AMD’s new Polaris-based Radeon Pro Duo will compete against the NVIDIA GeForce GTX Titan Xp
AMD’s new Polaris-based Radeon Pro Duo will compete against the NVIDIA GeForce GTX Titan Xp
AMD has announced a new version of the Radeon Pro Duo at the National Association of Broadcasters show in Las Vegas. This is the latest iteration of the dual-GPU workstation card it unveiled around a year ago, and it caters to professionals rather than gamers.
However, the surprising part is that the new Radeon Pro Duo is actually slower than its predecessor. It packs two first-generation (codename Ellesmere) Polaris GPUs and offers 11.45 TFLOPs of single-precision compute performance, compared to 16 TFLOPs from before. And while the 2016 model featured stacked High Bandwidth Memory (HBM), the 2017 card goes for GDDR5 memory, albeit a whopping 32GB of it.
The Polaris architecture is found on AMD’s mid-range Radeon RX 400- and 500-series cards, which are generally known more for power efficiency than raw performance. As a result, TDP has actually dropped from 350W to 250W on the new Radeon Pro Duo, as has the number of processing cores.
As a result, the card requires just one 8-pin and one 6-pin connector, instead of the three 8-pin connectors the Fiji card required.
That said, the new Radeon Pro Duo is no pushover by any means, and it looks set to go up against the NVIDIA GeForce GTX Titan Xp, which boasts a slightly higher 12 TFLOPs of computational power.
The table below summarizes the available specifications of the Polaris-based Radeon Pro Duo:
AMD Radeon Pro Duo (2017) | AMD Radeon Pro Duo (2016) | NVIDIA GeForce GTX Titan Xp | |
---|---|---|---|
GPU | Polaris 10 | Fiji | GP102 |
Process | 14nm | 28nm | 16nm |
Transistors | - | 8.9 billion x2 | 12 billion |
Stream processors | 2,304 x2 | 4,096 x2 | 3,840 |
Texture units | 128 x2 | 256 x2 | 240 |
Render output units | 32 x2 | 64 x2 | 96 |
Base clock | - | - | 1,480MHz |
Boost clock | 1,243MHz | 1,000MHz | 1,582MHz |
Memory | 32GB GDDR5 (16GB per GPU) | 4GB HBM x2 | 12GB GDDR5X |
Memory clock speed | 7,000MHz | 1,000MHz | 11,400MHz |
Memory bus width | 256-bit x2 | 4,096-bit x2 | 384-bit |
Memory bandwidth | 448GB/s | 512GB/s | 547.7GB/s |
TDP | 250W | 350W | 250W |
Launch price | US$999 | US$1,499 | US$1,200 |
The Radeon Pro Duo is targeted at creative professionals, and it’s not entirely clear why AMD chose to cut performance from the Fiji-based card seeing as this is a crowd that will definitely appreciate an extra dollop of horsepower. Still, its US$999 price tag is US$500 cheaper than what the older Radeon Pro Duo launched at, so it could be an attempt at offering a more affordable solution.
The new card is actually based on the Radeon Pro WX 7100, the flagship of AMD’s Radeon Pro WX series. The Radeon Pro Duo features two WX 7100 GPUs, which goes a way toward doubling its performance.
One of the primary benefits of the dual-GPU card is its support for parallel computing, which means professionals can utilize two software packages at the same time. This means dedicating separate functions to each GPU simultaneously, and according to AMD, you can handle multiple 4K video streams in real-time, or even perform live content creation with the first GPU, and real-time rendering or ray tracing on the second.
AMD says that professional workflows are becoming more and more complex, with creators working with multiple applications at once and switching between them frequently. The company is thus positioning the Radeon Pro Duo as a solution to these users, claiming superior multi-tasking performance and the ability to dedicate GPU resources where they are needed.
There’s support for the highest resolution displays as well, to the tune of a single 8K display, or four 4K displays at 60Hz.
AMD has set availability for the end of May.
Source: AMD