» Current | 2017 | 2016 | 2015 | Subscribe

Linley Newsletter

Imagination Unleashes Furian

March 28, 2017

Author: Loyd Case

PowerVR Furian is Imagination’s first wholly new GPU architecture since its 2012 design, code-named Rogue. Furian enables more scalability than its predecessor, increases throughput by implementing a new instruction set, streamlines data pipelines, offers greater layout efficiency, and supports full cache coherence. RTL has shipped to lead customers, and the first chips are expected to enter production in 1H18.

Furian will first arrive in the Series8XT GPU, giving Imagination new features and a performance boost over the current high-end Series7XT Plus. That older design topped out at 16 GPU clusters, whereas the new design supports up to 64 clusters. Furian remains a tile-based deferred-rendering (TBDR) architecture, which maintains rendering efficiency by writing only the visible pixels in a 3D scene. It also increases GPU-compute performance by beefing up the ALU blocks in the shading cluster and by incorporating a new architecture throughout.

To increase performance while keeping a similar die size, Furian introduces multiple changes relative to the previous architecture. For example, it doubles the number of primary ALU pipelines in the shader cluster without increasing the number of MAD units. This change raises throughput by increasing ALU utilization, and it allows greater flexibility when creating threads. Instead of the second MAD unit, Furian adds a smaller multiply (MUL) unit in each pipe.

Furian fully incorporates cache coherence, meaning the GPU and the main CPU can share data directly without explicit loads and stores. Series7 implemented only I/O coherence, which enabled it to snoop CPU caches. A new multistage memory-management unit (MMU) allows hardware virtualization to run in tandem with cache coherence. Furian’s cache coherence uses the standard Amba Ace protocol and supports bidirectional cache snooping, so it can work with a variety of CPU types.

Subscribers can view the full article in the Microprocessor Report.

Subscribe to the Microprocessor Report and always get the full story!

Purchase the full article

Events

Linley Processor Conference 2017
Covers processors and IP cores used in deep learning, embedded, communications, automotive, IoT, and server designs.
October 4 - 5, 2017
Hyatt Regency, Santa Clara, CA
More Events »

Newsletter

Linley Newsletter
Analysis of new developments in microprocessors and other semiconductor products
Subscribe to our Newsletter »