Search

Qualcomm Snapdragon 8 Gen 3 Brings Generative AI to Smartphones

Qualcomm Snapdragon 8 Gen 3 Brings Generative AI to Smartphones

The News: Qualcomm announced its latest premium mobile platform, the Snapdragon 8 Gen 3, at its annual Snapdragon Summit in Hawaii this week. Snapdragon 8 Gen 3 represents a pivot from previous generations of Qualcomm’s flagship mobile SOC in that it is the first mobile platform designed with generative AI in mind. Qualcomm Technologies’ latest processor is expected to be adopted for flagship devices by global OEMs and smartphone brands such as ASUS, Honor, iQOO, MEIZU, NIO, Nubia, OnePlus, OPPO, realme, Redmi, RedMagic, Sony, vivo, Xiaomi, and ZTE. Read more about Qualcomm’s Snapdragon 8 Gen 3 announcements on the Snapdragon Summit’s event hub.

Qualcomm Snapdragon 8 Gen 3 Brings Generative AI to Smartphones

Analyst Take: “Each year,” Chris Patrick, Qualcomm’s senior VP and GM of mobile handsets explained, “we set out to design leading features and technologies that will power our latest Snapdragon 8-series mobile platform and the next generation of flagship Android devices.”

In the past, those leading features have focused on 5G and mmWave connectivity, camera, security, processing power, gameplay, and even real-time translation. This year, the theme is on-device AI. Specifically, on-device generative AI. The new Snapdragon 8 Gen 3 “unlocks a new era of generative AI,” continued Patrick, “enabling users to generate unique content, help with productivity and other breakthrough use cases.”

Indeed, the Snapdragon 8 Gen 3 infuses high-performance AI across the entire system to deliver premium-level performance and entirely new mobile experiences to consumers.

Some key specs:

  • Qualcomm Kryo CPU (64-bit architecture with 1 prime core, up to 3.3 GHz, Arm Cortex-X4 technology + 5 performance cores, up to 3.2 GHz + 2 efficiency cores, up to 2.3 GHz)
  • Visual Subsystem/Adreno GPU (real-time hardware-accelerated ray tracing with Global Illumination, support for Unreal Engine 5 Lumen Global Illumination and Reflections System, Snapdragon Game Super Resolution, Adreno Frame Motion Engine 2.0, Snapdragon Game Post Processing Accelerator, HDR gaming with 10-bit color depth, and Snapdragon Shadow Denoiser
  • API support for OpenGL ES 3.2, OpenCL 2.0 FP, Vulkan 1.3
  • Hardware-accelerated H.265, VP9, AV1 decoder
  • HDR Playback Codec support for HDR10+, HDR10, HLG, and Dolby Vision
  • Memory support for LP-DDR5x memory, up to 4800 MHz, and memory density up to 24 GB
  • For power users and frequent travelers, charging should also be fast and efficient with Qualcomm Quick Charge 5 Technology

No surprises on the security side either: Trust Management Engine (Root of Trust) with platform-level security foundations, support for Android’s DICE-based remote key provisioning, Qualcomm Trusted Execution Environment & Services (TEE) for use cases requiring higher processing assurance, Qualcomm Type-1 Hypervisor (for isolation from the high-level OS secure processing unit [SPU] with support for Strongbox SW components, Qualcomm wireless edge services (WES) for secure attestation and provisioning, and Qualcomm 3D Sonic Sensor and Qualcomm 3D Sonic Max (fingerprint sensor).

Overall, these SOC specs are the best in the Android ecosystem, which is what I have come to expect from Qualcomm’s flagship Snapdragon SOCs. Qualcomm remains reliably consistent year after year on this front.

Snapdragon 8 Gen 3’s New AI Engine

Qualcomm’s new AI Engine is the secret sauce here. It was developed with on-device generative AI in mind, and supports multimodal generative AI models, including popular large language models (LLMs), language vision models (LVMs), and transformer network-based automatic speech recognition (ASR) up to 10 billion parameters entirely on-device. It is powered by an Adreno GPU, Kryo CPU, Hexagon neural processing unit (NPU), fused AI accelerator architecture, Hexagon scalar, vector, and tensor accelerators, and features Hexagon Direct Link, upgraded Micro Tile Inferencing, and an upgraded power delivery system. Qualcomm reports that its Hexagon NPU is 98% faster and 40% more efficient than its predecessor. The AI Engine also provides support for mix precision (INT8+INT16) and all precisions (INT4, INT8, INT16, FP16). Qualcomm’s improved Sensing Hub, which Qualcomm reports is 3.5x faster than its predecessor, features dual micro NPUs for audio and sensors, dual Always-Sensing ISPs to support two concurrent Always-Sensing Cameras, and support for the platform’s INT4 precision 5G Modem-RF System.

