iPhone 6 and Apple A8 SoC Performance Preview

Manufacturer: Apple

One Small Step

While most articles surrounding the iPhone 6 and iPhone 6 Plus this far have focused around user experience and larger screen sizes, performance, and in particular the effect of Apple's transition to the 20nm process node for the A8 SoC have been our main questions regarding these new phones. Naturally, I decided to put my personal iPhone 6 though our usual round of benchmarks.

View Full Size

First, let's start with 3DMark.

View Full Size

Comparing the 3DMark scores of the new Apple A8 to even the last generation A7 provides a smaller improvement than we are used to seeing generation-to-generation with Apple's custom ARM implementations. When you compare the A8 to something like the NVIDIA Tegra K1, which utilizes desktop-class GPU cores, the overall score blows Apple out of the water. Even taking a look at the CPU-bound physics score, the K1 is still a winner.

A 78% performance advantage in overall score when compared the A8 shows just how much of a powerhouse NVIDIA has with the K1. (Though clearly power envelopes are another matter entirely.)

View Full Size

If we look at more CPU benchmarks, like the browser-based Google Octane and SunSpider tests, the A8 starts to shine more.

View Full Size

While the A8 edges out the A7 to be the best performing device and 54% faster than the K1 in SunSpider, the A8 and K1 are neck and neck in the Google Octane benchmark.

View Full Size

Moving back to a graphics heavy benchmark, GFXBench's Manhattan test, the Tegra K1 has a 75% percent performance advantage over the A8 though it is 36% faster than the previous A7 silicon.

These early results are certainly a disappointment compared to the usual generation-to-generation performance increase we see with Apple SoCs.

However, the other aspect to look at is power efficiency. With normal use I have noticed a substantial increase in battery life of my iPhone 6 over the last generation iPhone 5S. While this may be due to a small (about 1 wH) increase in battery capacity, I think more can be credited to this being an overall more efficient device. Certain choices like sticking to a highly optimized Dual Core CPU design and Quad Core GPU, as well as a reduction in process node to 20nm all contribute to increased battery life, while surpassing the performance of the last generation Apple A7.

View Full Size

In that way, the A8 moves the bar forward for Apple and is a solid first attempt at using the 20nm silicon technology at TSMC. There is a strong potential that further refined parts (like the expected A8x for the iPad revisions) Apple will be able to further surpass 28nm silicon in performance and efficiency.


September 29, 2014 | 11:40 AM - Posted by Anonymous (not verified)

And why not include Snapdragon 805, 800? To see how A8 and K1 compare to them?

September 29, 2014 | 01:59 PM - Posted by pdjblum

ditto that

September 29, 2014 | 12:04 PM - Posted by Anonymous (not verified)

The Nvidia Denver is still on a 28nm process node, is the benchmarked K1 the custom Denver variant, or the reference A15 variant? The K1(whatever variant) is pretty impressive, but is it even a smartphone SKU, it appears to be in tablets, and maybe some phones will get it, but will the K1 variant in the phone, at first be clocked down, or maybe replaced wit a future K1 phone SKU on a smaller process node? Nvidia has the most impressive graphics, and until the actual PowerVr in the A8 is known and the argument is settled, the A8 has a respectable showing against the rest, other than the K1's take the game performance level in graphics. I want more K1 Denver tests, and is there going to be a K1 refresh at 20nm, or will Nvidia just wait for the M1(or what ever the K1's successor will be called). Back to the A8, what is the Browser used, and how does it compare to the Browser on the K1 system tested, both versions' Java engines, or does SunSpider have some form of standardized Java platform for testing built into the benchmark. The Apple A8 is probably fined tuned for power usage, to the point of loosing some performance, but the Denver K1, when it is completely tested, will probably be the one to see, when/if it gets made on a smaller process node, or is just replaced with the next generation, and Apple needs to get some resources into the PowerVR folks hands, to compete with the Nvidia graphics, especially on the Tablet form factor.

