Update – We’ve seen many reports and rumination surrounding the power consumption and other results of our review. We can only report what we observed therefore we’ll investigate the discrepancies diligently and report any and all findings to AMD.
Polaris marks the fourth iteration of AMD’s Graphics Core Next architecture. It was first seen in 2011 with the release of the 7xxx series of GPUs. It was a massive change in direction that enabled for a far more general purpose design that also carried over greatly into the world of gaming. The initial inception of the idea ran a bit hot and also consumed quite a bit of power, making it an interesting choice when performance per watt is taken into consideration, as it increasingly is. Hawaii took performance to another, even more competitive level that’s remained so even today. The architecture was still quite power-hungry, however. Fiji was an attempt to curtail some of those steep power requirements and also introduced a memory technology that, while not immediately successful, is paving the way for the introduction of faster memory technologies in the future. Performance didn’t quite scale linearly compared to Hawaii, as we’d all hoped, and power consumption was improved though not necessarily by much.
Polaris is a return to the small die strategy, but is it enough?
It is faster and newer drivers keep edging it ahead in performance, though it’s been clear to the new team at AMD, the Radeon Technology Group, that something has to change. The architecture needs some serious improvements internally, not to mention it’s more than time for a change in node process as well.
AMD RX 480
|AMD RX 480||AMD R9 390X|
|VRAM||4GB/8GB GDDR5||8GB GDDR5|
|Memory Bus Width||256-bit||512-bit|
We’ve heard time and again about the improvements that have been made to the Polaris architecture when compared to the previous generation of of Graphics Core Next. We’re expected to see a 15% improvement in shader efficiency per compute unit, a not inconsequential improvement. They’ve also said 1.7x performance per watt just with the implementation of 14nm FinFET and up to 2.8x performance per watt additionally as a result of the improvements that they’ve baked into the new GCN architecture.
If you’re looking for game benchmarks, just head on over to page 3. If you want to see how it performs in other compute environments, including a novel Deep Neural Network benchmark, then head on over to the next page. There we’ll put the RX 480 through its paces. And we’ll see if those performance enhancements carry over to GPGPU tasks.