Apple A8 Benchmarked

[Eug]

I'm not sure why people on android aren't waiting until 64bit comes to Qualcomm or Samsung. 10% IPC still puts A8 in line with Broadwell, with density as good or better.

That sounds way too simplistic.

If apple wanted more power they could've clocked A8 at 1.8 or 2Ghz and called it a day. That would be core i3 speed.

I wonder how high powered that would be. My guess is that it would be a pretty big battery sucker.

 

[TreVader]

That sounds way too simplistic.

Fair enough, but certain parts of it are simple. Being able to address more than 4GB of memory at a time (moving any file bigger than 3.7GB or something in android requires the "push" command or some 3rd party software).


I thought when the 5s came out, that 64bit was a gimmick. It would appear that it wasn't as I can tell the difference between my 5s and my brothers 5 or my friends 5C even just browsing.



Edit: about the battery, Apple has a lot of room to go with thermals and battery size. They packed A7 into a 7.2mm package with only 1500mAH (half what the S5 has). I think the biggest issue with an A8 @1.8 or 2Ghz is heat. My 5s gets warm very quickly running certain stuff (google earth comes to mind). The larger 6 and 6+ allow apple even more room to breath with battery tho. I think they cited better all around and I thought they cited 12+ hrs hard use for the 6+

 

[Rakehellion]

I don't even know what that means. What is the OS limitation?

iOS limits each app to a certain amount of memory. No single app can use all of your device's memory even if it's the only thing running.

 

[Rakehellion]

Fair enough, but certain parts of it are simple. Being able to address more than 4GB of memory at a time (moving any file bigger than 3.7GB or something in android requires the "push" command or some 3rd party software).

You don't need a 64-bit CPU to copy files over 4GB.

I thought when the 5s came out, that 64bit was a gimmick. It would appear that it wasn't as I can tell the difference between my 5s and my brothers 5 or my friends 5C even just browsing.

That's because the processor is faster, not because it's 64-bit.

 

[TreVader]

You don't need a 64-bit CPU to copy files over 4GB.







That's because the processor is faster, not because it's 64-bit.

You are arguing semantics. You need 64 bit to run 64bit OS to address >4GB


The same limitation was in XP

 

[Khato]

With respect to transistor density one major flaw in simply dividing number of transistors by die size is that not all transistors in a design are the same. Specifically, there are basically two ways in which maximum frequency of a circuit can be increased - decrease the amount of work done per clock cycle or increase the drive current of the transistors. (We overclock by the latter method as increasing voltage increases drive current.) However the typical method for increasing drive current of the transistors is to size them appropriately for their fan-out and the intended operating frequency. The end result being that, typically, transistor width in circuits operating at 2 or 3 GHz are going to be much larger than those operating under 1 GHz. (Or with FinFETs, number of fins per transistor.) Edit: Guess I should also explicitly mention rather than assuming that the complexity of the circuit is the other major factor with respect to transistor sizing - more complex logic typically has higher fan-out which requires more drive current in order to operate at a given frequency.

This effect is far more pronounced when you're comparing an SoC like the A7 or A8 against one of Intel's Core based processors. On Core M for example, about half of the die space is going to be operating at either core or uncore frequencies while the remainder is graphics - so half the die is capable of operating in the 2 GHz range while the remainder is around 800 Mhz. By comparison the A7 only has around 16% of the total die devoted to CPU core (I'd assume that's the fastest logic on die at ~1.3 GHz), and then likely the next fastest logic is 18% of the die used for graphics... but that very well might already be operating at frequencies where minimum transistor size is adequate. Regardless, the remaining two thirds of the die are likely operating at frequencies where all logic can be using the minimum transistor size which is far more suited to high density.

That's the brief explanation of why BK's quote from last year regarding density is quite easy to believe: "If you take a look at things like transistor density and you compare, pardon the pun, apples-to-apples and you compare, say, the A7 to our Bay Trail, which is a high density 22 nanometer technology, then our transistor density is higher or more dense than the A7 is."

 

[jfpoole]

While there are plenty of flaws in javascript benchmarks I don't understand why Geekbench is any better - at least half of its integer section are typically handled by dedicated hardware (with AES and SHA1 hardware acceleration affecting the score for Apple's SoC, then JPEG and likely Sobel filter would be handled by imaging DSP in actual usage.)