Thanks to its AI engine, Snapdragon 8 Gen 3 can run LLMs, process up to 20 tokens per second, and deliver the world’s fastest stable diffusion, allowing users to create AI-generated or enhanced content almost instantaneously right on the device.

Beyond how technically impressive this is, these on-device capabilities unlock entirely new horizons of performance and use cases. For starters, the ability to run LLMs directly on mobile devices helps address some of generative AI’s scale and cost issues we were concerned about at the start of the year. Data centers could not handle all of the training load coming down the demand pipeline (nor was it a cost-effective approach anyway). Having the ability to train LLMs on mobile devices helps spread that load and bring both balance and scale to the ecosystem. Second, giving users the ability to keep their prompts and data securely on their devices instead of pushing them to the cloud offers significant data security and privacy advantages. (More on Snapdragon 8 Gen 3’s security features in a moment.) Third, the ability to run extremely large on-device AI processes regardless of connectivity means little to no loss of productivity in situations where connectivity and bandwidth might be limited or unavailable. These three benefits together are not merely massive competitive advantages for Qualcomm’s flagship mobile SOC, they also represent a significant turning point for generative AI solutions in terms of greasing the proverbial skids of adoption, scale, and ubiquitous access.

Snapdragon 8 Gen 3’s New Bag of Premium Camera Tricks

Camera performance is still top of mind for premium mobile device features, and here too, AI brings significant performance upgrades. Qualcomm’s Intelligent Capture leverages the Snapdragon 8 Gen 3’s Cognitive ISP to allow users to prompt their phone to capture and edit photos and videos using only their voice. The feature can also optimize elements of the capture in real-time and up to 12 layers deep for more detail, vibrancy, and authenticity with Semantic Segmentation. Surprisingly, this feature is available on front and back cameras, which I thought was a very nice touch. Night Vision also comes to video with an AI boost for clearer, sharper low-light videos. Snapdragon 8 Gen 3 also brings Video Object Eraser (from ArcSoft) to the platform to easily remove unwanted objects and people from videos in seconds. Pretty great feature for content creators, but we will probably need to have a separate news and information-focused discussion about this feature at some point. Being able to edit objects and people out of videos on the fly is a very cool feature for most everyday users but could pose problems in the hands of malicious actors.

Speaking of enabling content authentication, Truepic photo capture with C2PA standard support will mark photos with Truepic’s cryptographic seal to prove that a photo is real and not created by generative AI. (I have not yet confirmed what solution Snapdragon 8 Gen 3 leverages to similarly help authenticate unaltered video content.)

Snapdragon 8 Gen 3 also features two always-sensing cameras in the front and back to allow for easy QR code scanning, face unlock, and more. Zoom Anyplace, powered by Samsung’s first 200 MP image sensor optimized for Snapdragon, allows capture of multiple videos, object tracking, and 2x and 4x zoom—all in 4K. This seems to strongly suggest that 1) Samsung’s next generation of flagship phones will be powered by the Snapdragon 8 Gen 3 platform, and 2) that the partnership between Qualcomm and Samsung looks to be deepening, which gives us some sense of where premium Android is headed.

First HDR photo technology from Dolby also enables capture and playback of images with a greater range of colors, tones, and shades to reflect the user’s full creative vision. Vlogger’s View lets users share both their selfie camera and rear camera video simultaneously to put them in the action as they share and narrate what they are capturing live. Lastly, Photo Expansion leverages generative AI to intuitively extend a photo beyond what was captured, which is probably the most “I didn’t realize I needed this until I saw it” feature of the entire camera section of the Snapdragon 8 Gen 3 presentation.

Key Spectra Image Signal Processor specs: Cognitive ISP, Triple 18-bit ISPs, real-time semantic segmentation photo and video processing (up to 12 layers), AI for 3A (auto-focus, auto-exposure, and auto-white balance), up to 10-bit color depth photo and video capture, 8K HDR video capture + 64 MP photo capture, 10-bit HEIF: HEIC photo capture, HEVC video capture, Google Ultra HDR photo capture, and of course Truepic photo capture with C2PA-compliant certificate authority.

