New Snapdragons: Some Context & Contrast
It hasn’t been a good year for Qualcomm so far. Every device featuring the Snapdragon 810 has had a less-than-stellar fate, be it because of overheating complaints or actual decreased sales. HTC is in a particularly worrisome situation, and Sony’s Z4 is slipping down Japanese charts just over a week after its release. The G Flex 2, Mi Note Pro and Z3+ all saw complaints as well. Even MediaTek is seemingly getting an advantage over this.
But a particularly bad SoC, albeit damaging, is not necessarily a coffin for a manufacturer of this size. We must remember that Qualcomm has held a practical monopoly of top-end premium flagship SoCs for close to three years, and the Snapdragon 810 has a somewhat rational explanation behind its failure. To put it shortly, Qualcomm was not ready to put out a 64-bit processor so early, and their new custom cores were not ready for primetime in early 2015. Apple began the 64-bit trend and Samsung’s Exynos 5433 already introduced (unsupported) 64-bit chips last year with the Note 4. Qualcomm’s Krait architecture successor was not ready, but it is coming, and we are now beginning to hear its details.
Here we will take a look at what we know and put it in proper context. Keep in mind this is not meant to be speculative but rather informative. While there is a lot of room for speculation, doing so on the shaky ground of early leaks wouldn’t be wise, so as always take the leaks with skepticism.
Snapdragon 620 and Cortex-A72
This week we heard about two Snapdragon SoCs, the main one being the 820 which is rumored to be coming in late 2015 or around early 2016 (manufacturing is only beginning on this half of the year). First, let’s tackle the little we know about the Snapdragon 620, their upcoming medium-range offering. We first heard rumors about it featuring the latest in mobile ARM cores, the Cortex-A72 design. Now we know for sure that this is the case. There are many improvements over the previous A57 design: performance and power efficiency are a given, but there’s also a hefty increase in bandwidth and reduced area. More importantly, the A72 is designed to not be a bursty processor like most mobile SoCs, and thus to be able to sustain its peak (or high) performance for longer periods
The A72 cores have up to a 60% increase in instructions-per-clock over their A57 predecessors, and also a respectable 20% power reduction when compared to the A57 (on the same 28nm processes). At the target of 14nm FinFet processes, these could be very impressive cores, and it wouldn’t be odd to see many devices in 2016 featuring them, particularly mid-range. Why mid-range? For one, Qualcomm’s “Kryo” cores so far look to be on Qualcomm’s premium SoCs. Samsung’s “mongoose” custom core architecture is also likely to make it to their high-end SoC. When you combine this with Intel’s x86 chipsets, the premium offerings with this solution narrow down.
Which leads us to the benchmarks: Geekbench scores were leaked for the 620 and they show remarkable results. This is a CPU benchmark, and while it does not give us directly commensurable results with many other SoCs (due to the different architectures), the output is impressive nonetheless: single core scores put it at 1513, higher than the Exynos 7420’s (1486), while multi core scores remain at a respectable 4051 (below the Exynos’ 5284 and the 810’s 4424). The single-core results are truly impressive for a mid-range chip, but the multi core scores are not as well-scaled those of its competitors. From what we see so far, this looks to be an excellent chip that raises expectations, as it won’t even be Qualcomm’s best. Which in turn leads us to…
Snapdragon 820 and Kryo
Since Qualcomm shot down the rumors of a Snapdragon 815 processor, enthusiasts became increasingly worried about the future of flagship chipsets in 2015, and each new 810 phone only increased the hysteria. Because of this, the 820 has big expectations to live up to. The Snapdragon 820 is said to come with the new Kryo cores clocked at 3GHz. Reports also say that the chips will be manufactured at Samsung’s foundries — this is very important, because Samsung is currently at the top of fabrication processes/manufacturing, and with their recent billionaire investments into new facilities, this could seal a permanent transition that would benefit Qualcomm.
The 820 is allegedly readied up and undergoing testing, and shipping will start soon. Xiaomi’s leaked roadmaps spoke of an upcoming device with this chipset, and these and many other sources reinforce the idea that the Xiaomi Mi5 will be one of the first devices to feature the new chip. Qualcomm had not sealed a deal with Samsung for the S6 (and most likely not for the Note 5 either) and thus lost a big source of revenue, so having a locked deal with Xiaomi (who does not have many other options but is still a huge player) is a great way to start the redemption. But how does the chip perform? We obviously can’t answer with certainty just yet, but leaked benchmarks (again) tell an interesting story:
GeekBench scores of what is presumed to be the Snapdragon 820 also showed up, and this tested chip (MSM8996) looks to be better than most of the current top dogs in the smartphone business. Its single core score is 1732, ~17% higher than the Exynos’ listed 1486 score and ~41% higher than the Mi NOTE Pro’s 1227 (from the 810). Multi core scores, however, are only 4970, which is ~6% lower than the Exynos’ and ~12% higher than the 810’s. This is seemingly a welcome upgrade over the Snapdragon 810, but when contrasted with the output of the Exynos 7420, a current chip, things get a little worrying.
