AMD announces Radeon Pro Duo graphics card
March 15, 2016 // 10:54 a.m.
AMD has nailed its colours to the mast in the battle for a slice of the virtual reality market with the announcement of the Radeon Pro Duo graphics card, boasting of having made the 'fastest gaming card in the world.'
Designed both for gamers who demand the highest possible performance and for content creators working in virtual reality and ultra-high-definition, the Radeon Pro Duo is a beast: AMD claims the card is capable of 16 teraflops of single-precision compute performance, more than double that of rival Nvidia's Titan X. A quick look at the name of the card reveals the trick, though: while it's a single board, it hosts two GPUs each the equivalent of AMD's Fury X card.
The Radeon Pro Duo will be AMD's top-end card for use with the LiquidVR API, though it won't be the only device to support it: the already-released R9 290 family, R9 390 family, R9 Nano, and R9 Fury boards all feature LiquidVR support for low-latency VR rendering on compatible headsets - of which both the Oculus Rift and HTC Vive have been confirmed as compatible - albeit at lower performance than its flagship card.
Given the presence of two top-end GPUs, it's no surprise that AMD has once again turned to an all-in-one sealed-loop liquid cooler to keep the heat of the Radeon Pro Duo in check. With a confirmed typical 350W thermal design profile (TDP) but power connectors for up to 500W draw, that's more of a necessity than a bonus. Specifications for the Radeon Pro Duo are, unsurprisingly, basically a doubling of the R9 Fury X with a drop from 1,050MHz boost clock to 1,000MHz: there are two GPUs each with 4,096 stream processors, 256 texture units, and 64 ROPS, while each has access to its own bank of 4GB high-bandwidth memory (HBM) running at 1Gb/s across a pair of 4,096-bit buses.
The price, however, has more than doubled: AMD has confirmed that when the device launches this summer it will fetch a premium figure of $1,499 - matching the launch price of the last-generation Radeon R9 295X2 while offering significantly improved performance at a lower TDP.
At the same event, AMD outlined its GPU technology roadmap: Polaris, its 14nm fourth-generation Graphics Core Next architecture, is scheduled for mid-2016 with a claimed 2.5-fold increase in performance per watt; Vega will follow in early 2017 with a switch to HBM2 memory; then a year later Navi will launch with improved scalability using what AMD has mysteriously called 'nexgen memory.'