Geekbench 3 only uses "dedicated hardware" for the AES and SHA-1 workloads. I wouldn't expect the other workloads (with the exception of SHA-2) to be accelerated by dedicated hardware in real-world scenarios. For example, I would expect JPEG compression and decompression to happen in software for most (if not all) applications. Are you aware of situations where this is not the case?

 

[Eug]

iOS limits each app to a certain amount of memory. No single app can use all of your device's memory even if it's the only thing running.

I am not a developer, so correct me if I'm wrong, but my understanding is that there is no hard ceiling to per-app memory allocation in iOS. Further to that, memory allocation per app increases with available memory. Sure, it won't use ALL the available memory, but that's a different argument.

This is illustrated by the fact that after a fresh reboot on my iPhone 5s, Safari can keep more tabs open without reloading. As you start more applications, the number of Safari tabs you can keep open like that diminishes. This is also illustrated by the fact that I see less of this behaviour (in albeit limited testing) on my wife's 32-bit iPhone 5. Why? Because the iPhone 5 effectively has more memory, when compared to A7 based devices, despite the fact that they all have 1 GB RAM.

This is discussed in Anand's iPad Air review:

http://www.anandtech.com/show/7460/apple-ipad-air-review/9

The move to a 64-bit platform however does complicate things a bit. Moving to a larger memory address space increases the size of pointers, which in turn can increase the footprint of 64-bit applications compared to their 32-bit counterparts. So although there’s clearly a performance uplift from app developers recompiling in 64-bit mode (more registers, access to new instructions), there’s also an associated memory footprint penalty. Since the iPad Air and iPhone 5s don’t feature a corresponding increase in memory capacity, I wondered if this might be a problem going forward.

To find out I monitored total platform memory usage in a couple of scenarios. Before measuring I always manually quit all open apps and performed a hard reset on the device. Note that the data below is reporting both clean and dirty memory, so it’s possible that some of the memory space could be recovered in the event that another process needed it. I hoped to minimize the impact by always working on a cleanly reset platform and only testing one app at a time.

I looked at memory usage under the following scenarios:

1) A clean boot with no additional apps open
2) Running Mobile Safari with 4 tabs open (two AnandTech.com tabs, two Apple.com tabs, all showing the same content)
3) Infinity Blade 3 (64-bit enabled) sitting at the very first scene once you start the game
4) iOS Maps in hybrid view with 3D mode enabled, with a WiFi assisted GPS lock on my physical location
5) Google Maps in the same view, under the same conditions. I threw in this one to have a 32-bit app reference point.

In general you’re looking at a 20 - 30% increase in memory footprint when dealing with an all 64-bit environment. At worst, the device’s total memory usage never exceeded 60% of what ships with the platform but these are admittedly fairly light use cases. With more apps open, including some doing work in the background, I do see relatively aggressive eviction of apps from memory. The most visible case is when Safari tabs have to be reloaded upon switching to them. Applications being evicted from memory don’t tend to be a huge problem since the A7 can reload them quickly.

The tricky part is you don’t really need all that much more memory. Unfortunately as with any dual-channel memory architecture, you’re fairly limited in how you can increase memory capacity and still get peak performance. Apple’s only move here would be to go to 2GB, which understandably comes with both power and financial costs. The former is a bigger concern for the iPhone 5s, but on the iPad Air I would’ve expected a transition sooner rather than later.

---

Summary: If the iPad Air 2 ships with 1 GB RAM, I will be very irritated. I'm less irritated with the iPhone 6/6+, but only because I have no desire to buy one.

 

[Techhog]

they know exactly what they're doing. The large screen will be worth it. Wait for the iPhone 7 if you want >1GB RAM

Oh please. You guys have been saying that for years. You're paying $650+ for $150 worth of hardware, plain and simple.

 

[Eug]

Oh please. You guys have been saying that for years. You're paying $650+ for $150 worth of hardware, plain and simple.

Well, you're also paying for the software and OS, which I much prefer. Also, I'm thinking the BOM will be around $225, or about 50% more than your $150.

 

[Techhog]

Well, you're also paying for the software and OS, which I much prefer. Also, I'm thinking the BOM will be around $225, or about 50% more than your $150.

That's not any better.

 

[cbn]

If apple wanted more power they could've clocked A8 at 1.8 or 2Ghz and called it a day. That would be core i3 speed.

I wonder how high powered that would be. My guess is that it would be a pretty big battery sucker.

