IPhone 4S: Anandtech review | Hacker News Reader

IPhone 4S: Anandtech review

anandtech.com

64 pointsincremental14 comments

I like seeing an in-depth review.

These days it feels like all the major sites rush out the reviews right before launch day with less than a day of actually testing being done. The reviews often sound like just a re-hash of the press release specs.

I, as I assume most on this site, only considers Anandtech the one true tech product reviewer on the web. The only other equal is John Siracusa's review of the latest Mac OS release.

Exactly. Which is why the only time I quote a review in a post is when it's one from Anandtech. It takes a solid hour and a little more for me to go through their reviews and it's always a pleasure because one comes out wiser at the end of it. In this review, I really enjoyed reading the A5 architecture and the optics section, it was interesting that they bothered to dig out where the optics might be sourced from (Genius Opt.).

Anandtech does awesome in-depth reviews but most of the stuff they are talking about is not relevant to the average consumer in the slightest. If other sites were all doing that many would not be relevant to their readers interests.

I agree that Anandtech serves a niche market, but I feel like most other review sites kind of serve the same niche, but more poorly. What I mean is that a gadget reviewer's experience is so far different from a normal consumer's as to be almost informationless. Gadget reviewers are always looking at new things, never have to use the old things for very long, and their use is probably not representative of most people's use. (For example, when reviewing an Android phone, why does it matter that the UI is different from another Android phone? This only matters for a gadget reviewer who has to switch between phones all the time. Most people will choose a phone and stick with it for at least 2 years. They'll learn to use that UI. They will rarely encounter other UIs.)

I think Brian Lam's Wirecutter[0] takes a step in the right direction. Gdgt[1], which has been around for a while, seems interesting, too, as it's user-generated content, so the experiences rendered are those of self-chosen peers.

P.S. For some reasons, I always read your username as the first word of the comment, so it always sounds in my head like you're saying, "Ugh, Anandtech does awesome..." In my head, you're quite an irritable fellow. There's someone else named "yawn" who always seems bored.

[0]: http://thewirecutter.com/ [1]: http://gdgt.com

Now that an 800mhz A5 gets better scores than 1.2Ghz chips... are we going back to the PPC vs Intel days?

No, it's just the javascript engine is different (which you would think would make it a pretty bad benchmark by AnandTech's usual standards).

Though the shift from A8 to A9 and single to dual core and the emphasis on GPU means the CPU speed isn't the key metric it might have once been.

Yep, hopefully these will be the days where we understand that comparing numbers such as processor speed when literally everything else in the stack, both hardware and software, is different is a completely pointless exercise.

(EDIT: I know that a JavaScript benchmark isn't an example of this, this is more a general comment on people comment on phone specs at the moment).

No, it doesn't. It's still mostly a software issue. As you can see a dual core 1.2 Ghz GS 2 gets 3300 ms, while a dual core 1.2 Ghz Droid RAZR gets 2000 ms. And that's just the stock browser. It can even get 1300 ms in Firefox.

The A5 still uses 2 Cortex A9 cores. Whatever optimizations Apple has done to it will have a minimal impact on performance. The biggest gains will still be obtained on the software side.

> "Whatever optimizations Apple has done to it will have a minimal impact on performance."

I agree in general, though I think Apple is going to continue adding custom DSPs to their package. And that will provide an advantage in those targeted areas that software alone will not be able to bridge.

The "Apple A5" SoC still uses ARM Cortex-A9 cores like nearly every other mobile SoC. Only one that doesn't is Qualcomm's Snapdragon, which uses their custom (but still based on ARMv7 like the Cortex chips) "Scorpion" design, soon to be succeeded by "Krait". They're still ARM though, the difference isn't near as drastic as PPC v. Intel; it's closer to Intel v. AMD.

As others have said, unless you're only changing one element in the hardware/software stack, it's not an apples-to-apples comparison. However, A5 is reported to use out-of-order execution, whereas A4 and earlier iPhone CPUs are purely in-order. Integer ALU pipeline length has reportedly also been reduced. The memory bus clock has also been doubled. All else being equal, a single A5 core should therefore beat the A4 on identical code. With a different JS engine on a different browser running on a different OS, I doubt these architectural effects will be noticeable.

If you want to talk about architecture, please use the name of the CPU core and not Apple's marketing name for the integrated SoC IC. The "A5" is not a CPU. The application CPUs on the chip are ARM Cortex A9's. These are the same cores that are available on the OMAP4, Exynos, Tegra 2, and a few others I can't remember off the top of my head.

It is indeed a SMP core (1-4 CPUs), which does out of order execution and register renaming. And it outperforms the Cortex-A8 (an in-order CPU) on a per-clock basis on many workloads.

But it's not an Apple part. Everyone is shipping these things now.

Overall a good article. Though it's got this whopper:

A DRAM package is then stacked on top of the SoC. Avoiding having to route high-speed DRAM lines on the PCB itself not only saves space but it further reduces memory latency.

Uh... come again? DRAM latency is a function of the analog circuit inside the chip, not the wires you connect to it. At best, you might be able to drive the chips at a higher transfer frequency (but even then, the limit is probably on-package in the DRAM, not due to board trace problems). But that has at best a minimal effect on latency (you're shrinking the handful of transfer cycles at the end of the read).

The advantage of the PoP configuration is precisely that it saves space -- quite a bit of space, and it's a great trick. But this bit is just way off.

Reply on news.ycombinator.com