Chipworks Publishes Samsung Galaxy S6 Teardown

47 Comments

View All Comments

• rd_nest - Friday, April 3, 2015 - link

  Waiting for S6 full review and M9 final part. It's basically waste of time reading other reviews.

• lilmoe - Friday, April 3, 2015 - link

  I really hope we get a VERY comprehensive review for the GS6 since it has LOTS of new design wins. I don't mind waiting a week or two longer to get that. We've already seen DOZENS of online reviews and comparisons, but more educated/technical input would be rather appreciated. I'd love to read more about sustained performance of the 7420 and how it compares to the SD810 and the 5433 (ie: the extent of efficiency we're getting out of Samsung's 14nm process vs TSMC/Samsung 20nm). Also, I'd like to see Geekbench, for example, being run 10 times to have a better look at performance degradation when throttling occurs. There seems to be so much inconsistency about reported efficiency/battery life, which means that various use cases and different firmware updates have VERY different battery life expectancies. I bet money the new model/transceiver combo also has a hand in this, and since it's Samsung's first attempt on LTE, it would be nice if compared with Qualcomm's MDM.
  It would also be incredibly useful if Andrei can tinker a bit with the phone and set the resolution at 720p (pretty please) to see the real impact the resolution has on battery life if at all possible (concrete data would shut me up for sure).
  
  Lastly, CAMERA. If comparison shots are made, PLEASE include samples from a proper DSLR for indoor, outdoor, and low light stills and videos. Samsung's ISP tends to be highly affected by conventional (yellowish) lighting. When I take pictures with my Note4, they tend to be warm/yellowish indoors at night. However, I've recently installed warm white LEDs (~4000K) and the difference was dramatic in the temperature of the shots (much more natural). It was MUCH better than even 6000K florescent lights which also made the camera take yellowish shots, even so the LEDs were warmer.
  
  Sorry guys, but I'm really setting the bar high for your review :P Oh yea, please let go of Chrome, or better, let go of browser benchmarks all together :D

• hung2900 - Friday, April 3, 2015 - link

  I really want to see how Samsung improves its implementation of Cortex A5x duo big.LITTLE compared to fhe Exynos 5433, and I also want to see how badly HTC cheated to reduce heat and increasd battery life, leading the M9 is even slower than the M8

• Samus - Friday, April 3, 2015 - link

  It's interesting how very different Samsung's SoC approach is than Apple's (who makes gigantic dies with only 2-3 cores) and here is Samsung making a tiny die with 4-8 cores. I guess the Android ecosystem is heavily optimized for quad-core, and big.LITTLE, so it makes sense to stick with it, but there is no denying Apple is the performance leader in ultramobile ARM SoC's in "per-core" power, so maybe they are doing something right by making a smaller number of powerful cores?

• Sunrise089 - Friday, April 3, 2015 - link

  Certainly not exclusively Samsung's fault, but the Android space is heavily optimized for many-core MARKETING. There's been enough incidences of inferior more-core editions of phones, especially internationally, to make it pretty clear what's driving the core counts.

• lilmoe - Saturday, April 4, 2015 - link

  The thing is, Apple's Cyclone lead over standard Cortex A57 (with similar voltage/tdp) isn't any more than 5-15% ATM. Android can make use of all the cores it can get, while iOS doesn't have much use for too many cores. It's a difference of OS utilization more so than individual apps. Apple's cores have more thermal headroom/boost since there isn't as many, and that's why you're getting faster single threaded performance. It's more about compromises than it is about the underlying tech.
  
  That said, I wouldn't call it "Apple's lead", far from it actually. Most Core i3s and i5s have better single thread performance than Core i7s (same class/TDP). But we all know which series is faster. So you can't just say that the A8 is faster than Exynos, it's definitely the other way around, and been so for quite a while.

• Morawka - Saturday, April 4, 2015 - link

  apple's SOC platform lead is genuine. Big ol cache keeps the whole system agile. 4MB of Unified Cache keeps the whole system agile. GPU or CPU, it all has access.
  
  Apple's SOC designs differ from a57/53 because apple has optimized the layout of the SOC. Using as little wire trace as possible to the most important components. All SOC components are placed for a very specific reason instead of a57/53 which uses auto layout with some tweaks. This is fundamentally why apple soc's are better than everyone else. They control the hardware and the software, and no android OEM is going to have that much control anymore.. Google quit making phones soo..
  
  iOS can use the extra cores if they wanted to add them. See ipad air 2, where pixolmator utilizes the 3rd core to it's full extent, drasticly decreasing render times when compared to the original iPad Air. The operating system is probably using it as well for background task, but nobody will know for sure unless you work for apple.

• lilmoe - Saturday, April 4, 2015 - link

  Apple can "afford" to make their own design. They have the volume needed for good pricing on their orders, and they don't have to worry about a fab that needs to keep production going to turn a profit or at least break even. Samsung, among others, can absolutely make their own proprietary/vertically-integrated designs (think Hummingbird), but for some reason, Samsung LSI prefers to make a more generic design that can be sold separately. This was a huge point of criticism for Samsung by many (myself included). There's absolutely no reason why Android can't be optimized for a more vertically integrated approach, but there are more reasons for why Samsung opted for this, the biggest one being Google themselves. Other reasons include built-in LTE modems, and price. Samsung Mobile, LSI, and Electronics have proven to be immature in component collaboration (the clash of the bureaucracy!). Their big boss stepped in last year and "ordered" them to get on it.
  
  "This is fundamentally why apple soc's are better than everyone else"
  
  "Better" is a bit ambiguous. "Better for the job at hand" would sound more realistic. And yes, I believe the A8 is better than Exynos, for example, in running iOS. One of many good reasons that's so is because iOS doesn't handle multi-tasking the same way Android does.
  
  Apple never shares info about their architecture. Most of what you said is guesswork. A more probable difference is that Cyclone (similar to Krait) is designed with a higher clock "range" and various efficiency/performance *levels* in mind. Again, Apple can afford relatively larger dies to hit more performance at lower clock speeds. A57's are mainly designed for higher performance/clocks and not much so optimized for efficiency. That's because big.LITTLE was designed to split the gap in performance/efficiency among different core architectures, thus a combination of A53/A57 (A7/A15 previously). Supposedly, splitting this performance/efficiency variation is "easier" (at the core level) and yields better efficiency and higher performance since the design (of the cores) doesn't need to be too complex (logically, optimizing for one thing should be easier). BUT, that theory isn't as easy in practice, at least not initially (think Exynos 5410). OEMs, like Samsung and NVidia are moving away from it because it adds MORE _collective_ complexity at the arrangement/design levels of the hardware, and more so on the software level. It's arguable that it's cheaper, and more efficient to get rid of all that complexity altogether, spend on designing a core architecture every 2-3 years, reuse that architecture, and utilized the saved space/complexity on improving/adding more features. SRAM is one way for that, and I believe Apple did the right thing in that particular regard.

• name99 - Saturday, April 4, 2015 - link

  The fundamental reasonS Apple's core is superior include
  (definite)
  - willingness to spend a LOT more transistors than the ARM cores
  - a more aggressive micro-architecture (6 wide, very large re-order and memory buffers)
  (probably)
  - more expensive (in transistors) branch prediction and memory prefetch
  (perhaps)
  - more sophisticated power control (eg automatically tracking whether code is memory limited, and if so shutting down one of the two compute clusters)

• danbob999 - Sunday, April 5, 2015 - link

  There is so much BS in this comment. As if A57 designs weren't optimized.