AMD Vega GPU presentation leaks

AMD Vega presentation leaks

AMD Vega presentation leaks

 
Slides for AMD’s Vega presentation have been leaked online, promising some huge architectural improvements including enhanced clock speeds and more performance per clock than previous AMD GPU architectures. 
 
To start off AMD has worked to improve Vega’s memory architecture, creating a new high bandwidth cache and making render back ends like AMD’s Pixel Engine clients of the GPUs L2 cache. These changes to AMD’s memory architecture are designed to allow data to be move throughout the GPU quickly and with lower latencies than before, allowing Vega to make better use of its internal resources. 
 
With Vega, AMD will also be moving to HBM2 memory, which has a 50% smaller footprint than GDDR5 based solutions and offers a huge increase in bandwidth when compared to both GDDR5 and HBM1.  
 
A lot of these changes in memory architecture will help AMD in the compute world, with professional data sets becoming increasingly large requiring larger and larger pools of VRAM and system memory to work with. 

  

AMD Vega presentation leaks  AMD Vega presentation leaks
AMD Vega presentation leaks  AMD Vega presentation leaks
AMD Vega presentation leaks  AMD Vega presentation leaks
  

In the gaming world, some of the most significant changes to Vega are in its Compute Engines/Geometry Pipeline, with AMD designing a whole new NCU (Next Generation Compute Unit) which is optimised for enhanced clocks speeds and higher numbers of operations per clock. 

Not only does AMD promise higher GPU IPC with Vega, but also higher clock speeds when compared to AMD’s older architectures (GCN/Polaris), which means that Vega should offer a significant increase in GPU performance per Compute unit. 

On the compute side, AMD’s NCU will also be capable of calculating 8-bit, 16-bit and 32-bit operations with perfect scaling, which is perfect for those that require higher levels of lower precision compute performance. This also opens up the option to do several of these varying precision levels of compute at the same time, with mixed precision compute capabilities. 

AMD’s new NCU offers much higher peak levels of geometry throughput, with AMD’s R9 Fury X offering 4 Geometry engines with  peak throughput of 4 polygons per clock, with Vega coming with 4 geometry engined that can handle up to 11 polygons per clock. a 2.6x increase.  

 

AMD Vega presentation leaks  AMD Vega presentation leaks
AMD Vega presentation leaks  AMD Vega presentation leaks
AMD Vega presentation leaks  

 

AMD has also been working to improve their GPU architecture by lowering bandwidth consumption through texture/colour compression and other technologies as well as eliminate needless GPU work using their new Draw Stream Binning Rasterizer. 

This will enhance performance and power consumption by allowing Vega to deliver more with less memory bandwidth and not do some GPU calculations which have no effect on the final image. This ensures that every bit of GPU performance is used to its fullest potential with Vega. 

 

AMD Vega presentation leaks  AMD Vega presentation leaksAMD Vega presentation leaks  

 

Vega is looking like a very interesting GPU architecture, offering some significant improvements over GCN and even their Polaris Architectures. 

Hopefully, we will hear more at AMD’s Vega event at CES, which is expected to start at 2pm GMT. We will offer a more comprehensive overview of AMD’s Vega architecture as more information becomes available. 

 

You can join the discussion on AMD’s Vega GPU architecture on the OC3D Forums. 

Â