This year although we’re not reporting from Hawaii, Qualcomm’s Tech Summit is still happening in digital form, representing the company’s most important launch event of the year as it showcases the new flagship products that will power next year’s smartphones. Qualcomm yesterday announced the new Snapdragon 888 SoC and platform, and today we’re going in-depth into the specifications and features of the new silicon design.

The Snapdragon 888 is a big leap for Qualcomm, so much so that they’ve veered off from their usual naming scheme increments this generation and even skipped the 87x series altogether. The 888 number is not there only for marketing purposes as it represents fortune and luck in Chinese, but the new SoC has some substantial generational changes that sets it apart from the usual yearly improvements of past.

Featuring the first ever implementation of a Cortex-X1 CPU core as its performance engine, new Cortex-A78 cores for efficiency, a massive +35% boost in GPU performance, a totally new DSP/NPU IP redesigned from the ground up, triple camera ISPs, integrated 5G modem, all manufactured on a new 5nm process node, the new Snapdragon 888 touches and updates almost every part of the SoC design with significant uplifts in performance and capabilities. There is a lot to cover, so let’s go over the details piece by piece:

Qualcomm Snapdragon Flagship SoCs 2020-2021
SoC Snapdragon 865

Snapdragon 888

CPU 1x Cortex-A77
@ 2.84GHz 1x512KB pL2

3x Cortex-A77
@ 2.42GHz 3x256KB pL2

4x Cortex-A55
@ 1.80GHz 4x128KB pL2

4MB sL3
1x Cortex-X1
@ 2.84GHz 1x1024KB pL2

3x Cortex-A78
@ 2.42GHz 3x512KB pL2

4x Cortex-A55
@ 1.80GHz 4x128KB pL2

4MB sL3
GPU Adreno 650 @ 587 MHz

 
Adreno 660 @ ?MHz

+35% perf
DSP / NPU Hexagon 698

15 TOPS AI
(Total CPU+GPU+HVX+Tensor)
Hexagon 780

26 TOPS AI
(Total CPU+GPU+HVX+Tensor)
Memory
Controller
4x 16-bit CH

@ 2133MHz LPDDR4X / 33.4GB/s
or
@ 2750MHz LPDDR5  /  44.0GB/s

3MB system level cache
4x 16-bit CH

@ 3200MHz LPDDR5  /  51.2GB/s

3MB system level cache
ISP/Camera Dual 14-bit Spectra 480 ISP

1x 200MP or 64MP with ZSL
or
2x 25MP with ZSL



4K video & 64MP burst capture
Triple 14-bit Spectra 580 ISP

1x 200MP or 84MP with ZSL
or
64+25MP with ZSL
or
3x 28MP with ZSL

4K video & 64MP burst capture
Encode/
Decode
8K30 / 4K120 10-bit H.265

Dolby Vision, HDR10+, HDR10, HLG

720p960 infinite recording
8K30 / 4K120 10-bit H.265

Dolby Vision, HDR10+, HDR10, HLG

720p960 infinite recording
Integrated Modem none
(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
X60 integrated


(LTE Category 24/22)
DL = 2500 Mbps
7x20MHz CA, 1024-QAM
UL = 316 Mbps
3x20MHz CA, 256-QAM

(5G NR Sub-6 + mmWave)
DL = 7500 Mbps
UL = 3000 Mbps
Mfc. Process TSMC
7nm (N7P)
Samsung
5nm (5LPE)

 

Re-integration of the 5G modem into the SoC

The most important aspect for this year’s design is the fact that Qualcomm is going back to an fully integrated modem design, contrasting last year’s surprising choice of the Snapdragon 865 not containing any modem at all and having instead to rely on the external X55 modem.

Last year’s rationale of going with an external modem was said to have been a practical one, stemming from the fact that 5G was still in its early stages and that many vendors had to make a lot of design efforts when designing their new handsets for 5G. A external 5G modem such as the X55 helped the 5G transition as it was available to vendors earlier than the Snapdragon 865 SoC itself, allowing them to design their RF systems before having access to the newest SoC.

