GLBenchmark 2.0 Results

[Lazy8s]

The GLBenchmark 2.1 suite has been released to the App Store and Android Markets. $5 for iOS and free for Android (I have both platforms yet won't complain too much about the inequality! I do appreciate the tool and the off screen option.)

My only iPad 2 run-through so far has produced some results, at least in the sub test for texture fill, that look noticeably higher than that of the database at the website. Getting close to 1G tex/sec now. I want to try the off screen test now, too.

I left a review in the market. To make it more user oriented in a future update, having a scene test where the user can toggle different graphics effects like AF, soft and hard shadows, certain texture effects, AA, etc. to see the visual and performance impact would be a useful option. I like that little aspect of the GL ES Extension Viewer app.

Anyway, thanks for the tool, Kishonti!

 

[Xmas]

My only iPad 2 run-through so far has produced some results, at least in the sub test for texture fill, that look noticeably higher than that of the database at the website. Getting close to 1G tex/sec now.

I don't see how that's noticeably higher than what the database shows.

 

[Lazy8s]

Yeah, I noticed it had changed from the ~0.95G tex/sec to the current 1G after I went back to look.

My results for the triangle "warm up" test (the definition of which I had questioned before but now understand to be a test where it renders iterations of white triangles with no lighting) was around 68M tri/sec if I remember correctly, so the results I submitted will hopefully update the previous 64M figure there.

 

[Lazy8s]

Looks like it might've updated already or else I keep accidentally toggling my filter for "Community Uploads".

 

[ltcommander.data]

http://www.glbenchmark.com/result.j...s=4&version=latest&certified_only=2&brand=all

So the new offscreen tests run the devices at equal resolution for fair comparison right? The results show the original iPad and iPhone 4 posting the same results even though the Apple A4 in the iPad is supposed to be clocked faster. What's more the iPhone 3GS results are barely slower than the iPad and iPhone 4. Is the test bottlenecking somewhere since I didn't think the Apple A4 being faster than the previous gen SoC was in dispute.

 

[Deleted member 13524]

Great, an apples-to-apples benchmark for mobiles!


Other fun facts:

- Amlogic's AML8726 with a Mali400MP1 and a Cortex A9 with 128KB L2 scores better than A4 and almost as good as a Hummingbird with the old Froyo driver.

- Rockchip's RK2918 with a Cortex A8 and a Vivante GC800 scores on par with the Hummingbird with the new driver and better than some smartphones with Tegra 2.

These chips are being bundled with 100->180€ tablets!

Looking at the GC800's results, i.MX6 with a GC2000 could very well beat Kal-El in 3D performance.



I hope they release a Symbian^3 version. Can't wait to see what the BCM2763 with 128MB of dedicated memory is capable of.

 

[metafor]

http://www.glbenchmark.com/result.j...s=4&version=latest&certified_only=2&brand=all

So the new offscreen tests run the devices at equal resolution for fair comparison right? The results show the original iPad and iPhone 4 posting the same results even though the Apple A4 in the iPad is supposed to be clocked faster. What's more the iPhone 3GS results are barely slower than the iPad and iPhone 4. Is the test bottlenecking somewhere since I didn't think the Apple A4 being faster than the previous gen SoC was in dispute.

I don't know how Apple clocks their SoC's but does the bus/GPU frequency diminish with its CPU clock? If not, then the only thing that may have changed is the CPU peak clock between the A4 in the iPad and iPhone 4.

This also applies to the iPhone 3Gs, which had the same SGX535 GPU which, I'd wager, is similarly clocked as the iPhone 4's.

 

[Ailuros]

This also applies to the iPhone 3Gs, which had the same SGX535 GPU which, I'd wager, is similarly clocked as the iPhone 4's.

With higher GPU performance on the iPhone4 at a higher resolution? I don't think so.

 

[metafor]

With higher GPU performance on the iPhone4 at a higher resolution? I don't think so.

Say what? The iPhone 4 doesn't show higher performance when normalized for resolution...

 

[Deleted member 13524]

With higher GPU performance on the iPhone4 at a higher resolution? I don't think so.

Check the link in post #45.

The new "offscreen" benchmark in 2.1 tests all the devices at 1280*720.
iPad, iPhone4 and 3GS all show the same result, within a small margin of error.

 

[Exophase]

About 7% in Egypt and 10% in Pro, I'd call that a little higher than "small margin of error", but yeah. It's telling just how different the low level tests are, considering you'd think a majority of them were resolution independent.

It seems like the scores should scale pretty closely with GPU clock so long as they haven't become CPU or bandwidth limited. But there could also be slight differences due to configurable parameters in the core, which may include metrics such as texture cache sizes. Different OS/drivers would also affect it - another thing that bothers me is that a whole bunch of OS versions and driver versions are listed (for both 3GS and iPhone 4). Is the final score the average of several?

 

[ltcommander.data]

About 7% in Egypt and 10% in Pro, I'd call that a little higher than "small margin of error", but yeah. It's telling just how different the low level tests are, considering you'd think a majority of them were resolution independent.

