Qualcomm Unveils Next-Gen Adreno GPU: 1.45GHz Clocks, 18MB Cache, AI Cores

Qualcomm has disclosed technical details of its next-generation Adreno GPU, which will power upcoming flagship mobile platforms. The new GPU retains a three-slice architecture while pushing clock speeds higher and, for the first time, introduces an on-chip high-performance cache, dedicated Matrix Cores for AI computation, and the Adreno Neural Fusion architecture. These changes aim to optimize sustained graphics performance and power efficiency at the architectural level.

The new Adreno GPU continues with a three-slice design, with each slice's operating frequency increased by 250MHz from 1.2GHz to 1.45GHz. Qualcomm said the GPU has undergone significant architectural adjustments focused on improving sustained graphics performance and power efficiency, with full performance data to be released later.

To reduce frequent access to system memory, the new platform is equipped with 18MB of Adreno High Performance Memory (HPM), an on-chip high-speed storage space dedicated to graphics processing that can serve as a configurable graphics workspace. During rendering, Adreno HPM can store render targets, depth and color buffers, textures from previous rendering stages, and other intermediate data. Developers can flexibly utilize the on-chip memory based on different graphics workloads, thereby reducing latency and power consumption associated with repeated data transfers to system DDR memory. Qualcomm's data indicates this design can deliver approximately 12% power efficiency improvements.

The new Adreno GPU introduces Qualcomm Matrix Cores for AI operations for the first time. These dedicated compute units can run AI models directly within the GPU graphics pipeline, extending matrix acceleration capabilities across major compute units such as CPU, NPU, and GPU, while reducing the overhead of migrating neural network tasks between different processing units.

Building on Matrix Cores, Qualcomm has further introduced Adreno Neural Fusion technology, which integrates AI computation more directly into the graphics rendering pipeline. This technology targets scenarios such as neural rendering, AI super-resolution, and frame generation, leveraging the GPU's internal AI compute power to enhance image quality and rendering efficiency while maintaining high graphics performance and reducing power consumption for improved sustained operation. AI super-resolution is expected to be one of the first key capabilities to debut with Adreno Neural Fusion. Specific details regarding the new Adreno GPU's model number, full operating frequencies, graphics performance improvements, and the initial list of games supporting Neural Fusion are yet to be announced.