One of the main workhorses here appears to be the brand-new engine for visual analytics (4.0), which includes hardware acceleration for iToF depth sensors up to 1080p30. Image sensor support includes DCG HDR, staggered HDR, QDOL HDR, Less Blanking HDR, and multiframe HDR image sensors. (Also, triple video capture from HDR image sensors with seamless switching between any HDR mode.)

Performance expectations: up to 36 MP triple camera at 30 FPS with zero shutter lag, up to 64+36 MP dual camera at 30 FPS with zero shutter lag, and up to 108 MP single camera at 30 FPS with zero shutter lag. AI-based face detection (up to 200 MP) in photo captures is one of the more intriguing features here, and one that I feel deserves a follow-up discussion, so I plan to circle back to it in an upcoming article.

On the video side, video capture HDR formats include HDR10+, HDR10, HLG, and Dolby Vision. Video performance expectations: 8K HDR video capture at 30 FPS, 4K video capture at 120 FPS, and slow-motion video capture at 720p at 960 FPS. Specs also highlight Bokeh Engine 2 for video capture, Pro Sight video capture, and the very exciting Night Vision video capture with RAW AI Noise Reduction in 4K 60 FPS. Computational HDR video capture can support up to four exposures (with QDOL image sensors, so OEM hardware implementation is going to be key here). Other video features include video super resolution, multiframe noise reduction (MFNR), and Locally Motion Compensated Temporal Filtering.

What I will be most interested in here, once I see how device OEMs implement these video features, is if any Android handset finally delivers video quality that is on par or better than what premium iPhones have been putting out until now. As an Android user, I continue to contend that in the higher price tiers, Android easily beats Apple in the photo department, but that Apple still wins when it comes to mobile video quality. I do not believe that is as much about Snapdragon’s capabilities as much as it is about OEM feature implementations, but there comes a point where the Android ecosystem needs to commit to challenging that narrative and the market expectations that endure because of it. Snapdragon 8 Gen 3 gives premium handset OEMs the hardware and AI acceleration specs they need to do that in 2024.

Snapdragon Elite Gaming Redefines Expectations for Mobile Gaming Yet Again

Snapdragon Elite Gaming features have been getting increasingly impressive year after year, especially given the form factor that they are limited to. Every year, mobile gaming user experience (UX) grows more immersive with lifelike, multisource lighting powered by hardware-accelerated (40% better) ray tracing and global illumination – a first-in-mobile feature thanks to Unreal Engine 5.2. Snapdragon 8 Gen 3 also supports ultra-smooth, lag-free 240 FPS graphics on 240 Hz displays. (Yes, 240 FPS. That is not a typo.) Mobile gaming performance can also be maximized by upscaling game scenes up to 8K external displays with Snapdragon Game Super Resolution, which is also a little mind-blowing given how small the SOC is compared with, say, a standalone gaming console. Adreno Frame Motion Engine 2.0 generates even higher quality scenes, doubling the frame rate while maintaining the same power consumption. Part of the secret sauce is Qualcomm’s 30% faster and 20% more power efficient Kryo CPU and new subsystem (featuring five gold cores), an upgraded 25% faster and 25% more power-efficient Adreno GPU, and improved battery life (for more or less 10% overall power savings).

Additional specs that play well into the platform’s gaming performance: on-device display support for 4K at 60 Hz and QHD+ at 144 Hz. Maximum external display support: up to 8K at 30 Hz or up to 1080 at 240 Hz. Variable refresh rate support for 240 Hz to 1 Hz. Also 10-bit color depth, Rec. 2020 color gamut, HDR10, HDR10+, HDR vivid, and Dolby Vision (as stated earlier). Snapdragon 8 Gen 3’s visual subsystem is a powerhouse, especially for such a small package. I cannot stress enough how impressed I am with that 240 FPS spec, especially on mobile.

Audio: The Unsung Hero of the Snapdragon Tech Stack Deserves More Attention

Sound is also a key piece of the UX pie, and here again, Qualcomm has been more or less quietly pushing the limits of wireless audio performance for years without receiving as much recognition on the consumer side of the industry as is perhaps deserved. (Snapdragon Sound has been crushing it.) This year, Snapdragon Sound, which supports 24-bit 96 kHz lossless music over Bluetooth, introduces a new feature called Qualcomm Expanded Personal Area Network Technology (XPAN), which enhances uninterrupted lossless audio even when a user is away from their phone (or moving from room to room)—without burning through power uselessly. One of the more significant improvements to Snapdragon Sound on 8 Gen 3 also brings phone-to-earbud lag mitigation improvements to Snapdragon Sound to sync content streaming and gameplay to the millisecond. (An area that still needed a tiny bit of improvement.)

