AMD's formal unveiling of the beastly new Radeon R9 Fury X at E3 earlier this revealed a lot about the graphics card, but several technical details were left glaringly undetailed. Today, AMD's taking the wraps off the rest of the information, giving us a full profile of its impressive new $650 flagship--a flagship where just as much care was spent on aesthetics as on raw technological firepower.
The rest of AMD's new Radeon R300 series cards are hitting the streets today, which we covered in a separate post.
We'll go through it all in detail, but let's kick things off with the premier feature: The Fury X's revolutionary high-bandwidth memory.
Traditional dies for GDDR5 DRAM need to be arrayed on the board around the graphics processor, which sucks up a ton of space on the card. HBM is a new technology that stacks DRAM vertically instead, connecting the dies and the GPU via interposers. You can read all about HBM here, but in a nutshell, it requires far less room on the graphics card and also delivers a ton of memory bandwidth, by pairing low clock speeds with a ridiculously wide memory interface.
Specifically, the HBM in the Radeon R9 Fury X is clocked at a mere 1Gbps. That may seem paltry when compared to the 7Gbps speeds standard to the traditional GDDR5 memory in Nvidia's flagship graphics cards. But Nvidia's GDDR5 memory travels over a 384-bit-wide interface, which the Fury X's 4GB of HBM utilizes a 4,096-bit bus. Yes, you read that correctly. That combination gives the Fury X 512GBps of total memory bandwidth, compared to the ferocious GTX 980 Ti's 336.5GBps.
HBM's drastically reduced footprint also lets AMD pack a ton of tech into its new Fiji GPU--literally. Fiji rocks 4,096 stream processors and 8.9 billion transistors, compared to the older R9 290X's 2,816 stream processors and 6.3 billion transistors. (Nvidia's Titan X packs 8 billion.) Clocked at up to 1,050MHz, it's able to pump out up to 8.6 teraflops of compute performance. You can see the full tech specs for the Fury X's HBM and Fiji processor in the chart at right.
All that power needs a pair of 8-pin connectors and 275 watts from the wall under heavy gaming scenarios, which is similar to the 980 Ti's needs. (Nvidia's card asks for 250W).
So now for the elephant in the room: How does the Fury X compare against Nvidia's similarly priced GeForce GTX 980 Ti? It's impossible to tell until we've put the Radeon through its review paces, given that AMD's stream processors and Nvidia's CUDA core technology are directly comparable, and HBM adds an unknown factor. But these AMD-supplied benchmarks--which were obviously chosen to place the Radeon in the best possible light--show the two cards performing fairly neck-and-neck in most games.
Sign up for MIS Asia eNewsletters.