Qualcomm Snapdragon X Elite, AI, Smartphones, AI, and Sound. (And AI.)

The highlight of Qualcomm’s annual media event in Maui, Hawaii is usually a new Snapdragon smartphone chip with incremental improvements. We got that this year, along with much more. Qualcomm executives were practically giddy about their new exceptionally powerful and efficient Oryon CPU platform that is dramatically accelerating Qualcomm’s reach in PCs. Oryon should also make Qualcomm more competitive in smartphones, XR, and automotive in 2025 - 2027. However, the focus of Summit was AI: Qualcomm is all-in on on-device AI with huge NPUs on its PC, smartphone, and even audio chipsets. It’s a bet that generative AI is not a fad, and that AI will change computing interfaces and how knowledge work is done.

Snapdragon X Elite PC Computing Platform

Qualcomm has been trying to grow into the PC laptop market for years with relatively little to show for it, but that could change with the Snapdragon X Elite. There have been two stumbling blocks to Windows on Arm taking meaningful market share from Intel and AMD: value proposition and software compatibility. I was at the Snapdragon Summit when Qualcomm first announced a Snapdragon processor for Windows that it promised would offer two weeks of always-connected battery life. This was, to put it charitably, not true. Battery life in the 12 – 18 hour range was realistic on Snapdragon-based laptops, but with the right combination of low-power x86 processor and battery, Intel and AMD could match that. The HP Dragonfly G4 I’m using to write this report is getting well over 12 hours of heavy use between charges, and longer is possible with lighter workloads. Qualcomm also promised constant connectivity, but that didn’t prove to be a differentiator that buyers cared about – and that, too, is possible on x86. (Until I doused it with diet Pepsi, the MediaTek 5G modem in this HP laptop was working fine.)

Windows on Arm also suffered from performance penalties and software incompatibility. Qualcomm’s past Arm CPU designs were adequate for office and web browsing, but little more. It took time, but more apps have been written or recompiled as native Win64 code, while Microsoft has improved Win32 emulation enough that it shouldn’t be a deal-breaker for most workloads. I have been testing Lenovo’s ThinkPad X13s, based on Qualcomm’s most recent Snapdragon 8cx Gen 3, and with Microsoft’s latest updates, it was able to run encryption software and scanner drivers without a hitch – major failings in the past.

Qualcomm’s Snapdragon X Elite is not just an iterative improvement on the Snapdragon 8CX. Thanks to new Oryon CPU cores, the platform significantly outperforms x86 architectures, and it has extensive on-device AI capabilities as well. Snapdragon X Elite is based on a 4nm process with a 12-core Oryon CPU, Hexagon NPU for AI, and Adreno GPU. Each Oryon core runs at up to 3.8GHz with dual core spikes up to 4.3GHz when additional performance is needed. I asked Oryon’s designer Gerard Williams why there are no efficiency cores, and he said that its performance cores are so power efficient that there was no need to reason to scale down.

Qualcomm is claiming 2x the performance at a given power level compared to Intel, matching the Intel’s peak performance at only 68% of the power draw, and 60% faster performance than an i7-3800H when given higher power budgets. These claims – and the power/performance curve slides – look an awful lot like Apple Silicon marketing, a coincidence that is deliberate (Gerard Williams was the chief architect of Apple Silicon before launching NUVIA, which was purchased by Qualcomm in 2021). Qualcomm is taking on Apple, too, claiming 50% faster peak multi-threaded performance vs Apple’s M2. Apple has more powerful chips than the M2 – and the M3 will be shipping well before Snapdragon X Elite reaches the market mid-2024 – but the fact that Qualcomm is legitimately in the mix is extremely impressive.

