Snapdragon 8 Elite Extreme Gen 6 Hits 5GHz With Neural Fusion Gaming, Wi-Fi 8

Today at its Snapdragon Summit event, Qualcomm is launching its next-generation flagship smartphone silicon with not one, but two new members of the Snapdragon 8 Elite family. The Snapdragon 8 Elite Gen 6 and top-end Snapdragon 8 Elite Extreme Gen 6 are both built on an unspecified 2nm process and introduce a new generation of Qualcomm's custom Oryon CPU, Adreno GPU, Hexagon NPU, Spectra ISP, and advanced connectivity hardware.

The two chips are extremely similar at the CPU and platform level, with the Extreme model distinguishing itself primarily through a more capable GPU and expanded multimedia capabilities. Both chips feature an eight-core Oryon CPU with two Prime cores running at up to 5GHz and six Performance cores running at up to 4GHz, along with 16MB of Qualcomm's new Oryon Flex Cache. Qualcomm describes the 5GHz Oryon design as the first mobile CPU to reach that clock speed.

Oryon CPU Reaches 5GHz

The Oryon CPU retains the basic two-Prime, six-Performance configuration used by previous Snapdragon 8 Elite processors, but Qualcomm has redesigned the CPU subsystem around a new cache architecture called Flex Cache, which provides a dynamically allocated shared cache pool that can be accessed by the heterogeneous CPU cores. Qualcomm says this allows the Prime cores to draw on the entire cache when workloads demand it, rather than being constrained by a fixed cache allocation. Larger working sets can therefore remain resident in the CPU cache instead of spilling into external memory, reducing latency and memory traffic as work moves between cores.

The Snapdragon 8 Elite Gen 6 and Elite Extreme Gen 6 both feature 16MB of Oryon Flex Cache. The Extreme model's CPU isn't simply a faster version of the standard Gen 6, then; both are specified with the same maximum CPU frequencies and cache capacity. The chips also support both LPDDR5X and LPDDR6 memory at up to 5,300MHz (10.6Gbps DDR), with configurations of up to 24GB of memory, as well as UFS 5.0 storage.

Qualcomm has not specified the memory bus width or resulting peak memory bandwidth for either processor, but the company does say that CPU performance improves 10% and 13% over last generation for the Elite and Elite Extreme, respectively. Those gains are broadly aligned with what we'd expect from a small clock bump and cache improvements. The more impressive detail is the efficiency story; Qualcomm says CPU power efficiency is up 37% over last-gen. Of course, that's mostly thanks to the move to 2nm, but it's still very impressive.

Hexagon NPU Targets Agentic AI

Qualcomm is putting much of the new generation's emphasis on the much-vaunted Agentic AI, where an AI system handles multiple steps, tools, and tasks rather than simply generating an answer to a single prompt. The revised Hexagon NPU features a new Element Accelerator designed specifically around transformer workloads alongside 12 scalar units, eight vector units, and one tensor unit. Qualcomm also lists Hexagon Direct Link, Micro Tile Inferencing, and 64-bit memory virtualization as features.

The chip firm says that the new NPU has a 50% larger shared-memory subsystem than the previous generation, although it didn't actually tell us the absolute size of that memory. The expanded local memory seems intended to keep model state, activations, and intermediate tensors closer to the NPU, once again reducing trips to external system memory. Qualcomm says the resulting architecture is designed for longer context windows, concurrent workloads, and low-latency "action loops" in which an agent can repeatedly reason, call a tool, process the result, and continue working.

The company is also positioning the NPU for much larger on-device models. Qualcomm says the platform can run Mixture-of-Experts (MoE) models with more than 30 billion total parameters, with roughly 3 billion routed parameters active for each token-generation step in its example. Because an MoE model doesn't execute all of its parameters for every token, that claim shouldn't be interpreted as equivalent to running a 30-billion-parameter dense model at its full computational cost; these are very different workloads.

For INT4 models, Qualcomm claims up to 50% higher prefill performance, along with faster decoding, enhanced speculative decoding, and higher overall tokens per second. The company is also emphasizing a broad INT2-to-FP16 precision range (including mixed-precision) as part of its effort to support different classes of local AI models.

