Qualcomm Announces Snapdragon 865+: Breaking the 3GHz Thresholdby Andrei Frumusanu on July 8, 2020 9:30 AM EST
- Posted in
- Snapdragon 865
Today Qualcomm is announcing an update to its extremely successful Snapdragon 865 SoC: the new Snapdragon 865+. The Snapdragon 865 had already seen tremendous success with over 140 different design wins, powering some of the best Android smartphone devices this year. We’re past the hectic spring release cycle of devices, and much like last year with the S855+, for the summer and autumn release cycle, Qualcomm is providing vendors with the option for a higher-performance binned variant of the chip, the new S865+. As a bit of a arbitrary, but also important characteristic of the new chip is that this is the first ever mobile silicon to finally pass the 3GHz frequency mark.
|Qualcomm Snapdragon Flagship SoCs 2020|
|CPU||1x Cortex A77
@ 2.84GHz 1x512KB pL2
3x Cortex A77
@ 2.42GHz 3x256KB pL2
4x Cortex A55
@ 1.80GHz 4x128KB pL2
4MB sL3 @ ?MHz
|1x Cortex A77
@ 3.1GHz 1x512KB pL2
3x Cortex A77
@ 2.42GHz 3x256KB pL2
4x Cortex A55
@ 1.80GHz 4x128KB pL2
4MB sL3 @ ?MHz
|GPU||Adreno 650 @ 587 MHz||Adreno 650 @ ?
|DSP / NPU||Hexagon 698
15 TOPS AI
|4x 16-bit CH
@ 2133MHz LPDDR4X / 33.4GB/s
@ 2750MHz LPDDR5 / 44.0GB/s
3MB system level cache
|ISP/Camera||Dual 14-bit Spectra 480 ISP
64MP ZSL or 2x 25MP ZSL
4K video & 64MP burst capture
|8K30 / 4K120 10-bit H.265
Dolby Vision, HDR10+, HDR10, HLG
720p960 infinite recording
(Paired with external X55 only)
(LTE Category 24/22)
DL = 2500 Mbps
7x20MHz CA, 1024-QAM
UL = 316 Mbps
3x20MHz CA, 256-QAM
(5G NR Sub-6 + mmWave)
DL = 7000 Mbps
UL = 3000 Mbps
We’ve come to know the Snapdragon 865 quite well over the last few months, detailing the performance of the chipset in our initial benchmark articles as well as a more extensive deep-dive in our Galaxy S20 review. The new Snapdragon 865+ is a new binned variant of the same chipset with higher peak frequencies on the part of the “prime” CPU as well as the GPU, promising +10% performance on both aspects.
The First Mobile Silicon to Break Past 3GHz – 3.1GHz Even
Whilst in relative terms the new chipset’s +10% clock improvement isn’t all that earth-shattering, in absolute terms it finally allows the new Snapdragon 865+ to be the first mobile SoC to break past the 3GHz threshold, slightly exceeding that mark at a peak 3.1GHz frequency. Ever since the Cortex-A75 generation we’ve seen Arm make claims about their CPU microarchitectures achieving such high clock frequencies – however in all those years actual silicon products by vendors never really managed to quite get that close in commercial mass-production designs.
We’ve had a chat with Qualcomm’s SVP and GM of mobile business Alex Katouzian, about how Qualcomm achieved this, and fundamentally it’s a combination of aggressive physical design of the product as well as improving manufacturing yields during the product’s lifecycle. Katouzian explained that they would have been able to achieve these frequencies on the vanilla Snapdragon 865 – but they would have had a lower quantity of products being able to meet this mark due to manufacturing variations. Yield improvements during the lifecycle of the Snapdragon 865 means that the company is able to offer this higher frequency variant now.
For context, in the mobile world, usually SoC SKUs are binned not by performance (clock-frequency), but by power (voltage variations). This comes in contrast to the desktop and server world where one single silicon design is binned by different performance SKUs, varying in frequencies or even functional blocks. In a sense, Qualcomm’s 855+ and 865+ are SKUs that expand the product line in the way that usual PC silicon vendors do. Other mobile vendors such as MediaTek for example also take advantage of such product segmentation by releasing a single silicon design as multiple product SKUs.
As to what this means for the power and efficiency of the new Snapdragon 865+: There will be a power increase to reach the higher frequencies, however this will only be linear with the increased clock speed, meaning energy efficiency of the new SoC will maintain the same excellent levels of that of the Snapdragon 865, so battery life will not be affected.
More + Designs This Year
This mid-year refresh was only introduced last year with the Snapdragon 855+, and while we’ve seen some vendors opt for the upgrade in their latest device releases, uptake was rather limited, with only a few handful more prominent devices such as the ASUS ROG Phone II.
This year, Qualcomm tells us that we should be expecting more adoption for the refreshed silicon, with more design wins. Amongst the publicly announced platforms today is naturally the AUSS ROG Phone 3, with full details on the phone to follow in the next couple of weeks. Lenovo is also part of the launch partners, promising to bring to market a smartphone under the Lenovo Legion branding.
