The AMD Radeon Pro Duo
Back in June of 2015, AMD held a launch event for their Fiji architecture and one of the products they teased was an in-development dual Fiji card. At the time of the announcement, the expected launch date was the end of 2015 and it was to become the company’s flagship gaming card. Fast forward to December 2015, at AMD’s Polaris event, and it’s been announced that the card will be pushed back to 2016 and realigned for the VR market.
Besides the VR market (which admittedly is very niche), the AMD Radeon Pro Duo is also positioned to be a content creator’s card, as opposed to be a gaming one. Although it is certainly more than capable of performing as one – the original pitch for the card, after all, was “the world’s fastest graphics card.”
Specs: How Fast Can it Really Go?
It’s known that VR requires a fairly beefy setup, so it will not be surprising that the AMD Radeon Pro Duo is built for power and speed, the question is just how fast and how powerful it is. The official numbers from AMD is that the card is capable of over 16 TerraFLOPS of performance, which is basically the equivalent of two Fiji GPUs clocked at 1Ghz each, making the Radeon Pro Duo the equivalent of two Fury Xs in a Crossfire setup.
Another indicator that this is basically two Fiji cards in one is the 64 ROPs and 256 texture units per GPU, along with 4 gigs of HBM per GPU @ 1 Gbps making for an effective memory bandwidth of 512 Gb per sec per GPU. It helps to note that 4 gigs per GPU do not necessarily mean that this card has 8 Gb of HBM, much in the same way that crossfiring two 4 Gb cards won’t give you 8. It’s still effectively 4 Gb, but with double the bandwidth.
Where this becomes useful is with VR applications, as the display for each eye can be rendered separately per card on a crossfire setting. Technically, this means devs can give each side 4 Gb on its own.
Another important concern here is that if the Radeon Pro Duo is essentially two crossfired Fiji cards @ $1499 MSRP, why not just buy two Fury Xes and use it in a crossfire configuration? A Fury X is $649 MSRP so buying two will still save you a whopping $200. There are a lot of reasons, but the easiest to explain for now is power efficiency.
It’s true that power efficiency is not much of a concern for people who would spend fifteen hundred bucks on a video card, but a cut down power consumption does allow the Radeon Pro Duo to ultimately pay for the premium through reduced power costs. It will take some time, but it will happen eventually.
As for actual numbers on power consumption – the card can be misleading because it requires 8 pin PCIe sockets. Coupled with the power draw coming from the PCIe socket, all signs point towards the card being able to draw as much as 525W.
But according to AMD, the extra pin requirement is just leeway for overclockers. The true TBP of the Pro Duo is much, much lower: 350W. Again, it may not matter much to the target market, but it most likely will allow OEM builders to be more conservative with their builds instead of trying to overshoot the PSU requirements.
Cooling Requirements and Ports
You don’t get to produce 16 Terraflops without producing tremendous amounts of heat, so it’s no surprise that AMD can’t go with passive or even simple air cooling. The Radeon Pro Duo boasts of a closed loop liquid cooler setup with a single 120mm radiator that is similar to the one used by the Fury X. Since there is no fan used, the Pro Duo will be quieter than even the 295×2 or nVidia’s Titan Z.
One has to assume that it also means a longer lifespan for the cooling solution due to the fewer number of moving parts. For its display I/O, there are 3 DisplayPorts and a single HDMI port, which is important because both the Oculus Rift and HTC Vive VR systems require an HDMI port.
While the pricing makes it less enticing than a pair of Fury Xes (or a pair of Nanos), one has to remember that the extra $200 you’re going to shell out does give you a smaller overall card and power consumption, so it can be a decent compromise if your casing is somewhat short on space (though one should never dream of using this card on a mini ATX) or if your PSU may struggle with two Fury Xes.
Ultimately, at the high end of the market, it can be argued that $200 might not be a big deal and that the convenience brought by not having to deal with two separate cards may be worth it – but the rest of us should either wait for future cards (either AMD’s next one or Nvidia’s upcoming offering) or settle with cross firing a couple of Fury Xes.