The Hexagon NPU isn't the only AI processor in the platform, either. Qualcomm's new Sensing Hub includes "Dual Micro NPUs" as well as dual always-sensing ISPs intended to process sensor information of all types while keeping power consumption low. Qualcomm says the new Micro NPUs provide 85% more performance and 20% greater efficiency, while supporting models of up to 200 million parameters. The Sensing Hub also supports Qualcomm's Personal Scribe and the creation of personalized knowledge graphs. Privacy concerns aside, this sounds rather similar to a mobile version of Microsoft's Copilot for Windows 11.

Adreno Adds Dedicated AI Hardware For Neural Fusion

The biggest generational improvements, as well as the largest architectural distinction between the two Snapdragon 8 Elite Gen 6 processors, both come on the graphics side. The standard Snapdragon 8 Elite Gen 6 uses a next-generation Adreno GPU (no name given) with Qualcomm's high-performance memory architecture, while the Elite Extreme Gen 6 adds dedicated Adreno Matrix Cores. Qualcomm describes the Matrix Cores as AI-dedicated GPU cores designed to run advanced AI models directly within the graphics pipeline, much like NVIDIA's Tensor Cores.

Those Matrix Cores are the foundation for Qualcomm's new Adreno Neural Fusion technology. Neural Fusion combines AI processing, performing super resolution and frame generation directly in the graphics pipeline. This will be familiar to most of the HotHardware audience, as it's directly analogous to NVIDIA's DLSS or Intel's XeSS. Qualcomm makes some bold claims about the quality and performance of Neural Fusion, and we're eager to test it ourselves. Notably, the technology is supported through Unity, Unreal Engine, and other development environments, allowing developers to integrate the feature without creating an entirely separate rendering pipeline.

The Extreme model's GPU also includes the full 18MB of Adreno High Performance Memory. Qualcomm describes HPM as a low-latency memory structure integrated closely with the GPU, keeping tiles, frame buffers, and compute data local to the graphics subsystem rather than repeatedly transferring that information through system memory. The new Adreno architecture uses three GPU slices, and Qualcomm says the combination of local memory and dedicated Matrix Cores is intended to reduce both latency and power consumption.

Performance-wise, the chipmaker claims 35% and 44% improved GPU performance over the Snapdragon 8 Elite Gen 5 for the Elite and Elite Extreme, respectively; these gains come along with an impressive 40% gain in GPU power efficiency, gen-on-gen. That means power actually goes down for the Snapdragon 8 Elite Gen 6, while the Extreme uses a bit more, although that's to be expected given all of the extra silicon present in that part.

The Elite Extreme Gen 6 also adds support for Unreal Engine 5's advanced graphics optimizations, including the new MegaLights feature, alongside hardware-accelerated ray tracing for Lumen global illumination as well as Nanite virtualized geometry. The standard Gen 6 retains support for the latter graphics technologies but does not receive the Extreme model's Neural Fusion feature set nor, apparently, support for MegaLights. We saw an Unreal Engine 5 demo running live on a demo unit at the summit and it looked very smooth, although it's not clear how much of that frame rate came from frame generation.

Spectra Pushes Mobile Video Further

Both processors feature Qualcomm's Spectra image signal processor with "triple 20-bit AI ISPs," real-time 3D LUT support through the Elite Color Engine, and support for a wide range of modern HDR sensor technologies. The image pipeline reportedly supports up to 64-megapixel triple-camera operation with zero shutter lag, up to 108-megapixel single-camera operation with zero shutter lag, and still-image capture at up to 320 megapixels.

The ISP also provides substantially more AI processing capability. Qualcomm specifies 16 bits of AI data per pixel, enhanced AI video segmentation at 4K60, and real-time AI skin and sky tone adjustments at 4K60. The system can perform AI-assisted noise reduction, including Bayer multi-frame processing, and supports ultra-low-light video capture at 4K60.

The Extreme model extends the platform's video capabilities substantially. Snapdragon 8 Elite Gen 6 supports 8K video recording at 30 frames per second and 4K slow-motion capture at 120 frames per second, while Elite Extreme Gen 6 raises those limits to 8K60 and 4K240. The Extreme model also adds support for the Advanced Professional Video codec, developed over years in close cooperation with rival Samsung.