Yes, especially if voltage had to increase by a good amount:

http://en.wikipedia.org/wiki/CPU_power_dissipation#Sources

The dynamic power consumed by a CPU is approximately proportional to the CPU frequency, and to the square of the CPU voltage

 

[Jovec]

I am not a developer, so correct me if I'm wrong, but my understanding is that there is no hard ceiling to per-app memory allocation in iOS.

John Carmack on IOS Rage in 2010. It could be different now, but IDK...

Managing over a gig of media made dealing with flash memory IO and process memory management very important, and I did a lot of performance investigations to figure things out.

Critically, almost all of the data is static, and can be freely discarded. iOS does not have a swapfile, so if you use too much dynamic memory, the OS gives you a warning or two, then kills your process. The bane of iOS developers is that “too much” is not defined, and in fact varies based on what other apps (Safari, Mail, iPod, etc) that are in memory have done. If you read all your game data into memory, the OS can’t do anything with it, and you are in danger. However, if all of your data is in a read-only memory mapped file, the OS can throw it out at will. This will cause a game hitch when you need it next, but it beats an abrupt termination. The low memory warning does still cause the frame rate to go to hell for a couple seconds as all the other apps try to discard things, even if the game doesn’t do much.

Interestingly, you can only memory map about 700 megs of virtual address space, which is a bit surprising for a 32 bit OS. I expected at least twice that, if not close to 3 gigs. We sometimes have a decent fraction of this mapped.

A page fault to a memory mapped file takes between 1.8 ms on an iPhone 4 and 2.2 ms on an iPod 2, and brings in 32k of data. There appears to be an optimization where if you fault at the very beginning of a file, it brings in 128k instead of 32k, which has implications for file headers.

 

[rtsurfer]

You are arguing semantics. You need 64 bit to run 64bit OS to address >4GB


The same limitation was in XP

Do you understand what address means there.?

You don't need a 64-bit OS to run an 8GB Blu-ray file.
Neither do you need 64 bit OS to copy a 4GB file, like you mentioned in the last case.(Android does it just fine).

You need 64 Bit to allow an application to allocate & use more than 4GB of RAM, aka heavy games, compression software, Vmware, etc.

Greater than 4GB files work just fine in XP as long as you had the right FileSystem, aka NTFS.
Running programs that consumed more than 4Gb of RAM was the issue.

 

[cbn]

The mobile CPU market has been a lot like the PC market in its early days -- big performance leaps enabled by better process tech, larger area budgets, and increasing power envelopes.

The "slowdown" in Intel's CPU performance increases only came about once Intel had to deal with trying to keep power consumption the same/drive it lower. Oh, and keep in mmd that die sizes are limited by the per-unit margin targets that these chip vendors aim for.

Further, not all of the thermal/transistor headroom could be spent on just the CPU in an increasingly integrated world. Everything from PC processors to server chips increasingly requires more sophisticated non-CPU blocks.

In short: welcome to the maturation of the smartphone/tablet markets

Maybe the next step for Apple will be to create another category of chip. This by separating CPU/GPU from PCH to increase performance and value.

Then with this higher performance and higher value combination they will enter markets beyond smartphone and tablet?

 

[Khato]

Geekbench 3 only uses "dedicated hardware" for the AES and SHA-1 workloads. I wouldn't expect the other workloads (with the exception of SHA-2) to be accelerated by dedicated hardware in real-world scenarios. For example, I would expect JPEG compression and decompression to happen in software for most (if not all) applications. Are you aware of situations where this is not the case?

It's quite possible that software hasn't caught up to the capabilities of software, especially on the Android side. I'll freely admit that the basis for my assessment was that I didn't see any reason why it wouldn't work and a quick search yielded this Anandtech article - http://www.anandtech.com/show/6747/htc-one-review/11 - which stated the below, with the bold text being my emphasis since it's what caught my eye on the matter in the first place:

The image compression/decompression tests are particularly useful as they just show JPEG compression/decompression performance, a very real world use case that's often seen in many applications (although hardware JPEG acceleration does limit its usefulness these days in mobile).

Could be that they're just referencing it with respect to actual camera capture, but it didn't seem like it.

 

[iiiankiii]

Do you understand what address means there.?

You don't need a 64-bit OS to run an 8GB Blu-ray file.
Neither do you need 64 bit OS to copy a 4GB file, like you mentioned in the last case.(Android does it just fine).

You need 64 Bit to allow an application to allocate & use more than 4GB of RAM, aka heavy games, compression software, Vmware, etc.

Greater than 4GB files work just fine in XP as long as you had the right FileSystem, aka NTFS.
Running programs that consumed more than 4Gb of RAM was the issue.