Key features support and performance specs: Qualcomm Aqstic audio codec, Qualcomm Aqstic smart speaker amplifier, Total Harmonic Distortion + Noise (THD+N), Qualcomm Audio and Voice Communication Suite, and Spatial audio with head-tracking. Playback: -108dB

Connectivity: Qualcomm Is Still by Far the Best in the Business

Naturally, as this is Qualcomm we are talking about, Snapdragon 8 Gen 3 comes with the best-in-class 5G, Wi-Fi, and Bluetooth technologies available to consumers. The new Snapdragon X75 Modem-RF System is the world’s first 5G modem with an integrated AI tensor hardware accelerator. Specs point to a 2.5x AI processing power improvement at the hardware level from its predecessor, for faster speeds, improved power efficiency, better location accuracy and coverage, and improved link robustness. Other key specs: Downlink speeds up to 10 Gbps and uplink speeds up to 3.5 Gbps, global 5G multi-SIM, including Qualcomm DSDA Gen 2, mmWave: 8 carriers, 2×2 MIMO, sub-6 GHz: 4×4 MIMO, and converged mmWave-sub-6 transceiver. Qualcomm’s 5G AI Suite Gen 2 includes a sensor-modem-RF solution for mmWave beam management, Al-enhanced channel state feedback, Al-enhanced antenna tuning, Al-enhanced GNSS Location Gen 2, Qualcomm 5G Al Processor Gen 2 with dedicated tensor hardware accelerator, Qualcomm 5G PowerSave Gen 4, Qualcomm Smart Transmit Gen 4 technology with Snapdragon Satellite support, Qualcomm Wideband Envelope Tracking, Qualcomm Power RF Efficiency Suite, Qualcomm 5G Ultra-Low Latency Suite, and of course 3GPP Release 17 and Release 18 support. Last, the multimode support list looks as complete as you would expect: 5G NR, NR-DC, EN-DC, LTE, CBRS, WCDMA, HSPA, TD-SCDMA, CDMA 1x, EV-DO, and GSM/EDGE. Digging into location specs a little, I spy Concurrent GPS, Glonass, BeiDou, Galileo, QZSS, NavIC, and triple frequency GNSS (L1/L2C/L5). Sensor-Assisted Positioning 6.0 comes with global map aiding, urban pedestrian navigation with sidewalk accuracy, and global freeway lane-level vehicle navigation.

Additionally, Qualcomm’s FastConnect 7800 mobile connectivity system features leading Wi-Fi 7 for premium speeds and reliability. As far as I can tell, it is the only Wi-Fi system supporting High Band Simultaneous Multi-Link (for blazing fast, low-latency performance even in busy and complex bandwidth environments). Key specs include: Wi-Fi 7 with peak speeds up to 5.8 Gbps (802.11be, 802.11ax, 802.11ac, 802.11a/b/g/n). Wi-Fi Spectral Bands: 6 GHz, 5 GHz, 2.4 GHz. Channel Bandwidth: 320/240/160/80/40/20 MHz. 8-stream sounding (for 8×8 MU-MIMO), MIMO Configuration: 2×2 (2-stream), MU-MIMO (Uplink & Downlink), 4K QAM, OFDMA (uplink and downlink), and High Band Simultaneous (HBS) Multi-Link. Wi-Fi Security: WPA3-Enterprise, WPA3- Enhanced Open, WPA3 Easy Connect, WPA3-Personal Integrated Bluetooth. Bluetooth Audio is Snapdragon Sound Technology with support for Qualcomm XPAN Technology (see previous section), Qualcomm aptX Voice, aptX Lossless, aptX Adaptive, and LE.

I have not had the opportunity to verify this yet, let alone quantify the difference, but my assumption here is that pairing a mobile device powered with the Snapdragon 8 Gen 3 platform (or running FastConnect 7800) with Bluetooth earbuds or headphones that support Snapdragon Sound, Qualcomm XPAN, Qualcomm aptX Voice, aptX Lossless, and aptX Adaptive, will likely deliver a higher level of nearly lagless audio performance than a less advanced audio product.

Qualcomm Delivers the Android Ecosystem’s Best SOC Yet Again

At the risk of sounding like a broken record, this year’s mobile Snapdragon SOC, the Snapdragon 8 Gen 3, is the most complete, well-rounded flagship Android mobile chipset on the market, thanks in part to its significant AI-powered upgrades. That is not to say that competitors will not release impressive chipsets with their own best-in-class features and specs. MediaTek comes to mind, obviously, and I very much look forward to seeing how its upcoming Dimensity 9300 chipset, based on TSMC’s 3 nm process, will stack up against Snapdragon 8 Gen 3. (In fact, I will be a little disappointed if Dimensity 9300 does not outperform Snapdragon 8 Gen 3 in a few critical tasks. Competition is good for the Android ecosystem and for users, so game on.) Google is also in the race now with its new Tensor G3 SOC, although it might take another year or two before we can realistically compare it with a flagship Snapdragon SOC.