This year, the market has evolved and is more mature, and Qualcomm chose to re-integrate the modem into the same silicon die as the SoC. The new X60 modem subsystem is the company’s 3rd generation 5G design and brings new capabilities in terms of carrier aggregation and 5G frequency band interoperability.

The platform’s reabsorption of the modem into the SoC die should signify better power efficiency, lower platform cost as well as lower PCB complexity for smartphone vendors.

2020 certainly was the year that 5G became a mainstream feature amongst device vendors, with essentially everybody adopting the new standard into their flagship and even mid-range devices. The new X60 modem will further mature the 5G experience by providing more flexibility to network operators in terms of frequency band support.

mmWave in particular has been a rather contentious aspect of 5G in 2020 as network deployments has been rather scarce and limited to US cities, with users reporting spotty reception with a larger impact on battery life. mmWave network expansion is progressing at a steady pace, and Qualcomm states that the new Snapdragon 888 platform completely solves the power efficiency concerns around mmWave usage. Hopefully 2021 will be the year where mmWave becomes a lot more useful and practical for users.

Whilst mmWave is expected to still be relatively niche for the vast majority of users, Sub-6GHz will be the workhorse of 5G, and here we’re seeing rapid expansion and deployments in countries all over the world. The new X60’s modem capability of allowing for carrier aggregation between FDD (Frequency Division Duplex, dedicated frequency bands between upload & download) and TDD (Time Division Duplex, upload & download in the same frequency band) means that network carriers will be able to mix and match more available Sub-6GHz spectrum together for even greater bandwidth.

DSS, or dynamic spectrum sharing, is also going to be a key technology enabling network operators to migrate existing LTE frequency bands to 5G NR dynamically based on the organic LTE/5G user demand – meaning that the frequency spectrum doesn’t need to be segregated for each technology, thus allowing more actual usable bandwidth for both types of users in the first few years and consumers switch over to 5G-capable handsets.

Manufactured on Samsung 5nm / 5LPE

The new Snapdragon 888 is making the transition from 7nm to 5nm, but the new design doesn’t merely make a process shift, it’s also making a foundry shift. After being with TSMC for the 7nm generations of the Snapdragon 855 and Snapdragon 865, Qualcomm is now switching back to Samsung Foundry and their new 5LPE process node for the new Snapdragon 888.

Qualcomm in recent years had been dual-sourcing from both TSMC and Samsung depending on the SoC design and product range, but in the high-end flagship SoC segment the company seems to have always chosen the technologically superior node as it had larger implications for the competitiveness of those parts. N7 and N7P were clear winning choices for the S855 and S865 as Samsung’s own 7LPP process was kind of late, and didn’t seem to be quite as good as TSMC’s variants. Qualcomm notably still used the 7LPP node on this year’s Snapdragon 765 SoC which has seen a lot of success in the premium range of device designs, however we had noted earlier in the year that it didn’t appear to be nearly as efficient as the TSMC-manufactured flagship SoC.

This year’s choice of switching back to a Samsung process for the flagship SoC seems to be a vote of confidence in the new process node- as otherwise Qualcomm likely wouldn’t have made the switch. Versus 7LPP, Samsung promises a 20% decrease in power consumption at the same performance, or a 10% increase in performance at the same power, together with a +-20% area reduction. How these figures will translate over to practical improvements for the new Snapdragon 888 remains to be seen.

Another rationale for the foundry switch could be manufacturing capacity. As Apple is eating up a lot of TSMC’s early 5nm capacity with the A14 and M1, Qualcomm probably saw Samsung’s 5LPE as the safer choice this year as the new Snapdragon 888 may be manufactured in the new dedicated EUV V1 line at Hwaesong.

It’ll be hard to gauge the process node switch for this generation as we don’t expect to see a similar design on TSMC’s 5nm node – unless MediaTek somehow has a new Cortex-X1 SoC in the pipeline for next year.

