Imagination re-works GPUs for latest Series9 offerings

“These new GPUs are the best-ever mainstream PowerVR GPU cores and will enable us to continue to increase our Android device footprint and our overall volume across mobile, set-top boxes, digital TVs, and automotive applications,” said Imagination executive v-p Nigel Leeder.

Series9XEP
Based on the the firm’s Rogue architecture, these cores are optimised for fillrate/mm² and feature PVRIC4 visually lossless compression to reduce bandwidth and memory footprint. They are aimed at GUIs, HMIs and casual gaming in set-top boxes, digital TVs, low-cost mobile devices, mid-range automotive dashboards and in-vehicle infotainment.

Series9XMP
The firm is claiming high compute density (Gflop/s/mm²) for these GPUs. Also Rogue-based, they are aimed at computation and gaming on mid-range mobile devices, set-top boxes and digital TVs, “and provide the foundation for AI and neural network processing on the GPU”, claimed the firm.

Series9XTP
Based on the newer Furian architecture, “these new GPUs have dramatically increased gaming-compute density and extensive power-performance-area optimisations throughout the design, making them ideal for premium mobile devices and high-end in-vehicle infotainement, as well as applications across compute-server segments”, according to Imagination – Electronics Weekly has requested power, performance and area data.

All the new GPUs include PVRIC4 – Imagination’s own image compression and include a new alpha buffer/block hint feature for reduced composition workload bandwidth.

They can be paired with the Series3NX neural network accelerator, using the GPU to manage classic vision and preprocessing algorithms and the Series3NX to handle optimised fixed-point neural network processing.

Optimisations in the recently announced PowerVR Automotive pack can be applied to any of them.

Last year the firm announced the 9XE and 9XM GPUs.

According to a statement, PowerVR Series9XEP GPUs were designed to go beyond Series9XE with:

• Reduced bandwidth through PVRIC4 and increased cache sizes
• Timing improvements to enable higher core clock speeds compared to the previous generation

Then for Series9XMP it said:

• Up to a 45% increased gaming/compute density for better gaming
• Up to 2x anisotropic filtering speed, up to 4x shadow sampler speed
• ALU up from 16-wide to 32-wide
• Compute improvements in data paths (caching), atomic operations and datamaster setup rates
• Reduced bandwidth through PVRIC4 and increased cache sizes
• Microarchitectural tuning and timing improvements

And for Series9XTP:

• Increased gaming and benchmark density with up to a 50% frame/s/mm² density increase
• low-level architectural tuning
• ALU width up to 40-wide per cluster for increased FLOP density
• Reduced bandwidth through PVRIC4 and increased cache sizes
• YUV speed up for 2- and 3-planar YUV

SDKs, tools and utilities are available for 3D graphics application development, and tools and utilities are available for GPU compute applications.

Physical design optimisation kits (DOKs) comprise optimised reference design flows, tuned libraries from partners, characterisation data and documentation.

The cores can be licensed now