AMD today announced the preliminary details of the upcoming Vega GPU architecture. After more than five years of design and development, Vega architecture to meet the traditional GPU can not effectively for PC games, professional design and machine intelligence to create new possibilities. As data-intensive workloads become the new mainstream, the parallel nature of the GPU itself is an effective solution to this problem. However, dealing with these large and new datasets requires fast access to large datasets, and the revolutionary memory subsystem of the Vega architecture allows the GPU to handle large data sets of different memory types. Vega architecture GPU high-bandwidth cache controller can be fine-tuned data movement, in a flexible and self-set way to access the processor internal cache and external memory.
Raja Koduri, Senior Vice President and Chief Architect, AMD Radeon Graphics Technology Group, said he was pleased that the GPU was being used to solve gigabyte-sized gaming data problems and exabyte-scale machine intelligence data. AMD designed Vega architecture is to extend this ability, while adding more flexibility to solve the GPU in the next five years, faced with different types of problems. AMD's key high-bandwidth caches have the potential to impact the entire GPU market. xc9536-5pcg44c
The Vega GPU architecture's advanced focus includes:
The world's most advanced GPU memory architecture: Vega architecture for the GPU to create a new memory hierarchy, the new subversive design using a new high-bandwidth cache and controller. Cache with cutting-edge HBM2 technology, can be transmitted per terabytes of data per second, the bandwidth of each pin is the previous generation HBM technology 2 times. In addition, HBM2 capacity increased significantly, the area is less than half of GDDR5 memory. The Vega architecture is designed to support a wide variety of memories, up to 512TB of virtual address space, for streaming huge data sets.
New Generation Geometry Channels: With the increased resolution of data capture devices, today's games and professional applications can handle extremely complex geometric data. Hundreds of millions of polygons have dense meshes in any picture, usually with multiple polygons per pixel being rendered. Vega's next-generation geometry allows programmers to handle complex geometries with superior efficiency, with more than 200% more throughput per clock than previous generation Radeon architectures. In addition, it has improved load balancing mechanism and intelligent workload allocator to provide consistent performance.
A new generation of computing engines: Vega architecture is the core of a new next-generation computing engine, using a variety of flexible computing unit, in the high-generation mode each clock cycle can handle 8-bit, 16-bit, 32-bit or 64-bit $ Job Note 2. These units are optimized for clock frequencies that are significantly higher than previous-generation units, and support a variety of data types, allowing the architecture to be highly flexible for handling different workloads. tisp3180f3sl
Advanced pixel engine: The new Vega pixel engine uses Draw Stream Binning Rasterizer rasterizer, designed to improve efficiency and energy efficiency. It supports pixel-fetching once, coloring once, using a smart chip to cache the cache cache, and early elimination of the final scene hidden pixels. The Vega pixel engine is now a built-in L2 cache slave that significantly reduces the amount of time it takes to read and write frequently as it processes drawing workloads.
GPU-based products based on Vega architecture is expected to launch in the first half of 2017.
- baitron's blog
- Log in or register to post comments