Snapdragon X Elite supports up to 64GB of LPDDR5x RAM, USB 4.0, and multiple high-resolution displays. All of Qualcomm’s other technologies are here as well: Snapdragon X65 5G, WiFi 7 FastConnect 7800, and Snapdragon Seamless (see below). There is a dedicated VPU for H.265 and AV1 encoding and decoding so video production is definitely a use case. The integrated Adreno GPU (4.6 TFLOPS) may be the weak point, relatively speaking. Qualcomm is claiming 2x the performance as Intel’s i7-3800H with no power constraints, and matching performance at 74% less power. Intel systems with that chipset often supplement the integrated graphics with a dedicated Nvidia GPU. As such, there are certainly going to be scenarios in engineering, rendering, and gaming where other solutions will outperform the Snapdragon X Elite (at least while plugged into the wall), but Qualcomm definitely has a competitive and differentiated offering overall.

The design wins bear this out: every significant Windows laptop vendor is signed up to build Snapdragon X Elite laptops in 2024: Acer, Asus, Dell, HONOR, HP, Lenovo, Microsoft’s Surface group, Samsung, and Xiaomi.

The Oryon CPUs are exciting, but it is only half the story: Qualcomm’s overall focus at Snapdragon Summit was on AI. Developing silicon takes years, and the payoff of Qualcomm’s bet on large NPUs ended up perfectly aligning with major advances in generative AI. From a hype and investor sentiment perspective, Qualcomm’s timing couldn’t be better. Snapdragon X Elite’s Hexagon NPU is enormous. It is capable of running 13 billion parameter LLMs on-device, along with AI Windows Studio Effects that improve hybrid work. Running AI on-device requires shrinking the models down from 100+ billion parameters down by a factor of 10, but there are big privacy and security advantages to keeping the data and queries local. There may be benefits to smaller LLMs as well if they are customized to the specific business or domain rather than a general-purpose AI like ChatGPT. Finally, AI in the cloud is expensive, but you’ve already paid for the computing power required to use AI on your device. If the performance and battery life that Oryon brings to Snapdragon X Elite is not enough to get enterprise customers to buy into Windows on Arm, the promise of highly secure, local access to AI resources just might.

Qualcomm may need that AI push, because Intel still has a distribution advantage, and there are still some edge cases where Win32 emulation doesn’t work well. Apple is also a competitor, steadily growing its ecosystem – and Apple still has performance and software integration benefits. Qualcomm is also getting more Windows on Arm competition whether the market can support more options or not. Just as Qualcomm was announcing Snapdragon X Elite, Reuters reported that Nvidia and AMD are also planning Arm-based CPU platforms for Windows. Both companies have deep GPU expertise, which could be an advantage – Nvidia could specifically target gaming. MediaTek is another potential Windows on Arm competitor. Intel certainly looks vulnerable, but if too many vendors enter the market, it could prevent any single entrant from building a profitable business.

Where Will Oryon Go Next?

Qualcomm officially announced that it is putting Oryon into smartphones in the Snapdragon 8 Gen 4, which should be revealed this time next year, with volume sales in 2025. Qualcomm management was clear that Oryon in Snapdragon X Elite is just version 1.0, and that future improvements are in the works. In my discussions with management, Qualcomm would not confirm anything beyond laptops and smartphones, but it is clear that Oryon should also find its way into automotive and XR. One key reason that Apple’s Vision Pro is so powerful is because it can leverage Apple Silicon. Having now seen the direction Apple is headed, some Snapdragon XR2 Gen 2 customers are already asking for more horsepower that can fit into standalone devices, and Oryon fits that bill perfectly.

Snapdragon 8 Gen 3 Smartphone Platform

Qualcomm leads off its Snapdragon 8 Gen 3 features with AI, and, as with its PC platform, Qualcomm perfectly timed the years-long silicon design cycle with the top of the genAI hype. The Snapdragon 8 Gen 3’s neural engine supports multi-modal generative AI models, including key large language models (LLM), language vision models (LVM), and automatic speech recognition (ASR) up to 10B parameters—solely on-device. This should allow lots of genAI customization in mobile apps and the user interface. We may also see hybrid device/cloud genAI implementations as well. It is entirely unclear when or whether genAI on phones will be a critical function of smartphone for consumers, but Snapdragon 8 Gen 3 is the first time you’ll be able to run multiple LLMs on-device in a serious capacity.

