Introduction

AMD is signing off the current-generation of graphics cards with the new Radeon Pro Duo, a dual-GPU monstrosity that packs a pair of "Fiji" GPUs, in the same configuration as the company's current flagship, the Radeon R9 Fury X. In the absence of any competition from the GeForce "Maxwell" family, it could end up being the fastest graphics card money can buy for some time. Although AMD is pricing the card at the same US $1,499 it asked for the Radeon R9 295X2, it's marketing the card in a category that's between the Radeon "consumer" and the FirePro "professional" lineup. Its marketing tagline reads "for creators who game, and gamers who create."



By "creators," AMD isn't necessarily referring to content creators in general, those who use top-of-the-line FirePro cards for their top-dollar visual effects production, but VR content creators. AMD, like NVIDIA, is betting heavily on virtual reality (VR) to become a mass-medium. Unlike TVs and other screens, VR adds a new element - the user's ability to choose which part of the content to consume by simply moving their head around.

Although the concept of VR is hardly new, the prohibitive amount of computational power it takes to present live-action and gaming VR content has kept it from consumers. Sure, your Oculus Rift or HTC Vive headset has fewer pixels than a $200 monitor, but it takes a lot more computational power to make sure it responds instantaneously to your head's movements to present content like in the real world. Any latency between your input and that response could cause nausea and vertigo because beyond a certain point, your brain begins to see VR as the real world, expecting it to obey the laws of physics, and that takes a lot of GPU pixel-crunching power.

The Radeon Pro Duo currently has no competitive landscape. As we mentioned earlier, NVIDIA did not launch a dual-GPU graphics card based on its GM200 silicon, and as such, the price-cut $1,500 GeForce GTX TITAN Z is the closest competitor, though based on the older "Kepler" architecture. It could be competed with and probably even be outperformed by a pair of GeForce GTX 980 Ti graphics cards, but that's a different kind of solution. People buy dual-GPU graphics cards because it gives them the convenience of a single card and because it allows them to have four GPUs in a machine with just two PCI-Express x16 slots.

This preview is a compilation of all the publicly available information of AMD's exciting new graphics card, which makes its retail debut on April 26, 2016. It will be sold in retail channel through AMD's AIB partners, and factory-fitted in various high-end gaming desktops.
Card images courtesy Expreview. Press-deck slides courtesy VideoCardz.

The Card

The Radeon Pro Duo retains the design-language AMD introduced with the Radeon R9 Fury X and the Radeon R9 300 series (reference cards). Matte-black meets nickel-chrome, and AMD's glowing red "Radeon" logo. By picking a liquid-cooling solution for the Pro Duo, AMD saved on a ton of space that would otherwise be occupied by a bulky air-cooling solution (such as the 2.5-3 slot reference cooler on the GTX TITAN Z).


The card itself has a very solid and "monolithic" appearance to it. This comes at a cost as all that heat soaked up by the coolant has to be expunged somewhere, and that's where the rest of the solution kicks in - a thick 120 mm radiator with a factory-fitted high-end fan and tubing that connects the card to it. The Radeon Pro Duo appears to have slightly longer coolant tubing than found on the R9 Fury X. This is because its fittings on the card are toward the front rather than the rear, like on the R9 Fury X. The additional tubing should give you increased flexibility for placing the radiator at just the right vent.

The Pro Duo is a long card (27 cm long), although it's shorter than some of its immediate predecessors, the R9 295X2 and HD 6990 (both upwards of 30 cm). This is one of the key dividends of AMD's decision to go with a stacked HBM multi-chip module for the "Fiji" silicon. AMD is spending crucial PCB real-estate it saved on a more meaty VRM, with its hot components (IR DirectFETs) better spread out.


The Cooler Master-sourced cooling solution consists of two AIO pump-blocks for the GPUs, connected to the coolant loop in serial; with an additional heatsink that transfers heat to the blocks, which cools the VRM and PCIe bridge chip.


On the PCB, you can find a 4+2 phase VRM system per GPU, which draws power from a whopping three 8-pin PCIe power connectors. Along with the slot power, this card is capable of drawing up to 525W of power. This isn't necessarily the card's rated power draw.


The Radeon Pro Duo features two 28 nm "Fiji" GPUs with all 4096 stream processors enabled on each chip. Their core clock speed is set at "up to" 1000 MHz, and memory sits at 500 MHz. The R9 Nano had its core clocked at up to 1000 MHz as well, but performed vastly different from the R9 Fury X, which was clocked just 50 MHz higher. Given this card's meaty VRM and cooling, we're inclined to believe that AMD could give these chips a more laxed power-management tuning.

Performance Claims by AMD

Since this is a preview, we can't put out numbers yet. AMD has, however, made its own performance claims for the Radeon Pro Duo, tested on a typical high-end desktop (HEDT) machine.


At the outset, AMD claims the Radeon Pro Duo to be 1.5X faster than the GeForce GTX TITAN X and 1.3X faster than the outgoing Radeon R9 295X2. This becomes a recurring theme for the way the Pro Duo performs versus the two cards in popular games that can take advantage of AMD CrossFire.


AMD is also claiming the Radeon Pro Duo to be faster than two GTX TITAN X cards in SLI under "ideal conditions" (4K Ultra HD "Ashes of the Singularity," with DirectX 12 Async Compute, a feature NVIDIA chips currently lack).

The Radeon Pro Duo scores the highest in SteamVR performance assessment.