iPhone 5 tops benchmark chart Apple's upcoming iPhone 5 is one of the fastest ARM devices out there, beating even the Asus-made Google Nexus 7, if benchmark data posted online are to be believed. An entry in the Geekbench database spotted this weekend - the figures were uploaded yesterday - shows the new handset clocking up a score of 1601. The readout …
But it isn't anything new - there *are* other phones out there, ignoring whatever OS is on them, because it seems mentioning "Android" suddenly polarises people (oops), which achieve virtually equal performance, equally good battery performance, in equally slim designs, but they've been on sale for months.
Like you, I'm not an iPhone lover, but neither am I knocking the new one - good design, bulletproof build quality, good innards, but let's stop calling the thing technologically special, let alone revolutionary.
The year's flagship Android phones seem to come out in the spring and summer, and the iPhone now seems to be released in September or October. Therefore there is a period of about three months of the year in which the iPhone lags badly in terms of hardware capabilities. Then the iPhone catches up, as it has here.
An extra 221 points with a new OS? Well, possibly.
Depends very much on the benchmarking but a change in the compiler which turned on bits of hardware could easily do that. This is why Intel is still in the compiler business.
But the comparison is spectacularly underwhelming given the predominance of FPU calculations. What do we use those for on hours (on a daily basis)?
It can also be down to a kernel sub-system getting an overhaul. To take an Apple example, they garnered a considerable improvement when one of the original authors of the virtual memory sub-system pointed out a potential improvement that had been overlooked. (This stemmed from unfinished work years before on the Mach kernel that had been absorbed into NeXTSTEP before it was reworked as the basis of Mac OS X).
A more Linux related example would be a switch in the behaviour of the scheduler, since there is often a trade off between optimal scheduling of server versus desktop related processes. This was the root of a kerfuffle in the Linux world a few years back, when the scheduler was more optimal for server stuff and an alternative scheduler was made available. The actual kerfuffle was down to Linus Torvalds refusing to make it easier for schedulers to be "pluggable" in the mainstream sources.
No, I mean Jelly Bean was specifically a performance release and certain performance improvements may have been obtained from tuning the kernel or supporting the latest chipsets. It's also possible that Samsung in moving to Jelly Bean benefited from better multi core support, or the manner in which power stepping was done, cores were idling or whatever. Of course that sort of thing would show up in a benchmark without specifically being done for the benefit of the benchmark.
We have an O/S tuned to optimally use the specific chipset, versus an o/s that is written for a broad spectrum of hardware, and the optimised version turns out to be faster.
My Lumia 800 appears to be very sprightly, inspite of the somewhat lackluster processor - it's heavily optimised.
iPhones /always/ appear snappy - Apple are good optimisers.
How does the benchmark differ between platforms? Surely it's not much use as a benchmark if it doesn't perform the same tests on both? I've never used it so I've no idea why it's not able to produce a comparable result.
That said, it doesn't look like a very reliable benchmark, half the Galaxy SIII results show the devices have 0KB in cache and the number of cores varies all over the place, with many showing 1 or 3 cores! There are also lots of overclocked SIIIs with scores the same as or lower than stock SIIIs which makes me wonder if people are somehow overclocking during the benchmark and then scaling back to stock when it reads the clock speed... (http://browser.primatelabs.com/geekbench2/chart?q=model%3A%22Samsung+Galaxy+S+III%22+platform%3A%22Android%22+architecture%3AARM+bits%3A32+)
I also question the validity of a benchmark that doesn't appear to test the GPU's performance. Given that both iOS and Android accelerate the UI using the GPU it would make sense to take GPU performance into account. Also, looking at the benchmark it tests things like image blur and image compress using the CPU, which are tasks more suited to using the GPU.
I'm sceptical as to the authenticity of this, until they are out in the wild I will assume their not real.
Also according to the article the S3 with JB is quite a bit faster, Samsung should push out JB and then advertise it just as the iPhone is made available..
Galaxy S3, faster than the iPhone :-D
One has to wonder how the members of the fruity cult have arrived to a figure of 1560 for the S3. Checking the shipping S3 devices benchmarks, most are way above that. Of course, it wouldn't look as good as saying that a single iYawn5 benchmark of an unreleased device was better than some S3 benchmarks, and much worse than many. It might dent the RDF...
"Compare that to the Nexus 7's score of 1591 and the Asus Transformer Prime TF201's 1497, and the iPhone 5 result, if genuine, shows the handset beating tablets."
Asus Transformer Infinity TF700 -
2041 http://browser.primatelabs.com/geekbench2/1036825 Asus Infinity
Are we trying to sell a point that doesn't exist by comparing it to old hardware? The S3 has scores ranging from 1500's to 2052.
Go over to XDA, and on most sub forums you can find users spending all day everyday whacking out benchmark and obsessing over the results, and how the latest update has them crying because they've lost a 100 points and now their phone is a pile of sh***.
Benchmarks are in essence an indicator of performance nothing more nothing less.
Does IO6 on iphone5 *feel* fast and smooth? Yes = WIN
Does Jelly bean on an S3 *feel* fast and smooth? Yes = WIN
Does win8 on 920 *feel* fast and smooth? Yes = WIN
Frankly, we should all be grateful the useability can now be taken for granted, regardless of our platform allegance.
Group hugs!
There's no such thing as a reliable benchmark; and GB is the worst of all: they do not describe at all their methodology, nor disclose oner line of source code.
It is a marketing tool to provide the fanboys the reassuring feeling that their overpriced gizmos are faster than no-name greyboxes.
It is merely (on comparable hardware) a measure of the strengh of a compiler on a given subset of instructions.
How can anyone trust or even talk abour it (not mentionning buying it) ?
Take something more real, as Cinebench, you'll get precisely the same number on the same hardware regardless of the OS platform.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
