AMD Radeon RX 470D

The AMD Radeon RX 470D is a 14 nm Ellesmere GPU built on GCN 4.0 and belongs to the Arctic Islands (RX 400) generation. It integrates 5,700 million transistors on a 232 mm² die, for a transistor density of 24.6M per mm², and is manufactured at GlobalFoundries. The engine runs at a 926 MHz base clock and a 1206 MHz boost clock, with 1792 shading units, 112 texture units, and 32 ROPs. Memory is 4 GB of GDDR5 on a 256-bit bus, clocked at 1650 MHz (6.6 Gbps effective), yielding 211.2 GB/s of bandwidth. The card’s derived pixel rate is 38.59 GPixel/s, its texture rate is 135.1 GTexel/s, and FP32 throughput is 4.322 TFLOPS, with FP16 at the same 4.322 TFLOPS at a 1:1 ratio. It was released on 2016-10-20 and is now marked end-of-life. The database entry places it at the 50th percentile of all GPUs; no individual benchmark scores are populated in the record.

Who Should Consider It

The database places the RX 470D at exactly the midpoint of all GPUs it lists, which is a meaningful positional marker: half of the database entries rank above this card and half rank below it. In the absence of benchmark scores, that percentile, combined with the listed compute resources, suggests a mainstream part rather than an entry-level or top-tier one. The 1792 shading units and 4.322 TFLOPS give it enough raw algebra throughput for many common rendering workloads, while the 112 texture units and 135.1 GTexel/s texture rate can sustain texture-heavy scenes. The 32 ROPs and 38.59 GPixel/s pixel rate, however, indicate that pure fill-limited effects are a weaker area.

The 4 GB GDDR5 frame buffer is the clearest boundary for settings selection. Users who keep texture working sets under that limit will be better served than users who expect to load high-resolution assets and post-processing buffers simultaneously. For lower-resolution targets with moderate settings, the memory capacity and bandwidth are not likely to be the primary limitation. For higher-resolution targets, the 4 GB capacity and 211.2 GB/s bandwidth become constraining factors, so settings may need to be reduced in areas that consume memory and bandwidth, such as texture quality and high-resolution effects.

Because the production status is end-of-life, this card is primarily relevant to users assembling systems from existing stock or used parts. The PCIe 3.0 x16 interface makes it compatible with a broad range of older platforms, but the dual-slot design and 241 mm (9.5 inches) length impose physical clearance requirements. The card’s moderate compute resources and memory footprint place it in a segment where users should target reasonable, not extreme, graphical configurations. Database information does not include benchmark scores, so any resolution-specific conclusion is derived from the memory subsystem and compute rates rather than from measured frame rates.

Memory Subsystem

The RX 470D is equipped with 4 GB of GDDR5 memory. The interface width is 256 bit, and the memory clock is 1650 MHz, or 6.6 Gbps effective. Multiply the bus width by the effective clock, and the bandwidth figure is 211.2 GB/s, as listed in the database. That is a substantial amount of bandwidth for a card of this class, and it allows the GPU to feed its output and texture units without an obvious mismatch. The 256-bit bus is the structural reason for that bandwidth: a narrower interface would require significantly higher clocks to reach the same number.

For high-resolution workloads, the memory size matters first. 4 GB is a fixed ceiling for all framebuffer data, texture allocations, geometry staging, and any other GPU-visible data that must reside on the card. Games or applications that exceed 4 GB will have to spill to system memory or reduce quality, and neither option is transparent to performance. The 211.2 GB/s bandwidth is sufficient for many scenes, but bandwidth-intensive effects such as large render targets, heavy post-processing, or high-resolution texture streaming can put pressure on the memory system. The combination of 4 GB capacity and 211.2 GB/s bandwidth makes this card a better match for controlled memory budgets than for environments that demand both very large allocations and very high transfer rates.

The memory type is GDDR5, which is standard for this generation. The 1650 MHz memory clock is the base figure in the database, with 6.6 Gbps effective signaling. Users should pay attention to the memory capacity as the binding constraint when selecting texture quality or resolution, since the compute side of the card has more headroom than the memory side in high-resolution scenarios.

Power and Cooling

The RX 470D has a TDP of 120 W. The suggested PSU rating is 300 W, and the card requires one 6-pin PCIe power connector. These figures define the minimum infrastructure for operation. A user should have a power supply that includes a 6-pin connector and is sized to the database’s 300 W suggestion; systems with additional components may need more, but the database does not provide other wattage figures. The power delivery requirement is modest, consistent with a 120 W TDP part.

Cooling uses a dual-slot design. The card’s length is 241 mm, or 9.5 inches, which is a typical clearance figure for a mid-range card. No height or width dimensions are present in the database, so the physical check should focus on the listed length and the dual-slot thickness. The 120 W TDP is not extreme, and the dual-slot form factor provides mounting space for a paired cooler. Users with cases that can fit a 241 mm dual-slot card and supply one 6-pin cable will satisfy the physical requirements. As with any GPU, airflow and cooling performance depend on the chassis, but the database entry does not specify a cooler type or thermal ratings.

How It Compares

The nearestRivals array in the database entry is empty, so there are no rival names, scores, or deltaPct values to report. Without that data, no direct score-to-score comparison against specific competing GPUs is possible. The only comparative figure available is the percentileVsAllGpus value of 50, which places the RX 470D at the median of the database’s GPU population. This is not a per-rival positioning, but it does show that the card is neither a low-end outlier nor a high-end performer in the database’s overall distribution.

In terms of product lineage, the database lists Pirate Islands as the predecessor generation and Polaris as the successor generation. That places the RX 470D between two named GPU eras, but the database does not include a consecutive product hierarchy within the RX 400 family. Because no nearest rival entries are populated, the comparison section can only point to the 50th percentile and the predecessor/successor record. Users looking for a rank versus specific competitors will not find it in this entry.

Ray Tracing and Feature Set

The database record contains no RT core count and no tensor core count for the RX 470D. This means the feature set is not shaped by dedicated ray-tracing hardware or dedicated tensor-style accelerators. The API support in the record is DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.3. Those APIs provide the forward-compatible graphics entry points for the card, but the absence of RT core and tensor core fields in the database means no hardware-level ray-tracing or machine-learning acceleration can be confirmed.

The FP16 rate is 4.322 TFLOPS, equal to the FP32 rate at a 1:1 ratio. This is a compute capability that can be used by workloads supporting half-precision arithmetic; it does not derive from a tensor core, since none is listed. The display feature set includes one HDMI 2.0b port and three DisplayPort 1.4a ports. Those outputs define the monitor connection options but do not imply any particular ray-tracing behavior. The card’s architectural feature set is rooted in GCN 4.0, and the database’s API list defines the software interface. With no RT cores and no tensor cores, users or developers evaluating this card should look to traditional shader-based rendering and general-purpose compute rather than hardware-accelerated ray tracing or tensor-accelerated workloads.