Powered by Cortex-X1 and Cortex-A78 CPUs

The Snapdragon 888 is the first publicly announced SoC powered by the new Cortex-X1 and Cortex-A78 CPU IPs by Arm. The Cortex-X1 in particular is the first of a new generation of CPU IP by Arm that focuses on maximising performance at the cost of lesser power efficiency, while the Cortex-A78 being the same-generation design but which still prioritises a balance between performance, power and area.

The new X1 core, based on Arm’s numbers, promised a +30% uplift in IPC over the last generation Cortex-A77 which was also deployed in the Snapdragon 865. Qualcomm advertises a 25% uplift over the Snapdragon 865, but that’s likely due to a few configuration differences on the part of the new Snapdragon 888 compared to Arm’s own internal figures.

The S888 continues to use a 1+3+4 CPU setup this generation, with the big difference being that instead of using the same CPU IP with a different physical implementation, the new 1+3 big cores are actually of different microarchitectures.

The “prime” performance core as Qualcomm likes to call it is the new Cortex-X1 design, clocking in at the same 2.84GHz as the Snapdragon 865’s prime core. The new core is configured with the maximum 1MB of L2 cache.

What stood out for me during our briefing of the new chip is that the clock frequency of the new design isn’t all very aggressive at all. Qualcomm’s 25% performance boost is in comparison to the vanilla Snapdragon 865 which also came at the same frequency. Compared to the Snapdragon 865+ which clocks in at 3.09GHz, this performance advantage should reduce to only 13%, which is less impressive.

Qualcomm’s 25% generational boost is also less than Arm’s advertised 30% as the new S888 continues to use a 4MB L3 cache for the CPU cluster, versus Arm’s envisioned 8MB configuration for a high-end 5nm SoC with the new X1 cores. Qualcomm explained to us that this was simply a balance between cost, implementation effort, and diminishing returns of a higher cache configuration design.

What this all means is that there’s a high chance that the Snapdragon 888 won’t be holding the Android CPU performance crown next year if Samsung’s next-gen Exynos SoC is even a little more aggressive in terms of clocks or cache configurations.

The high-performance X1 cores is joined by three Cortex-A78 cores clocking in at up to 2.4GHz, serving as the every-day workhorse CPUs for most computational tasks. In terms of cache, the new cores see their L2 doubled up from 256KB to 512KB.

One aspect I was interested in finding out is whether the new design still continues Qualcomm choice of fitting all the big cores together on a single voltage plane, which oddly enough, also seems to be the case for the new Snapdragon 888. This means that while the X1 and A78 cores can run at different frequencies, they’re all powered by the minimum voltage of either operating frequency at any one time. Qualcomm explains that this is again a practical choice surrounding the design complexity of the power delivery system, particularly mentioning that the X1 core can take advantage of the increased capacitance available from the larger shared power plane. Whilst this has worked well for the Snapdragon 855 and 865, I wonder that given the new X1 core’s increased performance and dynamic range, if the company isn’t leaving further performance or efficiency gains on the table for the sake of lower power delivery design cost. It’ll be interesting to see how other SoC vendors tackle their X1 implementations.

Finally, the big cores are again accompanied by four Cortex-A55 cores. This year the company yet again clocks them at 1.8GHz, which makes this the 4th generation SoC with an essentially identical configuration of little cores, which is a bit disappointing. Qualcomm can’t do much here as there’s simply a need for a new little core CPU IP, something which we’ll hopefully see released next year in 2021 for 2022 SoCs.

