When AMD launched its Radeon Pro Duo a year ago, it was clearly an attempt to simultaneously drive VR development and carve out a niche for a high-end Radeon solution that could drive each eye from a separate GPU. At the same time, however, the $1,500 price tag and limited VRAM (just 4GB per GPU) were less than ideal for a four-digit development board.
AMD has announced(Opens in a new window) a new Radeon Pro Duo this week, and despite wearing the same name, it's a very different GPU under the hood. This Pro Duo is based on Polaris with 32GB of RAM in total (16GB per GPU), 2,304 stream processors, a 1243MHz engine clock, and 224GB/s of memory bandwidth per GPU. This suggests that AMD had to slightly lower the total memory clock, since the standard RX 480 offers 256GB/s of memory bandwidth per GPU.
Specs on the new Radeon Pro Duo
This GPU is derived from the Radeon Pro WX 7100, which means it has just 128 texture units instead of Polaris' 144 TMUs. Total board power is down dramatically, to just 250W total, compared with 350W for last years' dual Nano solution. There's no way the new Radeon Pro Duo would outperform last year's model -- the dual Nano solution offered nearly double the cores, twice the ROPs, and twice the texturing performance -- but none of that matters if you don't have enough VRAM on-board to handle workloads. You aren't going to see performance improvements no matter how powerful the rest of the GPU is.
AMD's cooling solution for the new Polaris-based Duo is also much simpler than the water cooling loop it used for the Nano-derived variant, and the board is probably significantly cheaper to boot. The Pro Duo was a way for AMD to sell a few more Nano boards and move some development cards, but Polaris is a better fit for the AMD hardware in most systems, and its additional RAM should make VR development easier. Polaris also adds DisplayPort 1.4 and HDMI 2.0b, both features the Nano lacked. The new GPU is expected to ship next month for $999.
With Vega supposedly shipping in the next few months, it's a little surprising to see AMD pushing a midrange dual GPU rather than waiting for a higher-end variant of its upcoming architecture. One possible explanation is that the Pro Duo is explicitly intended for developers who want to optimize per-eye VR rather than using a single GPU to render first one eye than the other. Vega may not be mature enough yet to support a dual-GPU configuration; such cards do tend to ship well after their single GPU counterparts. And AMD could be planning to refresh the Pro Duo line with either a $1,500 part based on Vega that would ship at the same price as last year's Nano-derived Pro Duo, or simply to wait another year before it introduces another dual GPU part.
