AMD Radeon Pro Duo Polaris

The AMD Radeon Pro Duo Polaris is a dual-slot Radeon Pro GCN generation card built on the Ellesmere chip, using GCN 4.0 architecture on GlobalFoundries’ 14 nm process. Its launch MSRP was 999 USD. The database records no benchmark score for this part, leaving its percentile rank at 50 among all GPUs. Because no nearest rivals are attached to the entry, the analysis below is drawn from the listed specification set rather than measured competitor deltas.

Benchmark Performance

The average benchmark score shown for this card is 0, and the nearestRivals list is empty, so no percentage delta against a specific competitor can be derived from the record. What the record does contain is a complete compute specification: FP32 throughput is 5.728 TFLOPS, and FP16 throughput is also 5.728 TFLOPS on a 1:1 basis. The card has 2304 shading units, 144 texture mapping units, and 32 render output units. Texture fill rate is 179.0 GTexel/s, while pixel fill rate is 39.78 GPixel/s.

The 1:1 FP16 figure is a notable data point: the listed half-precision rate is identical to the single-precision rate, with no separate or reduced FP16 path recorded. The relationship between texture and pixel rates also stands out. At 179.0 GTexel/s, texture throughput is substantially higher than 39.78 GPixel/s pixel throughput, and the 144 TMUs outnumber the 32 ROPs by a wide margin. That points toward a workload profile where texture-heavy shader work is better serviced than raw pixel-pushing.

With no measured benchmark result, the only global performance anchor is the percentile rank. The 50th percentile among all GPUs places this model at the median of the database’s distribution. That is a neutral position: there is no evidence in the data of a leading or trailing score relative to named alternatives. The specification sheet suggests a card that offers solid compute rates and texture throughput, but without a score, any claim of being faster or slower than a direct rival would be unsupported.

Power and Cooling

Power requirements are explicit in the record. The card has a 250 W TDP, and the suggested power supply is 600 W. Power delivery uses one 6-pin and one 8-pin PCIe power connector, a concrete installation requirement. The board is dual-slot, with a length of 305 mm (12 inches) and a height of 111 mm (4.4 inches). These dimensions define the physical clearance needed in a chassis. The bus interface is PCIe 3.0 x16.

Cooling details beyond slot width are not listed; the data does not include a specific cooler model or cooling capacity. What is available is the dual-slot designation, which indicates the card occupies two expansion slots. The power and physical specification set is therefore clear: a 600 W power supply, one 6-pin and one 8-pin connector, a two-slot footprint, and a 305 mm board length. The 14 nm GlobalFoundries process provides the fabrication context for the 250 W figure. The die measures 232 mm² and contains 5,700 million transistors, yielding a transistor density of 24.6M per square millimeter.

How It Compares

The nearestRivals array in the record is empty. There are no competitor names, no comparative scores, and no deltaPct values to analyze. The only broad positional data is the percentile field: percentileVsAllGpus is 50, which places this card at the median of all GPUs in the database. That rank means the aggregate database metric places half of the tracked GPUs above it and half below it.

In the product lineage, the predecessor is listed as FirePro GCN and the successor as Radeon Pro Polaris, situating this model between those two professional GCN-era entries. The production status is listed as end-of-life, and the release date is 2017-04-23. Without any nearestRivals entries, a per-rival comparison cannot be constructed from the data, so the 50th percentile and the predecessor/successor relation are the only comparative facts available.

FAQ

Q: What is the memory configuration of the AMD Radeon Pro Duo Polaris?

A: The card has 16 GB of GDDR5 memory on a 256-bit bus, with 224.0 GB/s of bandwidth. The memory clock is 1750 MHz, listed as 7 Gbps effective.

Q: What power supply and connectors does it require?

A: The TDP is 250 W, and the suggested power supply is 600 W. The card requires one 6-pin and one 8-pin PCIe power connector.

Q: What display outputs does it provide?

A: The card provides one HDMI 2.0b output and three DisplayPort 1.4a outputs.

Q: Which graphics APIs are supported?

A: The listed API support includes DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.3.

Q: Does the record list dedicated ray tracing or tensor cores?

A: No. The RT cores and tensor cores fields are both null, so no dedicated ray tracing core count or tensor core count is present in the specification record.

Q: What is the FP16 throughput compared to FP32?

A: FP16 throughput is 5.728 TFLOPS, and FP32 throughput is also 5.728 TFLOPS, giving a 1:1 ratio.

Who Should Consider It

Because no benchmark score is recorded, this guidance is based on the memory and compute specifications rather than competitive measurements. The standout feature is 16 GB of GDDR5 memory. Workloads that keep large textures, large frame buffers, or large data sets on the card are the ones that can use that capacity. In contrast, tasks with modest memory footprints would not engage the full buffer resource.

The compute and texturing figures also define the candidate workload. The 5.728 TFLOPS FP32 and FP16 rates provide shader and compute throughput, while 144 TMUs and 179.0 GTexel/s support texture-heavy rendering. Users whose scenes rely on detailed texture sampling or compute ratios benefit from those resources. With 32 ROPs and 39.78 GPixel/s, pixel output is less emphasized, so the card is better suited to setups with substantial shading and texturing work rather than pure fill-rate-bound output.

The 16 GB capacity is the main reason to consider this card at high resolutions or with settings that increase memory pressure. The data does not list specific resolutions, but the combination of a large frame buffer and 224.0 GB/s of bandwidth is the relevant basis for those scenarios. Because the production status is end-of-life and no score exists in the database, this recommendation is conditional on the workload matching the memory and compute profile.

Memory Subsystem

The memory subsystem is defined by four numbers: 16 GB of GDDR5, a 256-bit bus, 224.0 GB/s of bandwidth, and a 1750 MHz memory clock with 7 Gbps effective data rate. The 256-bit bus is the path between VRAM and the 2304 shading units, and 224.0 GB/s is the aggregate data rate available to feed those units.

High-resolution rendering tends to increase both capacity pressure and bandwidth demand. In the data, capacity is the larger strength: 16 GB is a large frame buffer, while 224.0 GB/s is the bandwidth figure that must sustain transfers. Texture fill rate is 179.0 GTexel/s, and pixel fill rate is 39.78 GPixel/s, so the memory system is matched to a card with more texture work than pixel output. The effective 7 Gbps memory rate is the listed speed for the GDDR5 modules, and no memory overclocking or additional memory features are recorded.

Ray Tracing and Feature Set

The record lists no RT core count and no tensor core count, so dedicated ray tracing and tensor acceleration hardware is not documented in the data. API support is DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.3, defining the software interface set. The architecture is GCN 4.0, built on the Ellesmere chip at 14 nm.

Display connectivity consists of one HDMI 2.0b port and three DisplayPort 1.4a ports. These outputs, along with the API list, are the exposed feature set. The compute resources available to those APIs are 5.728 TFLOPS FP32 and 5.728 TFLOPS FP16 on a 1:1 basis. Without RT or tensor core counts, the data provides no basis for claiming dedicated ray tracing acceleration or AI acceleration hardware.