Ever since AMD announced Fiji and its new memory interface, there’s been one major question on the minds of enthusiasts — could this new memory architecture
give Team Red the performance it needed to compete with Maxwell? Fiji, after all, is based on the R9 285 (aka Tonga), a modest update to AMD’s 2013 Hawaii GPU. The entire product family is 3.5 years old at this point. While GCN more than held its own against Kepler, Maxwell’s debut swept AMD’s performance
and power efficiency.
We weren’t supposed to run these numbers publicly until the NDA lifts on the Radeon R9 Fury X, but evidently that message didn’t get out to everyone, so AMD has lifted the embargo on its own performance projections — and thus, we find ourselves returning to the waters Fiji again to discuss the card — hopefully for the last time before we can bring you a full review.
As the data shows, the Radeon R9 Fury X is neck-and-neck with the GTX 980 Ti across a wide spectrum of titles. This fits our general expectations for both cards, given their respective configurations and the fact that Fiji is a giant-sized implementation of the already well-known GCN architecture.
Can you trust vendor-provided benchmarks?
This brings up an interesting aspect of benchmarking that I don’t often get to write about (mostly because I don’t want to bore you to tears). Any time a vendor runs benchmarks, there’s always the question of whether or not the results are accurate or not. Generally speaking, the results that AMD, Nvidia, or Intel claim for their respective hardware isaccurate, by which I mean that if you take their system configuration and test it with their settings, you’ll see the same results with some modest variation.
The devil, in these cases, is in the details. Small shifts in detail levels and settings can significantly tilt competitive comparisons, often by 10-15% or more. The question for any given vendor comparison isn’t “Will I see these same results,” but “What settings were used to generate the data?” Since the point of a Reviewer’s Guide is to actually present useful information, the results are typically ballpark-accurate — it’s easy to tell when any company is gaming the system by using odd settings, since our own benchmark data will show a very different spread between two different cards.
The Radeon R9 Fury X’s performance, meanwhile, is just one part of the equation. Power consumption and noise also matter, and these are areas where AMD has promised some truly significant improvements. We’ll know soon if those achievements arrived on schedule with better benchmark figures.