Today Qualcomm is announcing the successor to last year’s quite successful Snapdragon 765 line-up, a “premium” tier that the company had debuted, featuring the same higher-end features as on the flagship Snapdragon 800 series, albeit at lower performances.

The new Snapdragon 780G follows up on its predecessor with some large upgrades in terms of performance and multimedia capabilities, doubling up the number of large cores – increasing GPU performance by significant amounts, and featuring the new more performant fused AI engine with the new Hexagon 770 DSP. Furthermore, camera capture abilities have also seen great improvements with the new Spectra 570 triple-ISP.

Qualcomm Snapdragon Premium SoCs
SoC Snapdragon 765
Snapdragon 765G
Snapdragon 768G

Snapdragon 780G

CPU 1x Cortex-A76
@ 2.3GHz (non-G)
@ 2.4GHz (765G)

1x Cortex-A76
@ 2.2GHz

6x Cortex-A55
@ 1.8GHz
1x Cortex-A76
@ 2.8GHz

1x Cortex-A76
@ 2.4GHz

6x Cortex-A55
@ 1.8GHz
1x Cortex-A78
@ 2.4GHz

3x Cortex-A78
@ 2.2GHz

4x Cortex-A55
@ 1.9GHz
GPU Adreno 620 

 
Adreno 620

+15% perf over 765G
Adreno 642

+50% perf over 768G
DSP / NPU Hexagon 696
HVX + Tensor

5.4TOPS AI
(Total CPU+GPU+HVX+Tensor)
Hexagon 770
Scalar+Tensor+Vector

12TOPs AI
(Total CPU+GPU+DSP)
Memory
Controller
2x 16-bit CH

@ 2133MHz LPDDR4X / 17.0GB/s
ISP/Camera Dual 14-bit Spectra 355 ISP

1x 192MP 
or
1x 36MP ZSL
or
2x 22MP with ZSL
Triple 14-bit Spectra 570 ISP

1x 192MP
or
1x 84MP ZSL
or
2x 64+20MP ZSL
or
3x 25MP ZSL
Encode/
Decode
2160p30, 1080p120
H.264 & H.265

10-bit HDR pipelines
Integrated Modem Snapdragon X52
Integrated

(LTE Category 24/22)
DL = 1200 Mbps
4x20MHz CA, 256-QAM
UL = 210 Mbps
2x20MHz CA, 256-QAM

(5G NR Sub-6 4x4 100MHz
+ mmWave 2x2 400MHz)
DL = 3700 Mbps
UL = 1600 Mbps
Snapdragon X53 Integrated

(LTE Category 24/22)
DL = 1200 Mbps
4x20MHz CA, 256-QAM
UL = 210 Mbps
2x20MHz CA, 256-QAM

(5G NR Sub-6 4x4 100MHz)
DL = 3300 Mbps
UL = ? Mbps
Mfc. Process Samsung
7nm (7LPP)
Samsung
5nm (5LPE)

At heart, the new Snapdragon 780G is a very different SoC to its predecessor as it changes up the CPU configuration quite substantially. We’re moving from a 1+1+6 config, to a newer 1+3+4 setup, including a prime Cortex-A78 core at 2.4GHz, three Cortex-A78 cores at 2.2GHz, and four Cortex-A55 cores at 1.9GHz. Qualcomm promises CPU uplifts of up to 40% - the doubling of the large cores as well as the new microarchitecture employed should indeed offer a good boost in everyday user experience.

On the GPU side, we’re seeing the use of a new Adreno 642. As usual Qualcomm doesn’t disclose much details on the design here, but they disclose a generational performance uplift of up to +50% over the Snapdragon 768G, meaning over the 765G that should grow to +72%. Based on our past benchmarks, this should end up with similar performance as the Adreno 640 of the Snapdragon 855 flagship from a few years ago – meaning the GPU is seemingly aptly named in terms of its performance.

Qualcomm is employing its newest fused scalar+tensor+vector DSP and AI engine in the new Snapdragon 780G, meaning it should be equal in terms of its architectural design as the new unit on the Snapdragon 888, albeit at lower performance levels. Qualcomm advertises 12TOPs of AI performance across all the IP blocks of the SoC, which is over 2x over that of the predecessor.

In terms of DRAM, the SoC remains a 2x16b LPDDR4X-2133 design, which seems to be crucial for cost reduction in this market segment.

A very large upgrade in capabilities is found on the part of the camera ISPs. Again, much like the DSP, the new design follows up with the similar new IP architecture as employed in the Snapdragon 888, employing a new triple Spectra 570 block that is capable of operating three RGB camera sensors concurrently. 192MP captures are possible for single modules (with shutter lag), or in terms of zero shutter lag operation we can see either 1x 84MP, 64+20MP or 3x 25MP sensor configurations. In terms of video encoding, we don’t see mention of much changes compared to the predecessor so we assume that video capture abilities remain the same.

