The GPU module supports the following
features:
- Base architecture, fully
compliant with the following APIs:
- OpenGL ES 3.2
- OpenCL 1.2 EP
- Vulkan 1.2
- Tile-based deferred rendering
architecture for 3D graphics workloads, with concurrent processing of multiple
tiles
- Support for DRM security
- Support for GPU virtualization
- Up to 8 virtual GPUs
- Separate IRQ for each
OSID
- Multi-threaded Unified Shading
Cluster (USC) engine incorporating pixel shader, vertex shader and GP-GPU
(compute shader) functionality
- USC incorporates an ALU
architecture with high SIMD efficiency
- Fully virtualized memory
addressing (up to 64 GB address space), supporting unified memory
architecture
- Fine-grained task switching,
workload balancing and power management
- Advanced DMA driven operation for
minimum host CPU interaction
- 256KB System Level Cache
(SLC)
- Specialized Texture Cache Unit
(TCU)
- Compressed Texture Decoding
- Lossless data compression (PVRGC)
- The PowerVR's geometry compression, which is performed in the Geometry
Processing phase of the 3D graphics workload.
- Dedicated RISC-V processor for
firmware execution
- Separate power island for the
firmware processor for low latency power domain transitions
- On-Chip Performance, Power and
Statistics Registers.
- Resolution Support
- Frame buffer max size =
8K × 8K
- Texture max size = 8K ×
8K
- Anti-aliasing
- Performance
- Floating Point Operations
(F32) - 64 operations per clock
- Floating Point Operations
(F16) - 128 operations per clock
- Texture performance - 4
texels per clock (@32 BPP)
- Pixel performance - 4
pixel(s) per clock (@32 BPP)
- Geometry Performance -
0.25 triangles per clock
- Max performance
- ~50 GFLOPS, 4
GTex/s or 4 GPix/s @ 800MHz