Gen 9 Graphics
Lossless Render Target Compression
• Compress data before sending to memory
• Saves bandwidth and DDR power
• Cache line granularity
• Peak compression ratio: 2:1
• Performance improvement: 3-11%
• Supported all SKUs starting Gen9
Compute Capability Enhancements:
• Shared Virtual Memory (SVM) improvements over Broadwell
- Improved cache coherency performance
• Larger L3 cache per slice
• Additional atomic operations
- 32-bit float, min, max, compare and exchange
• Support for smaller thread groups
- Barriers, shared local memory (SLM)
• Improvements in preemption granularity
Low power enhancements
- Standalone Fixed-Function media in Unslice
- Reduced memory bandwidth
• New Intel® Quick-Sync Video mode
- Fixed function encoder designed for low power
and low latency real-time applications
• New codec decode and encode support
- Support for HEVC, VP8, MJPEG
• New RAW imaging capabilities
• Broad enabling of applications
- Support for DirectX* 12 and OpenCL™ 2.0, extends CPU/GPU programmability
Summary and Next Steps
• Another large step function in scalability and performance
- Gen9 GT4 1.5x larger than Gen8 GT3
• Many 3D/Compute performance and visual features
- Throughput increases, DirectX* 12 features, compression, preemption, etc.
• New low power media support
- Fixed-Function decoders and encoder
• Media quality improvements and 4K camera RAW support
• Multiple high-resolution display pipes
- Additional planes per pipe