We know the CPU went from 600Mhz in the iPhone 3GS to 1GHz in the iPad. It would seem lob-sided for Apple to scale the CPU 67%, but only clock the GPU 10% higher given the move from 480x320 to 1024x768.

http://browse.geekbench.ca/geekbench2/view/479944
http://browse.geekbench.ca/geekbench2/view/479881

Synthetic Geekbench memory bandwidth tests show the iPad having 50% more memory performance than the iPhone 3GS on the same iOS 4.3.3 and we know the memory controller went from 32-bit to 64-bit so memory bandwidth shouldn't be the bottleneck.

 

[Exophase]

I don't know exactly what the GPU was clocked at for 3GS, but sources suggest somewhere between 150-200MHz, which is on the aggressive side for 65nm; compare with OMAP34/35xx which clocked its SGX530 at 110MHz. 600MHz for the Cortex-A8, on the other hand, was not aggressive at all. So I wouldn't be that surprised if CPU clocking got a much more substantial boost. It was well known that iPad's GPU performance relative to screen resolution was lacking compared to 3GS.

 

[Ailuros]

Say what? The iPhone 4 doesn't show higher performance when normalized for resolution...

True. Yet see all the other posts above and:

http://www.glbenchmark.com/compare.jsp?benchmark=glpro21&showhide=true&certified_only=1&D1=Apple%20iPhone%204&D2=Apple%20iPhone%203G%20S

That for a 4x times increase in pixels. One can never be sure with Apple, but I had heard from developers in the past that the 3GS is clocked at 150MHz and the 4 at 200MHz; iPad1 at 250MHz.

In fact the last is so far my major point to believe that the GPU block in iPad2 is clocked also at 250MHz, or better frequency between iPad1 and iPad2 should be the same otherwise the up to 9x times increase in GPU performance breaks.

 

[Lazy8s]

Transformer TF201 results were removed.

I wonder if they'll make noticeable performance changes before shipping.

 

[Arun]

Transformer TF201 results were removed.

I wonder if they'll make noticeable performance changes before shipping.

Given that the Kal-El GPU is basically a clockbumped Tegra 2 GPU with an extra pixel shader pipeline, which itself was basically a clockbumped Tegra 1 GPU, you'd certainly hope the drivers are mature by now and there are only small incremental improvements left for major benchmarks. I suppose something could have been wrong with that test platform or the clocks weren't final, but the numbers look roughly believable to me.

 

[Lazy8s]

So, I got the iOS 5 update for my iPad 2 -- very painless set up considering all of the iCloud configurations that could potentially be made -- and ran GLBenchmark 2.1 to make sure my stuff was matching up to the new standards. The rendering errors went away, and my run set some new high scores on the benchmark: 10034 for Egypt offscreen.

I received a very negative surprise, though, when I started testing the updates to the browser: a very slight delay in response to pinch-zoom! Now, I know I said "very slight", but the foundation of Apple's software superiority has always been in responsiveness. It'll probably be imperceptible to most and still feels better than most Android devices, but It's so un-iOS-like that it's jarring to me. Apple definitely needs to put that back to the way it was...

... someone else feels my pain (and it looks like it's been that way for several months in beta):
http://forums.macrumors.com/showthread.php?t=1188890

 

[Pressure]

So, I got the iOS 5 update for my iPad 2 -- very painless set up considering all of the iCloud configurations that could potentially be made -- and ran GLBenchmark 2.1 to make sure my stuff was matching up to the new standards. The rendering errors went away, and my run set some new high scores on the benchmark: 10034 for Egypt offscreen.

I received a very negative surprise, though, when I started testing the updates to the browser: a very slight delay in response to pinch-zoom! Now, I know I said "very slight", but the foundation of Apple's software superiority has always been in responsiveness. It'll probably be imperceptible to most and still feels better than most Android devices, but It's so un-iOS-like that it's jarring to me. Apple definitely needs to put that back to the way it was...

... someone else feels my pain (and it looks like it's been that way for several months in beta):
http://forums.macrumors.com/showthread.php?t=1188890

Mine is working as it always did.

Did you make a full restore or just a normal update?

 

[Lazy8s]

Well, I didn't restore factory settings as a separate step before selecting to update at the "a new version of iOS is available" prompt, but I think iTunes master reset the device (all of the apps were removed) as part of the update process.

I really think it feels like the result of a design change... not the goal of the change but some side effect of it. I might try wiping the device and reinstalling again, but I have a feeling it's not a glitch. I'm guessing they'll roll a correction into a near-future update with the group of other bugs they find post-launch.

Other than that, I'm mostly very pleased with the performance of iOS 5.

 

[idsn6]

Try disabling four-finger multitasking gestures in Settings. When I do so, pinch-zoom is once again instant.

I believe this is a heuristics issue, not performance: iOS needs more motion to determine which gesture is being input. Personally, the multitasking gestures are more than worth the slight delay.