The Spectra ISP can also work closely with Hexagon to provide considerably more granular AI information during image processing. Qualcomm describes this as providing up to 256 times more granular AI data per pixel, although the precise baseline behind that comparison isn't exactly clear. We suspect it simply refers to a move from 8-bit (256 values) to 16-bit (65,536 values) precision on the AI tag data, but that's not confirmed.

New Connectivity for the 5G Advanced Era

Both Snapdragon 8 Elite Gen 6 platforms integrate Qualcomm's X105 5G Modem-RF system, which supports peak downlink speeds of 14.8Gbps and peak uplink speeds of 4.2Gbps. That's approximately 18% faster on the downlink and 14% faster on the uplink versus the previous generation X85 modem. The X105 supports 5G Advanced, both standalone and non-standalone operation, mmWave and sub-6GHz connectivity, up to eight mmWave carriers, six-carrier aggregation with up to 500MHz of downlink bandwidth (+25% over last-gen), four-layer uplink carrier aggregation, 1024-QAM on sub-6GHz, and six-antenna receive configurations.

Qualcomm also says the X105 is hardware-ready for 3GPP Release 19 and supports both NB-NTN and NR-NTN, providing a path toward satellite connectivity using cellular standards. An integrated fifth-generation Qualcomm 5G AI processor provides additional AI-based network optimization, while the modem supports global 5G bands and Qualcomm's Turbo DSDA multi-SIM technology.

The wireless connectivity subsystem gets an even more substantial upgrade with Qualcomm's FastConnect 8800. The new system supports Wi-Fi 8 and a peak PHY rate of 11.6Gbps, using a 4x4 design and high-band simultaneous multi-link operation. FastConnect 8800 also integrates Bluetooth 6.0, Bluetooth Channel Sounding, Qualcomm's High Data Throughput technology, Ultra Wideband, and Thread 1.4, making it remarkably versatile. Qualcomm is positioning these technologies as part of a broader effort to make the smartphone aware of its surroundings and its nearby devices, rather than simply providing higher raw wireless throughput.

Two Flagships, One New Platform

The Snapdragon 8 Elite Gen 6 family is really less about a dramatic increase in CPU speed than it is about broad improvements throughout the smartphone SoC. The new Oryon CPU reaches 5.0GHz and delivers up to 10% more performance on the standard Gen 6 and 13% more on the Extreme, while Qualcomm claims substantially greater power efficiency (37% on both). Even larger gains arrive elsewhere, with the Adreno GPU apparently improving by up to 44% and the Hexagon NPU by up to 35% on the Extreme model.

That performance split also explains why Qualcomm is offering two flagship platforms. The standard Snapdragon 8 Elite Gen 6 carries the new CPU, NPU, connectivity, and graphics architecture into a broader range of premium smartphones, while the Extreme adds dedicated graphics AI hardware, additional high-performance GPU memory, Neural Fusion, and higher-end media capabilities that many users may not want or need. Rather than making the Extreme a faster CPU, Qualcomm has elected to build a more capable accelerator-heavy platform for enthusiasts and power users.


Motorola is providing the first concrete example of that strategy with the "Motorola Signature 27," which it announced on stage at the Summit and which it plans to launch as one of the first smartphones powered by the Snapdragon 8 Elite Extreme Gen 6. The handset pairs Qualcomm's highest-end mobile silicon with an advanced multi-camera system including a 50MP Sony LYTIA 910 main sensor and 200MP periscope telephoto, seven years of operating system upgrades, a new Audio by Bang & Olufsen partnership, and a substantially upgraded ArcticMesh cooling system, which Motorola emphasizes considerably in its press release, signaling to us that the new chip can get pretty spicy.

The result is a Snapdragon generation where the most significant advances aren't necessarily visible in a spec sheet or technical readout. Qualcomm has increased CPU, GPU, and AI performance while targeting substantially lower power consumption, and it has added increasingly specialized hardware throughout the SoC to keep data close to the engines processing it. The coming generation of flagship phones will be the ultimate measure of just how much of that silicon can be translated into real-world performance, battery life, camera quality, and gaming experiences.