What’s very interesting of the new design and probably telling of the wider market at large, is the fact that the new part no longer advertises mmWave capability on the part of its modem. The new X53 modem has seemingly chopped off this feature from its spec sheet. Generally, mmWave remains an extremely niche feature that’s currently only widely deployed in select US cities globally. Given that the SoCs target devices at lower price points, and we’ve seen some extremely cheap Snapdragon 765 phones in the past year, mmWave capabilities were probably contradictory to the market segment these phones were targeting – vendors always have the possibility to use higher-end solutions such as the Snapdragon 870 if they want to include mmWave connectivity.

Finally, the new SoC is manufactured on Samsung’s 5LPE process node, which is an upgrade over the 7LPP node of last year’s Snapdragon 765. While the node doesn’t seem to be as promising when compared to TSMC’s 5nm node, it being employed in a SoC in this price category is definitely a positive and should show notable gains against its predecessor.

Qualcomm plans to bundle the Snapdragon 780G SoC with the FastConnect 6900 Wi-Fi chips which feature Wi-Fi 6E connectivity, hopefully signalling a wider spread of adopting of the new 6GHz spectrum technology.

The Snapdragon 780G is expected to see deployment in commercial devices in the second quarter of 2021.

Related Reading:

POST A COMMENT

38 Comments

View All Comments

  • eastcoast_pete - Thursday, March 25, 2021 - link

    That 780 looks a lot more attractive than its big brother/sister, the 888. The one fly in the ointment is the restriction of DRAM choice (no LPDDR5), but that doesn't affect my everyday use that much; the added speed and lower power draw would have been nice, though. The apparent omission of mmWave in the 5G modem is no issue even stateside IMO; as pointed out, it's really only available in large urban centers, and even there it only really works in "line of sight" situations. Verizon could get it to work at full speed in large stadiums, but not many other places. I am due for a new phone, so waiting for Summer makes even more sense now.
    @Andrei: One thing I do wonder about is the power draw by the integrated 5G modem in the 888; did you have a chance to test that with the cellular modem on and off? 5G is reported to be a power hog; any truth to that?
    Reply
  • eastcoast_pete - Thursday, March 25, 2021 - link

    And, had to get this standing complaint out of my system: why are stock ARM designs for the LITTLE cores (currently A55) still stuck on in-order execution after several years? Apple's small cores also have much better perf/W because they have had out-of-order execution for several generations by now. ARM seems to be all focused on big cores for servers and laptops these days (X1, Neoverse..), and the small cores haven't gotten much attention in recent years. However, those do matter a lot for smartphones; being able to stay on the LITTLE cores makes your battery last longer! Reply
  • jeremyshaw - Thursday, March 25, 2021 - link

    Apple small cores aren't that small vs A55. In the end, everything is a tradeoff.

    That being said, the A55 is somewhat old at this point. ARM has 3 main design teams, so it's likely one of them is working on it right now.
    Reply
  • eastcoast_pete - Thursday, March 25, 2021 - link

    How much larger (transistor numbers) are the small cores from Apple vs. the A55? I might be off, but AFAIK they're not the 2-3 times larger that the performance differences suggest. Also, I read/heard that "ARM is working on new LITTLE cores" about 18 months ago. As you wrote, the A55 is getting a bit long in the tooth, and a successor is overdue. I guess we see the downside of the ARM monopoly for SoCs for Android phones here; there really isn't an alternative small core design out there right now. Reply
  • KarlKastor - Monday, March 29, 2021 - link

    I compared an A13 Die shot with a Kirin 990 Die shot. The small cluster of the A13 is 4.41 mm² (Anandtech). The A55 Cluster is 1.48 mm². So it is 3 times the size.
    And that is pretty obvious. The small cluster has already a huge 4 MB L2 Cache.
    The size is closer to the A76 Cluster (5.4 mm²) in the Kirin (without L3).
    But it is still impressive. The IPC is already A76 level. Total performance is 4x that of A55.

    So the performance per mm² is not that impressive, though very good.
    Most impressive is the efficiency. And that is something ARM has to catch up. The big cores getting more powerful but less efficient. So they need better small cores. The Performance gap is too huge. Apple's Icestorm have 1/3 of the Performance of the firestorm. The A55 have 1/6 of the middle cores.
    Reply
  • TheinsanegamerN - Monday, March 29, 2021 - link

    smaller in order cores are more efficient. They onyl handle basic instruction after all. Reply
  • DanNeely - Thursday, March 25, 2021 - link

    Does "triple" for the ISP refer to the "3x 25MP ZSL", or does it mean that the 780G can simultaneously run 3 cameras each capable of any of the listed resolutions/modes of operation? Reply
  • Wardrive86 - Thursday, March 25, 2021 - link

    Did they ditch the System cache? Reply
  • Stochastic - Thursday, March 25, 2021 - link

    I’d like to see this in the Pixel 5a. Reply
  • Bob Todd - Thursday, March 25, 2021 - link

    Indeed. We have finally hit a point where the midrange SOCs are getting quite powerful and exciting. The jump to quad A78 is awesome. Reply

Log in

Don't have an account? Sign up now