AMD Radeon Pro Duo Preview: Dual Fiji Unleashed
If you would like a refresher on the underlying technologies at work on the Radeon Pro Duo, we have a few articles we'd recommend perusing. Our Radeon R9 Fury X review covers many of specific features and capabilities of the Fuji GPU employed on this card. We've also got some detail regarding the High Bandwidth Memory (HBM) used on Fiji, and cover Asynchronous Compute as well. Though the Radeon Pro Duo is a new product in AMD's stack, and the first in the company's VR Ready Creator line, it is fundamentally similar to the R9 Fury series, since they're based on the same underlying GPU technology.
With that said, we've got all of the speeds and feeds for the Radeon Pro Duo listed for you below and will cover many of its specific features on the paged ahead...
 |
|
| Radeon Pro Duo | R9 Nano | R9 Fury X | |
| Process | 28nm | 28nm | 28nm |
| Stream Processors | 8192 (4096 x 2) | 4096 | 4096 |
| Compute Units | 128 (64 x 2) | 64 | 64 |
| Engine Clock | Up To 1GHz | Up To 1GHz | Up To 1.05GHz |
| Compute Performance | 16.38 TFLOPS | 8.19 TFLOPS | 8.6 TFLOPS |
| Texture Units | 512 (256 x 2) | 256 | 256 |
| Texture Fill-Rate | 512 GT/s | 256 GT/s | 268 GT/s |
| ROPs | 128 (64 x 2) | 64 | 64 |
| Pixel Fill-Rate | 128 GP/s | 64 GP/s | 67.2 GP/s |
| Z/Stencil | 512 | 256 | 256 |
| Memory Configuration | 8GB HBM (4GB x 2) | 4GB HBM | 4GB HBM |
| Memory Interface | 4096-bit x 2 | 4096-bit | 4096-bit |
| Memory Speed / Data Rate | 500 MHz / 1.0 Gbps | 500 MHz / 1.0 Gbps | 500 MHz / 1.0 Gbps |
| Memory Bandwidth | Up To 1024 GB/s | Up To 512 GB/s | Up To 512 GB/s |
| Power Connectors | 3 x 8-Pin | 1 x 8-Pin | 2 x 8-Pin |
| Typical Board Power | 350 W | 175 W | 275 W |
| PCIe Standard | PCIe 3.0 | PCIe 3.0 | PCIe 3.0 |
| API Support | DX12, Vulkan, Mantle | DX12, Vulkan, Mantle | DX12, Vulkan, Mantle |
| FreeSync Support | Yes | Yes | Yes |
| Virtual Super Resolution | Yes | Yes | Yes |
| Frame Rate Target Control | Yes | Yes | Yes |
As we've mentioned, the Radeon Pro Duo is built around a pair of AMD’s Fiji GPUs. Like most GPUs from the last couple of generations, Fiji is manufactured using TSMC’s 28nm process – the ASICs used here are the very same ones featured on the Fury X and Fury Nano. We've got the full list of specifications side-by-side with a couple of other high-end Radeons above for comparison.
The AMD Radeon Pro Duo, Front View
Each AMD Fiji GPU is comprised of roughly 8.9B transistors and has a die size of 596mm2. Keep in mind though, other parts of Fiji, like its 4GB of High Bandwidth Memory (HBM) and Interposer (which is used to connect the GPU to the HBM) are not accounted for in that die size. The total for the entire assembly is roughly 1011mm2 per GPU, and there are two on the Radeon Pro Duo.
The AMD Radeon Pro Duo With Liquid-Cooling Courtesy Of CoolerMaster
The Fiji GPUs used on the Radeon Pro Duo feature 4096 stream processors each, for a total of 8192 per card. The memory bus width is 4096-bits wide per GPU, and each is packing 4GB of integrated HBM memory, for a total of 8GB. The 256 texture units per GPU get doubled to 512 on the card too, as do the total number of ROPs – there are 64 per GPU, for a total of 128 on the Radeon Pro Duo.
At its reference clocks of up to 1GHz (GPU) and 500MHz (HBM), the Radeon Pro Duo offers peak compute performance of 16.38 TFLOPs, up to 512 GT/s of texture fill-rate, 128 GP/s of pixel fill-rate, and a whopping 1024GB/s of aggregate memory bandwidth – it’s essentially equivalent to a pair of Radeon R9 Nanos running in CrossFire on a single card. The compute performance, memory bandwidth, and textured fill-rate are huge upgrades over any previous-gen, single-GPU powered graphics card, thanks to the pair of GPUs used on the Pro Duo.
The AMD Radeon Pro Duo, Front View
Each AMD Fiji GPU is comprised of roughly 8.9B transistors and has a die size of 596mm2. Keep in mind though, other parts of Fiji, like its 4GB of High Bandwidth Memory (HBM) and Interposer (which is used to connect the GPU to the HBM) are not accounted for in that die size. The total for the entire assembly is roughly 1011mm2 per GPU, and there are two on the Radeon Pro Duo.
The AMD Radeon Pro Duo With Liquid-Cooling Courtesy Of CoolerMaster
The Fiji GPUs used on the Radeon Pro Duo feature 4096 stream processors each, for a total of 8192 per card. The memory bus width is 4096-bits wide per GPU, and each is packing 4GB of integrated HBM memory, for a total of 8GB. The 256 texture units per GPU get doubled to 512 on the card too, as do the total number of ROPs – there are 64 per GPU, for a total of 128 on the Radeon Pro Duo.
At its reference clocks of up to 1GHz (GPU) and 500MHz (HBM), the Radeon Pro Duo offers peak compute performance of 16.38 TFLOPs, up to 512 GT/s of texture fill-rate, 128 GP/s of pixel fill-rate, and a whopping 1024GB/s of aggregate memory bandwidth – it’s essentially equivalent to a pair of Radeon R9 Nanos running in CrossFire on a single card. The compute performance, memory bandwidth, and textured fill-rate are huge upgrades over any previous-gen, single-GPU powered graphics card, thanks to the pair of GPUs used on the Pro Duo.
