Qualcomm has released a new generation of flagship XR chip Snapdragon Reality Elite, which will be the first to be installed in the computing unit of the Xreal Aura Android XR device launched this fall.Qualcomm’s naming this time is different from its previous XR chip product line, but it is still positioned as a flagship solution for high-performance independent XR devices, with special emphasis on supporting multiple forms: the chip can be directly integrated into the head-mounted display, or placed in a wired external computing box (compute puck). It is also compatible with different display systems such as video see-through and optical see-through.
Compared to the previous generation flagship Snapdragon XR2+ Gen 2, Snapdragon Reality Elite offers significant improvements in graphics, general computing, and AI capabilities. Qualcomm claims that the GPU performance of the new chip has been improved by about 60%, the CPU performance has been improved by about 30%, and the NPU performance for machine learning tasks has been improved by 160%, with a computing power of up to 48 TOPS. In addition, the EVA (Visual Analysis Engine) module inside the chip has been expanded to accelerate more computer vision tasks, including 3D environment reconstruction.
In terms of imaging and power consumption, Qualcomm said Snapdragon Reality Elite can bring a higher quality, lower latency camera see-through experience. Compared with XR2+ Gen 2, the new chip further reduces latency on the "photon-to-photon" link by 10%, reduces power consumption by approximately 33% under the same workload, and integrates a more advanced image noise suppression algorithm. At the same time, the overall system power consumption optimization also extends the battery life by about 20% under the same load, and the chip temperature can be reduced by up to about 12 degrees Celsius when fully loaded, which is considered to be one of the key factors that makes it suitable for use in the user's pocket.
In terms of storage and connectivity, Snapdragon Reality Elite supports higher-speed UFS 4.0 storage and a memory frequency of up to 4.2 GHz, which is significantly improved compared to the 3.2 GHz of the previous generation. In terms of interface configuration, the chip natively supports up to 2 USB 3.1 interfaces and integrates support for Bluetooth 6.0 for the first time, providing a basis for higher bandwidth and lower latency for XR devices to connect peripherals and wireless accessories.
Qualcomm believes that the jump in NPU performance will provide the computing foundation for "a new generation of locally generated AI experiences", from lifelike digital avatars to large language model agents running locally, as well as applications such as rapid real-time 3D object generation. According to official examples, a large language model with 3 billion parameters can achieve a generation speed of about 45 tokens per second on the NPU, and a large visual model with a resolution of 512×512 can also complete reasoning within a delay of about 1.7 seconds.
Faced with the industry's concern about visual analysis capabilities, Qualcomm said in an inquiry that from the perspective of chip capabilities, the expanded EVA module can provide efficient real-time continuous scene grid reconstruction and other functions for the head display without a depth sensor. However, whether such capabilities can ultimately be implemented in specific products will also depend on the implementation of equipment manufacturers and application developers.
In terms of product launch rhythm, Xreal has confirmed that its Aura Android XR device will be on sale this fall, and will adopt a matching computing unit solution, placing Snapdragon Reality Elite in an external "puck" to provide computing power support for the headset. Another XR brand, Play For Dream, also publicly stated that its next flagship device will use Snapdragon Reality Elite as the core platform to further expand the coverage of this chip in the high-end XR market.
Qualcomm did not give a direct answer when asked whether placing the chip in a computing box in the user's pocket would limit performance due to poor cooling conditions compared to a head-mounted display with a built-in active fan. The company only reiterated that the Snapdragon Reality Elite was originally designed to cover multiple forms, and the specific cooling and performance scheduling strategies will be determined by each machine manufacturer based on its own product form.