Hexagon 780: A Whole new IP for AI & DSP; Adreno 660 GPU
Comments Locked

123 Comments

View All Comments

  • jaj18 - Thursday, December 3, 2020 - link

    It will come with adreno 7××🤔
  • StormyParis - Wednesday, December 2, 2020 - link

    "This year although we’re not reporting from Hawaii". Heh heh. I'd feel sorry for you if I wasn't jealous for all the other years ? ;-p
  • Krysto - Wednesday, December 2, 2020 - link

    No AV1 decode support in 2021? Really?
  • tuxRoller - Thursday, December 3, 2020 - link

    I'm more interested in accelerated encode at this point.
    We've not had industry wide buy-in of a new lossy codec since jpeg, and hevc haven't quite achieved the ubiquity that h.264 managed after the same time in market.
  • GeoffreyA - Thursday, December 3, 2020 - link

    While hardware AV1 encode would be quite nice to see, there's a possibility it will lose much of software AV1's gains over software HEVC (that is, one might encode quickly but end up with less compression than x265). Also, leaving aside the Slough of Patents for a moment, VVC will have to be taken into account once x266 comes out. If the studies are right, the reference VVC encoder (not x266) already shows better compression and speed than AV1. Hopefully, it won't inherit HEVC's less than pleasing picture too (to my eyes at least).
  • tuxRoller - Friday, December 4, 2020 - link

    That's a great point. In my haste to mention the lack of encoding ability I'd forgotten about the actual implementation of such a complicated codec. Which of the 30 or so tools, and their combinations, provide the most bit savings per mm²?
    Iirc, vvc owes a lot of its gains via integration with ml (there's at least one commercial av1 implementation that does this as well to, supposedly, great effect). IOW, I'm uncertain how much easier vvc will be too implement in hardware. Otoh, EVC looks quite interesting.
  • GeoffreyA - Friday, December 4, 2020 - link

    Oh, yes, it will probably make their heads spin implementing this thing in hardware, and when they do, which they will, they're going to make it a marketing point (even if, in practice, it fell behind x265).

    Yesterday I was experimenting with libaom-av1 on FFmpeg and discovered a useful parameter: -cpu-used. Controls compression/encoding speed and takes values between 0 and 8. 0 being the slowest, 1 the default, and 8 fastest. To my surprise, 8 brought encoding speed to reasonable levels: about 10x slower than x265, if I remember right, which isn't half bad. I was using a video shrunk to 360p though.

    As for VVC, can't wait to give it a go. Hopefully, it'll deliver and be of AVC's calibre. I wasn't familiar with EVC but took a look at it now, and it does appear to be quite an interesting concept.
  • tuxRoller - Saturday, December 5, 2020 - link

    You might be interested in the doom9 forums (https://forum.doom9.org/forumdisplay.php?f=17). In the av1 thread you'll often see people posting updates about the various av1 en/decode implementations, new settings and, in general, some interesting thoughts from folks in the industry.
    BTW, starting from this post (https://forum.doom9.org/showthread.php?p=1929560#p... there's an interesting discussion regarding qcom & their interest in not pushing av1.
    Regarding fast encoders, I'm assuming you've tried svt-av1? That's supposed to have nearly caught up with aom's encoder quality but is still a good deal faster.
    Lastly, thanks for the paper. It looks interesting and a quick skim didn't reveal any mention of ml enhanced transform, or even a new entropy code(!); they seem to be continuing to iterate on h.264->h.265. However, only started reading it and realized I'm not getting through that tonight:)
  • GeoffreyA - Saturday, December 5, 2020 - link

    Thanks for those doom9 threads. Looks like a treasure trove of information on AV1 there. As for SVT-AV1, yes, I have tried it. While the speed was good, the picture didn't seem that impressive. Anyhow, I'll have a crack at it again and see how it stacks up against libaom, now that I've got the latter running faster.

    You're right. I remember getting the impression that this was similar to how HEVC improved over H.264. Mostly, extending techniques already laid down. Yet another reason to tip one's hat to the MP3 of video.
  • GeoffreyA - Saturday, December 5, 2020 - link

    I found this some weeks ago. It goes into some lower-level details of VVC.

    https://www.cambridge.org/core/services/aop-cambri...

Log in

Don't have an account? Sign up now