The funny part is this; it's a 64-bit OS with 1GB of ram. That's how you take advantage of a 64-bit OS

 

[Essence_of_War]

You are arguing semantics. You need 64 bit to run 64bit OS to address >4GB


The same limitation was in XP

It is not at all a semantic argument. It is explaining to you why you're incorrect to attribute the qualitative "faster" to 64-bit-ness, when there is a much more parsimonious explanation, a much faster CPU.

You seem to be confusing something like: the maximum addressable RAM space, the maximum size of a file in a filesystem, and the maximum bootable partition size. All of those maximum sizes derive from addressable space limits, but they don't all derive from the same limit, and they aren't all resovled by "64-bit!".

 

[TreVader]

Do you understand what address means there.?

You don't need a 64-bit OS to run an 8GB Blu-ray file.
Neither do you need 64 bit OS to copy a 4GB file, like you mentioned in the last case.(Android does it just fine).

You need 64 Bit to allow an application to allocate & use more than 4GB of RAM, aka heavy games, compression software, Vmware, etc.

Greater than 4GB files work just fine in XP as long as you had the right FileSystem, aka NTFS.
Running programs that consumed more than 4Gb of RAM was the issue.

A) android does in fact have that limitation. Any file over 4GB is a pain to deal with

B)Practically speaking, the result is the same (limited OS).

So I still think 64bit is important and it will give the 5s 6,6+ AND Tegra K1/Denver a big edge in the future.


If I were buying an android device, it would be K1

 

[TreVader]

It is not at all a semantic argument. It is explaining to you why you're incorrect to attribute the qualitative "faster" to 64-bit-ness, when there is a much more parsimonious explanation, a much faster CPU.



You seem to be confusing something like: the maximum addressable RAM space, the maximum size of a file in a filesystem, and the maximum bootable partition size. All of those maximum sizes derive from addressable space limits, but they don't all derive from the same limit, and they aren't all resovled by "64-bit!".

My point is apple backed their claims up and I don't think anybody would say ARMv8 is a "gimmick".


The technical points of your argument are irrelevant from a consumers point of view (I.e. Mine). It's all Greek to me man.

 

[rtsurfer]

A) android does in fact have that limitation. Any file over 4GB is a pain to deal with

B)Practically speaking, the result is the same (limited OS).

So I still think 64bit is important and it will give the 5s 6,6+ AND Tegra K1/Denver a big edge in the future.


If I were buying an android device, it would be K1

A) Source.? Worked fine on mine the few times I had to deal with it.

B) Practically speaking, Apple's approach to this is like selling a Ferrari with V4. The infrastructure is there (ala 64Bit OS) but without the power (4+ GB of RAM), it useless and can't go faster than the competition.

 

[MiddleOfTheRoad]

Well, you're also paying for the software and OS, which I much prefer. Also, I'm thinking the BOM will be around $225, or about 50% more than your $150.

Isn't that kinda of a bogus reason to overpay for a phone? Because people who buy Apple desktops get the OS for free now.

 

[mavere]

Isn't that kinda of a bogus reason to overpay for a phone?

People buy something because they prefer the experience of that product over the competition. Getting worked up over how much other folks benefit from your purchase is wallowing in bitterness.

 

[jfpoole]

It's quite possible that software hasn't caught up to the capabilities of software, especially on the Android side. I'll freely admit that the basis for my assessment was that I didn't see any reason why it wouldn't work and a quick search yielded this Anandtech article - http://www.anandtech.com/show/6747/htc-one-review/11 - which stated the below, with the bold text being my emphasis since it's what caught my eye on the matter in the first place:

Could be that they're just referencing it with respect to actual camera capture, but it didn't seem like it.

Thank you for the pointer!

I took a quick look through the Android 4.4 source code and found that the Android graphics library uses libjpeg to decode JPEG images. This is the same library Geekbench 3 uses for its JPEG workloads. Maybe the hardware JPEG acceleration referenced in the article just relates to camera capture?

 

[cbn]

B) Practically speaking, Apple's approach to this is like selling a Ferrari with V4. The infrastructure is there (ala 64Bit OS) but without the power (4+ GB of RAM), it useless and can't go faster than the competition.

Maybe we see the 64 bit fully utilized first in a Cyclone+ desktop/laptop APU + PCH first? This desktop/laptop class APU with higher drive current/higher leakage xtors than the SOC that is used in the phone and tablets.