While AI is the hot new thing, consumers consistently pick phones based on their camera capabilities, and Qualcomm has plenty of new goodies. Highlights include Qualcomm working with partners. Zoom Anyplace uses a Samsung 200 MP image sensor specifically optimized for Snapdragon to capture multiple videos with object tracking and 2x and 4x zoom in 4K. Dolby is getting into HDR photo capture, not just Dolby Vision on the display end. Truepic’s C2PA standard is supported; it adds a cryptographic seal to pictures to prove that they weren’t generated by AI. Qualcomm also now supports two always-sensing cameras in the front and back allow for easy QR code scanning, face unlock, and privacy concerns.

Like the Snapdragon 8 Gen 2, the Snapdragon 8 Gen 3 is based on 4nm process, but it still has impressive speed and efficiency gains over its predecessor. The Snapdragon 8 Gen 3 does not use Oryon CPU cores, but Qualcomm claims its processing is 30% faster and 20% more efficient. Qualcomm is using an unusual big/medium/small core arrangement: a single big 3.3 GHz Arm Cortex A4, five medium cores with three running at 3.2GHz and two at 3.0GHz, and two small efficiency cores. The Adreno GPU is also 25% faster and 25% more efficient. The 8 Gen 3 supports 240fps gaming on 240Hz displays, which is impressive and perhaps a bit silly on a phone, but you can be sure that ASUS’ next ROG phone will offer it.

AI-based modem, building on Qualcomm’s lead in radios and RF for faster and better connectivity in more situations. Smartphones may be a mature category, but networks have been changing considerably. Consumers with older phones upgrading to a Snapdragon 8 Gen 3 phone will get much better reception across frequency band combinations that their old phone might not even support. Wi-Fi 7 comes along for the ride, too. Wi-Fi 7 is already on the Snapdragon 8 Gen 2, so it isn’t new, but still good for future-proofing.

The Snapdragon 8 Gen 3 builds on Qualcomm’s existing smartphone franchise. Aside from the AI capabilities, it is fairly iterative – but that’s to be expected; smartphones are a mature category. Even if they don’t know what an LLM is, anyone upgrading from a phone two or more years old will find better imaging, better performance, and better battery life.

Snapdragon Seamless

There have been many vendor ecosystem and Microsoft initiatives aiming at improving device interaction: enabling mice, keyboards, and headphones to work across phones, tablets, and PCs; easy file transfer; and getting XR glasses to connect to any phone. This is a great idea that requires a lot of cooperation to pull off, and the key companies that would make it universal – Apple and Samsung – are conspicuously absent. Google’s hardware division isn’t listed either, just “Android.” For U.S. consumers and businesses, the only relevant vendor signed onto Snapdragon Seamless is Lenovo. The Snapdragon Seamless demos in Maui were impressive — people absolutely want these capabilities. However, it seems unlikely that they will

Snapdragon Sound

Qualcomm’s new S7 and S7 Pro Sound Platform chipsets come with impressive audio capabilities including XPAN for better range, low-power WiFi, and improved 4th generation ANC. Qualcomm sells a lot of these chips to headphone OEMs at higher price points, and if you are paying for a premium pair of earbuds, XPAN in particular should be a significant differentiator. Most OEMs are likely to use Qualcomm’s ANC, though they don’t have to – Bose will almost certainly use its own (exceptionally good) IP. Because this is Qualcomm, there is also room on the chip for AI, which could improve virtual assistants or digital health capabilities. The big news is XPAN, which seamlessly switches between Bluetooth and Wi-Fi. This should all but eliminate dropouts, no matter how far you roam from your phone or PC.

That additional functionality is necessary because Snapdragon Sound as a platform is stuck in ecosystem neutral: it requires the entire audio chain to be compatible, few devices and services support it, and many consumers can’t hear the difference. Audio quality has never been a primary device purchase driver.

To discuss the implications of this report on your business, product, or investment strategies, contact Techsponential at avi@techsponential.com.