingbirds normal 512mb.

So not only did Apple not play any part in the Hummingbirds creation, their custom variant as used in the iPad is actually inferior.

I found this by accident, earlier today:
Samsung Galaxy S I9000

JM5+SamSet1.9d+LagFix2.3





The guy has some nice landscape photos taken in Kerry as well: http://www.flickr.com/photos/obecny/4888389946/in/photostream/

partyatmygaff

cnocbui wrote: »

A slightly disingenuous and factually incorrect article.

Apple is not a leading SoC player. Maybe at some point in the future they might be, but not at present.

And that bit is just pure bull dust.

The A4 is not a custom chipset and Apple didn't commission it or have any part in it's design.

Samsung contracted Intrinsity to optimise the ARM design for the Hummingbird, which they did. It was only afterwards that Apple bought Intrinsity. The A4 is just re-badged Hummingbird, the design of which Apple played no part in, AFAIK.

I suppose you could say the A4 variant that is used in the iPad is custom tailored by Apple - they ordered a version that only had 256mb of RAM instead of the Hummingbirds normal 512mb.

So not only did Apple not play any part in the Hummingbirds creation, their custom variant as used in the iPad is actually inferior.

I found this by accident, earlier today:
Samsung Galaxy S I9000

JM5+SamSet1.9d+LagFix2.3





The guy has some nice landscape photos taken in Kerry as well: http://www.flickr.com/photos/obecny/4888389946/in/photostream/

We all know the Galaxy S can score higher on Quadrant. That's a full system benchmark after all. It takes into account file system speed (NAND Flash is faster than a Class 2 SDCard) as well as it's GPU. Neither of which matter the slightest in this thread which is purely for comparing both CPUs.

cnocbui

Looks like Samsung is fixing things:

The Samsung Epic 4G is the fastest Galaxy S phone to hit the market. With Samsung’s latest optimizations, this phone flies. We’d venture to say it is faster than the Google Nexus One, Motorola Droid, Motorola Droid X, or any other competing Android 2.1-sporting device. With Android 2.2 Froyo coming soon this year, this phone will undoubtedly rock the mobile world.
While other Galaxy S phones have forced users to place custom launchers (namely LauncherPro) over TouchWiz, we can say without a doubt that the Epic 4G’s Touchwiz UI runs smooth as butter. So, if you like what Samsung’s done to the user experience, you’ll be able to keep the Touchwiz interface without losing any speed advantage.
This phone won’t need a custom Ext2, Ext3, or Ext4 lag fix come launch time, it is snappy and ready to go right out of the box. Overall, Samsung gets a 10/10 for getting it right with this one. Well done.

http://briefmobile.com/samsung-epic-4g-benchmarked

partyatmygaff

cnocbui wrote: »

Looks like Samsung is fixing things:



http://briefmobile.com/samsung-epic-4g-benchmarked

There doesn't appear to be any spec change, it's still running the hummingbird. This isn't about whether or not Samsung have fixed their software and the incessant lag issues on touchwiz. It's a question of raw power which isn't affected by poorly optimized UIs. Aka completely irrelevant to this thread.