AMD Radeon R9 Fury X Review: Fiji And HBM Put To The Test

Article Index

Introduction and Specifications

Last week, AMD lifted the veil in its Radeon R7 and R9 300 series of products and publicly demoed its latest flagship, the Radeon R9 Fury X, at a small venue adjacent to E3. While the Radeon R7 and R9 300 series leverages existing GPU designs that have previously been featured in Radeon 200 series products, albeit with different clocks and updated memory / board configurations, the Fury X is an altogether new kind of animal that uses some bleeding-edge technology.

We’ve got a number of details regarding AMD’s various announcements posted here and we’ve already evaluated a Radeon R9 390 card if you’d like check out some of the backstory. Today though, we can finally dive deep and tell the Radeon R9 Fury X’s whole story, complete with our own independent benchmarks.

There is a lot to cover, so grab a snack and strap in as we dish out all of the pertinent details. First up, let’s check out the specifications and then we’ll cover some of the Fury X’s standout innovations before digging into some testing...
Update 11/30/2015 – Since the initial release of the Radeon R9 Fury X, AMD has completely revamped its software suite for the graphics card. The AMD Catalyst Control Center is being retired, in favor of the new Radeon Software Crimson Edition. The Radeon Software Crimson Edition has been re-architected using QT, with an emphasis on speed and reliability, and to modernize the interface. Some game specific performance enhancements have been introduced as well.

The initial release of the Radeon Software Crimson Edition has caused some issues with fan speed states on some Radeon graphics cards, but an update is due to hit the web today to address the problem.

AMD Radeon R9 Fury X
Specifications & Features
amd fury x specifications

The Radeon R9 Fury X (and upcoming air-cooled Fury and Fury Nano) are built around AMD’s Fiji GPU. Like most GPUs from the last couple of generations, Fiji is manufactured using TSMC’s 28nm process. We should point out, however, that AMD has made some tweaks that significantly improve efficiency over older products that used the same manufacturing process. The Fiji GPU is comprised of roughly 8.9B transistors and has a die size of 596mm2. Keep in mind though, other parts of Fiji, like its 4GB of High Bandwidth Memory (HBM) and Interposer (which is used to connect the GPU to the HBM) are not accounted for in that die size. The total for the entire assembly is roughly 1011mm2.
fiji GPU
AMD Fiji GPU Block Diagram

Above is a high-level block diagram of Fiji’s inner-workings. If you’re familiar with AMD’s Tonga or Hawaii GPUs, which we’ve previously covered, this diagram should look familiar. The Fiji GPU itself is an evolution of Tonga and Hawaii, but has been beefed up considerably in a number of areas. Fiji features an updated scheduler, video engine, and the new 16-bit integer and floating point instructions, along with delta color compression technologies introduced with Tonga. Other areas of the GPU, however, are built out to be roughly 50% more capable than Hawaii. For example, Fiji features 4096 stream processors, versus 2816 in Hawaii. The memory bus width on Fiji grows to 4096-bits wide thanks to HBM, versus 512-bits on Hawaii, though cards are only packing 4GB of memory. Fiji also gets a boost in texturing capabilities thanks to its 256 texture units—Hawaii has 176. Hawaii and Fiji both have 64 ROPs, however, so it’s not an across the board upgrade for Fiji.

At its reference clocks of 1050MHz (GPU) and 500MHz (HBM), Fiji and the Radeon R9 Fury X offer peak compute performance of 8.6 TFLOPs, up to 268.8 GT/s of texture fill-rate, 67.2 GP/s of pixel fill-rate, and a whopping 512GB/s of memory bandwidth. The compute performance, memory bandwidth, and textured fill-rate are huge upgrades over Hawaii and even outpace NVIDIA’s GM200, which powers the Titan X and 980 Ti, although pixel fillrate is right in-line with Hawaii and Radeon R9 290X.
AMD Radeon R9 Fury X 4
AMD Radeon R9 Fury X Breakout

AMD was able to pack everything onto a PCB that’s not much bigger than mainstream GPUs at about 7.5”. It does, however, have a TDP that clearly falls into enthusiast territory at 275W. As such, the card requires dual 8-pin supplemental PCI Express power feeds.

To keep the entire assembly cool, AMD strapped a close-loop liquid cooler onto the Fury X that’s capable of dissipating 500W. That may sound extreme, but there’s a reason AMD went that route on this card, and it’s not because they had to. There will be air-cooled Fury and Fury Nano cards coming in a few weeks that feature fully-functional Fiji GPUs, though we’re not sure what the clocks are going to be just yet. What the high-powered liquid-cooler on the Fury X does is allow the use of an ultra-quiet fan, with the side benefit of keeping the GPU very cool (relative to most other GPUs) under both idle and load conditions. And keeping the GPU cool (think around 60’C max under load and 30’C at idle) helps reduce overall power consumption by limiting leakage current, which wastes power. Leakage increases as temperatures go up.

Let's get a closer look at the Fury X card and it's cooling assembly, next...

Related content

Comments

Show comments blog comments powered by Disqus