AMD Showcases Graphics, Energy Efficient Computing and Die-Stacking Innovation at Prestigious Hot Chips 27 Conference - Seite 2

Die-Stacking Journey Culminates In High-End GPU
The path to the new "Fiji" family of AMD Radeon R9 Fury GPUs began with exploring the best die-stacking option for bringing large amounts of memory into the same chip package with the GPU while dramatically increasing the memory bandwidth available to a high-performance graphics engine, without increasing power consumption. Working with memory partner SK Hynix, the new GPUs based on AMD Graphics Core Next (GCN) architecture offer up to 4 GB of high-bandwidth memory (HBM) over a 4096-bit interface to achieve an unprecedented 512 Gb/s memory bandwidth. The new memory is stacked close to the GPU in the package by implementing the first high-volume interposer as well as the first through-silicon vias (TSVs) and micro-bumps in the graphics industry. HBM and the interposer provide 60 percent more bandwidth than previous generation GDDR5 memory³ and 4x the performance-per-watt of GDDR5.⁴ At the same time, the "Fiji" family is capable of up to 8.6 TFLOPS performance, a nearly 35 percent increase over the previous generation (Radeon™ R9 290 Series GPUs).The result is an improvement of up to 1.5x performance-per-watt.

APU Energy Efficiency
The new 6th Generation AMD A-Series APU harnesses up to 12 compute cores (4 CPU + 8 GPU)* leveraging AMD "Excavator" CPU cores and the third generation of AMD GCN architecture. The result is a groundbreaking processor that boasts more than twice the battery life of its predecessor⁵ and up to 2x faster gaming performance than competitive processors.⁶ The processor incorporates a number of genuine power management firsts for AMD, including the first implementation of Adaptive Voltage and Frequency Scaling (AVFS), the first High Efficiency Video Compression (HEVC)/H.265 decoder on-chip for notebook, and the first APU to utilize color compression and bandwidth-saving technologies. In addition, AMD will share details on clock and power gating management, disclosures on the new integrated Southbridge, and improvements in entering sleep and idle modes. AMD will also discuss numerous additional efficiency improvements, including the enhanced ability to quickly put the SoC in the lowest power state and self-refresh DRAM.