Context and Contrast
The fact of the matter is that the Exynos, for example, is running A57 cores clocked at 2.1 GHz, while the Kryo is allegedly running at 3 GHz. When you consider that the next flagship revision of Samsung’s SoC’s will probably incorporate A72 cores (unless their custom cores are ready), you see that they can easily have a very good base increase in performance. Assuming that Samsung manages to optimize their device’s hardware as much as they have been doing now (keep in mind Samsung is not using custom architecture yet, and that they rely on ARM designs), expecting a revision that can match the alleged 820’s ~17% advantage is single core scores sounds very reasonable. When it comes to multi core output, it must be noted that Samsung’s Exynos line has always been excellent at multi-threading, even without Homogeneous Multi-processing which could surprise us with a refined implementation. Other chip manufacturers can go down the A72 route as well.
If Qualcomm truly opts for a 3 GHz frequency in their cores, this is also something to keep in mind. Qualcomm is known to go for higher frequencies than other competitors, namely Samsung’s Exynos line of SoCs. For example, the Note 3, S5 and Note 4 all ran at the same 1.9 GHz maximum frequency, yet each saw iterative performance improvements. Much of the performance gains that the Note 4 saw over the Note 3, despite both featuring big.LITTLE octa core configurations and the same frequency, came from the design changes (we must also remember that the Note 4’s software did not support the introduction of its 64-bit nature either).
The Note 3 came with 28nm Cortex A15 cores with an ARMv7 instruction set, while the Note 4 had 20nm Cortex A57 cores with ARMv8 instructions. The S6 only increased the maximum frequency by 2.1GHz, but migrated to 14nm processes among other improvements, and that netted it an ~18% increase (over the Note 4) in single core Geekbench scores, and a ~21% increase in multi core scores. This goes to show that design changes can greatly enhance the output of a processor. In phones, performance does not scale too linearly with frequency (there are many bottlenecks and constraints as well), but even then, Samsung or other manufacturers could find ways to break off their conservative frequency boosts to match up to the 820.
We must also keep in mind that the stock A72 solution aims to bring persistent performance that is not too bursty, and while many Snapdragon devices in the past (like the 805) have had consistent frequencies with (relatively) little throttling, we know nothing about the Kryo. It’d be unfair to speculate on said cores by contrasting it with the Snapdragon 810, which in a way was out of Qualcomm’s comfort zone. Nevertheless, we must stay on the lookout for details about throttling and performance consistency, because this is easily one of the biggest possible strengths of its upcoming competition.
The Snapdragon 820 looks to be a sizable upgrade from the Snapdragon 810, and most importantly, we expect it to address the throttling constraints that the 810 had. There are big net gains in the limited data we have from these benchmarks, and considering that this is likely not running on final software nor hardware, it is entirely possible (and likely) that scores will get higher with time and optimization. This being said, when put into context, the Snapdragon 820 is not looking to be the most likely champion so far. The fact that the A72 architecture alone can produce such performance gains (as seen in Qualcomm’s very SD620 scores) with additional promises of sable performance should put pressure on Qualcomm to improve upon their design and bring out their best.
Qualcomm must be using their Kryo cores for a good reason, though, and even if early scores don’t reflect the chunky gains we all desire in our phones, Snapdragon SoCs always bring many other neat additions to the table beyond simple performance boosts over previous models. Take a look at some of the latest Snapdragon feature videos or their MWC 2015 press release to get an idea of the kind of new features and improved functionality current and upcoming offerings allow for. Now, keep in mind that other manufacturers are also speeding up their advances. Intel, for example, has put out a cheap yet respectable Atom processor in the ZenFone 2, and at MWC 2015 we heard a lot of good things about their x7 line of premium processors.
The battle for the best performance can yield great speculation… however, it’s always best to look at the context of the market not to raise our predictions but to ground our expectations. These early details are just that — early details. As such, they give us a limited glimpse at what will most surely be refined and optimized through software, clever implementations and components. Nevertheless, there is a lot to be excited about.