Gigabyte Radeon RX 460 WindForce 2X OC 2GB Review

Mere days after the release of its latest graphics card, the RX 470, AMD is back with another launch that brings its Polaris architecture to the entry-level. AMD's pitch is that the RX 460 is designed to give ageing systems an upgrade to a cool and efficient card capable of >90fps 1080p gameplay in popular esports titles and of future-proofing them as well with a slew of modern display capabilities and features. Whereas the RX 470 was based on a cut-down version of the Polaris 10 GPU that powers the RX 480, the RX 460 is using a new, significantly smaller chip known as Polaris 11.
The 
die size is a measly 123mm², which is almost half the size of a GTX 1060 or RX 480 GPU. Even so, AMD manages to cram in 3 billion transistors, and you can get an idea of how these are arranged in the block diagram below.

This is a far smaller implementation of GCN than we're used to seeing, although the design principles and arrangement are all familiar. At the front end there's the usual Graphics Command Processor and pair of Hardware Schedulers, as well as four Asynchronous Compute Engines – until now we've been used to eight here. These are used to dispatch graphics and compute tasks in parallel and we've seen them give AMD an edge in DirectX 12 applications as a result. Polaris brings with it extra asynchronous compute functions, but these will be of limited benefit to a card of this calibre – the focus clearly isn't on maximising performance in the latest DX12 titles.The main pipeline is still arranged into Shader Engines, each with its own geometry processor, rasteriser, Compute Units and Render Backend units. For Polaris, the geometry processor has had some tweaks to better handle heavily tessellated scenes, but the main hardware blocks are unchanged from what we've seen before. In this GPU there are eight Compute Units per SE, but for the RX 460 two of these are disabled leaving it with just 14 in total, which is less than half the count found in the RX 470. This leaves it with a stream processor count of 896 and 56 texture units. Four of AMD's Render Backend units mean we're looking at a card with just 16 ROPs – a card for high resolution gaming this is not. The GPU is set at a base clock of 1,090MHz and a boost clock of 1,200MHz.

In terms of memory, the card has a 1MB internal L2 cache. As for the external GDDR5, there will be two designs: one 2GB and one 4GB. In both instances this will be clocked at 1.75GHz (7Gbps effective), netting us a total memory bandwidth of 112GB/sec over the 128-bit interface. This is only the raw bandwidth – Polaris benefits from AMD's latest delta colour compression algorithms, allowing this card to make better use of the available bandwidth than previous generation products. 2GB of memory feels like a bit of squeeze these days, even at 1080p, but it's likely enough to cope with the low demands of esports titles.