That said, looking at all of the top SOC specs and performance benchmarks, here is what I keep coming back to: Snapdragon’s unique advantage is that in addition to consistently delivering most of the best real-world performance and features in the industry, it is also the most well-rounded platform for OEMs to build premium experiences on. Even if it does not always turn up the best performance in every benchmark test (and not every benchmark is relevant to real-world use cases), the platform’s overall performance across a breadth of features completely overwhelms the few areas where it might not be number one.

Taking this line of thinking a bit further, even if, for the sake of argument, you were to brush aside all other spec and performance discussions, Qualcomm enjoys yet another critical advantage in the mobile SOC market: Qualcomm’s connectivity solutions are second to none (which is probably why Qualcomm continues to supply Apple with cellular modems). No matter how good mobile CPUs and GPUs get, a phone is first and foremost a connected device: Calls, emails, videoconferencing, navigation, search, streaming media, social apps, productivity, gaming – none of these are possible without a solid connectivity foundation, and the more premium the handset tier, the faster, smoother, more reliable the connectivity piece of the overall mobile UX equation has to be. Qualcomm still holds the high ground here too.

Last, a few thoughts about the transformative, disruptive force that is on-device AI, and how it could help reset the mobile SOC competitive order in the next few years. Taking a quick inventory of what I have seen so far this year, we could potentially see Google and MediaTek leverage on-device AI to close the performance and features gap with Qualcomm’s Snapdragon 8 Gen 3. Maybe. (We will likely see them compete more evenly in the more budget-priced tiers.)

At the premium/flagship level though, Qualcomm’s long bet on on-device AI is already paying off. (If it seems that Qualcomm was hitting the gas on on-device AI long before most technology users were even aware of its potential, it is because that is exactly what Qualcomm engineers were doing.) Qualcomm identified the opportunity early, invested the right amount of resources to accelerate its development, and found itself in the right place at the right time. Solid execution and convergence of capabilities and market needs. Is Qualcomm’s head start advantage insurmountable? I do not think so. AI presents other semiconductor giants with a new opportunity to compete against Snapdragon and perhaps, if they are clever and imaginative enough, level the playing field a bit. For now, though, Qualcomm continues to dominate the flagship Android SOC space and set the bar for premium UX, features, and performance.

Smartphones powered by Qualcomm’s Snapdragon 8 Gen 3 should start to enter the market later this year.

Disclosure: The Futurum Group is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.

Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of The Futurum Group as a whole.

Other insights from The Futurum Group:

Qualcomm Raises Bar for On-Device Generative AI at Snapdragon Summit

Google Tensor G3 SOC Elevates Pixel 8 and Pixel 8 Pro Performance

Apple iPhone 15 Defects?

Image Credit: Qualcomm

Author Information

Olivier Blanchard has extensive experience managing product innovation, technology adoption, digital integration, and change management for industry leaders in the B2B, B2C, B2G sectors, and the IT channel. His passion is helping decision-makers and their organizations understand the many risks and opportunities of technology-driven disruption, and leverage innovation to build stronger, better, more competitive companies.  Read Full Bio.

SHARE:

Latest Insights:

All-Day Comfort and Battery Life Help Workers Stay Productive on Meetings and Calls
Keith Kirkpatrick, Research Director with The Futurum Group, reviews HP Poly’s Voyager Free 20 earbuds, covering its features and functionality and assessing the product’s ability to meet the needs of today’s collaboration-focused workers.
Paul Nashawaty, Practice Lead at The Futurum Group, shares his insights on the Aviatrix and Megaport partnership to simplify and secure hybrid and multicloud networking.
Paul Nashawaty, Practice Lead at The Futurum Group, shares his insights on AWS New York Summit 2024 and the democratizing of Generative AI.
Vendor Leverages Amazon Q on AWS to Drive Productivity and Access to Organizational Knowledge
The Futurum Group’s Daniel Newman and Keith Kirkpatrick cover SmartSheet’s use of Amazon Q to power its @AskMe chatbot, and discuss how the implementation should serve as a model for other companies seeking to deploy a gen AI chatbot.