The Tegra brand is for a line of SOCs, the K1 being currently offered in 2 variants, it would help to always label the variant, to disambiguate what version of the K1.

September 29, 2014 | 12:08 PM - Posted by Anonymous (not verified)

It's the one inside the shield tablet which is the none denver variant.

September 29, 2014 | 12:17 PM - Posted by Anonymous (not verified)

Thanks, is the Denver expected in a Tablet device soon, I can not remember, if it had any design wins, or if Nvidia is going to refresh the shield this year with the Denver?

September 29, 2014 | 12:06 PM - Posted by Nikolaus (not verified)

one key element is missing in this test, the new Intel M core (5y70); this Intel part has only one huge problem-exorbitant price, that is going to weight heavily on the performance gap. How big is that gap? that is the question.

September 30, 2014 | 01:24 AM - Posted by renz (not verified)

is there actual product using intel new M core right now?

September 29, 2014 | 12:19 PM - Posted by derz

I guess Ryan doing a blind-folded tear down was out of the question? ;>_>

September 29, 2014 | 01:05 PM - Posted by Anonymous (not verified)

phone vs tablets.

September 29, 2014 | 04:22 PM - Posted by Anonymous (not verified)


September 29, 2014 | 06:15 PM - Posted by Anonymous (not verified)

Is benchmarking tablet SKUs against Phone SKUs the proper way to compare CPUs/SOCs, hardly. Is writing articles that include Custom microarchitecture SOCs, that have vastly more execution resources than the ARM holdings reference designs, but still only discussing them in terms of the ARM holdings Reference designs, correct, hardly. The same CPU/SOC in a Phone form factor, and a phone's thermal limits, is going to perform differently in a tablet form factor, with a more space for a larger heatsink, More RAM Memory, and most SOCs have thermal throttling sensors, and software/firmware controlled thermal throttling. So phone SKUs need to be benchmarked against Phone SKUs, short of having a standard testing Mule, and a CPU/SOC chip that can be socketed/wired to the testing platform[not possible with Phones/tablet SOCs]. There are quite a few powerful custom ARM ISA based SOCs(Apple A7/A8, Nvidia Denver, others) that have no business being described in terms of the many other slightly tweaked(100s of tweaked SOCs) designs based on ARM Holdings reference designs, as they are entirely different with respect to execution resources, the A7, and A8(?)can execute at least 6 IPC, and the Denver K1 at least 7 IPC. Phones and tablets are harder than gaming rigs, gaming rigs can at least be standardized to some extent, but phones and tablets, forget about ever trusting benchmarking at all, as there are to many undocumented, non standardized platforms, with each phone/tablet coming with a custom motherboard, and software tuned to that systems specs, and only that systems specs. Even the Java/dalvik/other VMs are customized to some degree, for the specific hardware. Testing individual phone/tablet platforms is even more inaccurate than the gaming rigs, when it comes to rating the CPU/SOC itself.

What is irrelevant is any claims or benchmarks as facts, because it is a losing proposition, believing in any Phone/tablet benchmark, with so many uncontrollable variables, in most phone and tablet platforms. As far as testing the devices' Overall performance of the completed device, a little less irrelevant, but still, it's the user having to purchase the device and running the user preferred applications, that will give the best results, keep track of the return policy, so the device can be returned, if it fails the final user benchmark, that only the user can do.

September 30, 2014 | 12:53 AM - Posted by DerekR (not verified)

It's actually very relevant in this case because Apple itself compared its new A8 to the K1 in iPhone 6 announcement presentation. Phones and tablets are different devices, but Apple opened itself up to this comparison when it presented its own data comparison.