Amongst other new novelties of the Snapdragon 865+ platform is the ability for vendors to bundle with the new FastConnect 6900 Wi-Fi chips from Qualcomm, the company’s new Wi-Fi 6 chipsets with 6GHz band capability (Wi-Fi 6E).
We’re looking forward to devices with the new Snapdragon 865+ in the coming weeks and months.
- Qualcomm Announces Snapdragon 855 Plus: A Higher Bin SKU
- Qualcomm Announces Snapdragon 865 and 765(G): 5G For All in 2020, All The Details
- The Snapdragon 865 Performance Preview: Setting the Stage for Flagship Android 2020
- The Samsung Galaxy S20+, S20 Ultra Exynos & Snapdragon Review: Megalomania Devices
- The Snapdragon 855 Phone Roundup: Searching for the Best Implementations
Post Your CommentPlease log in or sign up to comment.
View All Comments
ikjadoon - Wednesday, July 8, 2020 - linkA fair point: we actually have a real-world example, i.e, Apple's battery & CPU throttling.
iPhones suffer a ~11% performance drop at low battery charge states, even on brand-new batteries, because the CPU asks for such a high voltage and/or current draw that the battery can't provide.
That is, ostensibly partially exacerbated by the CPU's high current draw, the battery degrades so much that it can't provide the power the CPU demands. But hopefully w/ just a single core demanding higher current in the 865+, it won't stress batteries as much.
GC2:CS - Wednesday, July 8, 2020 - linkNo matter which phone or battery you use you can discharge it to a level where you will see a “performance drop” called a shutdown.
Nobody is complaining about that.
ikjadoon - Wednesday, July 8, 2020 - linkThis has nothing to do with unexpected shutdowns: those are a much more severe symptom that has long affected any ho-hum $10 device.
These performance variations, however, are only exhibiting on high-performance, high-consumption ARM-based flagship CPUs. The results are clearly showing that batteries have held back SoC performance.
SoC improvements have easily outpaced the rather slow improvement in Li-on battery output.
deltaFx2 - Thursday, July 9, 2020 - link@ ikjadoon: That's part of it, yes. But in general, 1AH battery could provide 100mA for 10hrs but if you draw 1A, it may not last even 1hr. And yes, as you get to lower battery charge, it may be unable to provide sufficient voltage at the current draw required.
https://learn.adafruit.com/all-about-batteries/pow... (first link on a google search, there may be better references) Discharge rate adversely affects the energy delivered by the battery.
Then there's long-term battery life and the effect of current draw on that. A sealed degrading battery that costs >$100 to replace is a nice way of planned obsolescence.
s.yu - Thursday, July 9, 2020 - linkOne thing is definitely the cost, the batteries themselves cost a few bucks, if you know where to look. If you buy from service centers they'll rip you off for sure.
tuxRoller - Wednesday, July 8, 2020 - linkGood point.
If since this is linear, the increase should be, at most, 8-9%.
Where does does the 865 fall within the designed discharge rates of li-ion chemistry?
ikjadoon - Wednesday, July 8, 2020 - linkAh, I see now. Of course, in the end, the power draw must be reduced somewhere else to maintain the same energy usage (i.e., battery life). Here, it's not another part, but simply running the CPU for a shorter time.
That is, a race to idle is the other side here: use slightly more peak power and then sleep for much longer. 12 W for 5 seconds (60 W-s consumed) versus 10 W for 6 seconds (60 W-s).
Faster performance at higher power draw with equal energy usage / battery life.
Though not my initial confusion, deltaFx2's does note higher discharge rates bring an amplified loss of capacity: I can only think of Apple, but hopefully Android OEMs won't fall into the same trap.
Kangal - Thursday, July 9, 2020 - linkAlso, the efficiency of 865+ has increased over the 865.
If you try and run an 865 at 3.1GHz, it is possible, as long as the thermal limits aren't reached and if the energy required is supplied. However, whilst the early 865 chipset would have to slowly and exponentially increase the Voltage to sustain that frequency... whereas the 865+ seems to only require a linear increase in voltage.
This is the "lowest-hanging fruit" way of increasing performance, because you can simply hit higher frequencies more stably, thanks to having a more accurate silicon. This happens as yields of nodes improve during production. In the PC Gaming space they call this "binning". Regardless, this "linear increase" is better than the "exponential increase", so it is still an efficiency increase.
So you might be wondering if it's even possible for the standard 865 to overclock that high?
And to show you it is, an easy example is trying to overclock the Tegra X1 processor found in the Nintendo Switch... that's what we observe on the Nvidia Shield TV, which runs with a higher power-draw and a larger cooling solution.
iphonebestgamephone - Friday, July 10, 2020 - linkAnd your source is?
helloworld_chip - Friday, July 10, 2020 - linkIn order to have the power increase to be "linear with the increased clock speed", we need the cpu to be running at similar voltage range right? How is it possible to run 2.84GHz and 3.1GHz at similar voltage under the same process node and same design? Voltage has to be increased aggressively to get to 3.1GHz