September 30, 2014 | 09:14 AM - Posted by Anonymous (not verified)

both are arm base hardware both are capable and do run the same set up so yes it is irrelevant

September 30, 2014 | 09:38 PM - Posted by Anonymous (not verified)

The A7/A8, and the Nvidia K1, are different underneath in their custom wide order superscalar designs, they are completely custom SOCs that are designed to execute the ARMv8 ISA, their hardware resources are custom implementations, In the same manner that AMDs implementation of its x86 microarchitecture is different from Intel's x86, they just happen to be able to execute the x86, 16/32/64 bit ISA. The ARM reference designs that execute the ARMv8 ISA, are a narrower superscalar design, the Arm Holdings' reference designs that execute the ARMv8 ISA, are only 3 wide, the A7/A8, are 6 wide, the Nvidia K1(Denver variant) is 7+ wide, so the ARM holdings reference design ARMv8 ISA core can only execute 3 instructions per clock cycle(IPC), whereas the Apple A7 can execute 6 IPCs, the A8 6 IPCs(At least, ?), and the Nvidia Denver K1, 7+ IPCs.

Do not mistake a CPUs ISA, with the CPU's underlying implementation of the microarchitecture, designed to run the specific ISA, ISAs can be implemented in microcode/microarchitecture in many varying ways, some more efficient, and powerful than others.

September 30, 2014 | 08:11 AM - Posted by Anonymous (not verified)

someone is a novice at benchmarking. At least compare similar things. Honestly whats the point? next up desktops vs tablets THEN tanks vs cars crash tests!

if you want to compare things to see whats the better version of that thing they should at least be the same thing.

September 30, 2014 | 11:14 AM - Posted by Anonymous (not verified)

As far as testing the CPU/SOC unless the chip can be obtained and socketed/wired into a standardized motherboard on a standardized testing platform, the test results are not scientific! Phone/tablets platforms have too many independent variables that can not be isolated or properly brought into unity, to be considered scientifically testable. Were the government to require CPUs/SOCs to be subjected to the same style testing, that the automobile industry is subjected to, such as fuel economy/other testing, then there may be some sort of standardized platform developed for all CPUs/SOCs, but good luck with that. It's not a matter at being a Novice at benchmarking, the AnTuTu controversy proved that, the so called independent benchmarks can be gamed, and the lack of independent scientific testing methods rules out trusting any results. Never completely trust any review from any website that gets more than 50% of its revenues from the manufactures of the products, that the website is reviewing.

Proper testing platforms, testing mules, etc. can easily cost 6 figures or more to develop, and that's not including the programming costs to enable the testing platform to work with a specific CPU's/SOC's pinout, never mind the proprietary Op codes, that are hidden from all but the device's maker, to allow for such testing. Testing non standard motherboard(one off style) platforms in most smart phones, and tablets(less so) is very costly, and so full of hidden variables, that even the manufactures themselves are never completely sure, mix that with the marketing monkeys, of the marketing "profession", and things get beyond simple lies, damn lies, and statistics.

September 30, 2014 | 02:36 PM - Posted by biblicabeebli

Could someone knoledgeable comment on the effects of the new texture compression mode/format called ASTC. I know "it increases fill rates," but what does that mean in real world terms? Does that impact ui responsiveness on the 6+, or is it just relevant to gaming-style situations? Is this the thing Apple uses to claim a 50% graphics performance improvement?

its mentioned in the a8 analysis article from anantech:

and explained more thoroughly in another article from anandtech:

September 30, 2014 | 08:37 PM - Posted by Anonymous (not verified)

PowerVR Series6XT GPUs include PowerVR GX6250, GX6450 and GX6650.

September 30, 2014 | 08:41 PM - Posted by Anonymous (not verified)

I want to see the PowerVR wizard, in Apples new tablets!

September 30, 2014 | 11:22 PM - Posted by Anonymous (not verified)


This article from, anandtech:

And more in-depth iPhone review:

October 2, 2014 | 08:08 PM - Posted by Fasic (not verified)

Where is Snapdragon 800, 801, 805?! :D

October 25, 2014 | 01:55 PM - Posted by atintop

SOC depends on the OS so no additional comparable with SOC on another OS.
Renault provides motors for some Mercedes Class A, work it better on his own Megane? Is it comparable? Not

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

By submitting this form, you accept the Mollom privacy policy.