ATI Radeon HD 3470

The ATI Radeon HD 3470, built on the RV620 chip and TeraScale architecture, is an end-of-life entry-level part from AMD. Fabricated on TSMC's 55 nm process, the chip packs 181 million transistors into a 67 mm² die, yielding a transistor density of 2.7M per mm². Released on January 22, 2008, it succeeded the Radeon R500 PCIe series and was followed by the Radeon R700. With a 64-bit memory interface, 40 shading units, and a 30 W TDP, this card targets basic 3D acceleration and media playback rather than high-end gaming. The data shows a part that is firmly anchored in its era, with no modern acceleration features and a median position in the overall GPU distribution.

Memory Subsystem

The HD 3470 ships with 256 MB of GDDR3 memory on a 64-bit bus. The memory clock runs at 950 MHz, translating to 1900 Mbps effective, which yields a peak bandwidth of 15.20 GB/s. This is a very narrow pipe by any standard. The 64-bit interface halves the data path compared to mainstream parts of its generation, and the 256 MB capacity is insufficient for modern high-resolution textures. At resolutions above 1080p, the frame buffer will spill over, forcing the card to swap to system memory via the PCIe 2.0 x16 interface. The pixel rate is 3.200 GPixel/s and the texture rate is 3.200 GTexel/s, meaning the card can fill 3.2 billion pixels and texels per second. These figures are tightly coupled to the memory bandwidth; the 15.20 GB/s limit will throttle any attempt to push high resolutions. For 1080p and lower, the bandwidth is adequate for basic tasks, but for 1440p or 4K, the combination of 256 MB VRAM and 15.20 GB/s will cause severe stuttering and texture pop-in. The 40 shading units and 4 ROPs further constrain the resolution ceiling. The data shows that this card is best suited for 720p or lower with reduced detail settings. The 4 TMUs and 4 ROPs mean that texture filtering and pixel output are extremely limited, making any high-resolution workload impractical. The 64-bit bus is the primary bottleneck; even if the core could process more data, the 15.20 GB/s bandwidth would starve it. For legacy applications that fit within 256 MB, the card performs adequately, but the memory subsystem is clearly the weakest link.

Ray Tracing and Feature Set

The HD 3470 does not include dedicated ray tracing cores or tensor cores; the fact pack lists null values for both. This places it firmly in the pre-DXR era. The API support includes DirectX 10.1 (specifically the 10_1 feature level) and OpenGL 3.3. There is no Vulkan support, which means modern titles relying on Vulkan will not run. The feature set is limited to the capabilities of the TeraScale architecture. DirectX 10.1 introduces some improvements over 10.0, such as better shader model support, but it lacks the compute shaders and advanced geometry features found in later APIs. For ray tracing, the absence of RT cores means no hardware-accelerated ray tracing is possible. Software-based ray tracing would be impractically slow given the 64.00 GFLOPS of FP32 compute. The 4 texture mapping units and 4 render output units further limit the card's ability to process complex lighting effects. In summary, the HD 3470 is a fixed-function part for its era, offering only the basic rasterization pipeline with no modern acceleration features. The lack of Vulkan is particularly telling; it means the card cannot run any modern game engine that has moved to Vulkan as a primary API. OpenGL 3.3 is also quite old, lacking the compute shaders and multi-draw indirect features that are standard today. The DirectX 10.1 support is a small step above DirectX 10, but it is still far behind DirectX 11 and 12. The 64.00 GFLOPS FP32 performance is a theoretical ceiling that would be quickly saturated by any ray tracing or compute workload. For users interested in modern features, this card is entirely unsuitable.

Power and Cooling

Power consumption is a highlight of this card. The TDP is just 30 W, which is remarkably low even for its time. The card requires no external power connectors; it draws all its power from the PCIe 2.0 x16 slot. The suggested PSU is a 200 W unit, which is extremely modest and compatible with almost any system. The card is a single-slot design, measuring 168 mm in length (6.6 inches). This makes it suitable for small form factor cases and office PCs. The absence of power connectors simplifies installation, and the low TDP means a passive or small active cooler is sufficient. The 30 W figure also implies low heat output, so thermal management is trivial. The single-slot form factor ensures it won't block adjacent slots. The 200 W PSU recommendation is conservative; even a 150 W unit would likely suffice, but the data specifies 200 W as the guideline. The 168 mm length is compact, fitting into most cases without clearance issues. The lack of power connectors means no cable management is needed, and the card can be installed in a system with a minimal power supply. The 30 W TDP is a strength, but it comes at the cost of raw performance. For a database entry, the power figures are notable because they show how far low-end parts have come in efficiency. The 55 nm process node, while large by modern standards, allows for this low power draw. The single-slot cooler is adequate for the heat generated, and the card runs quietly under load. The 200 W PSU recommendation is the only power-related number, and it is a very low bar for any system builder.

How It Compares

The nearestRivals field in the dataset is empty, so direct comparisons with specific competitor cards are not available from this data. However, the card's position within its own lineage is clear. It succeeds the Radeon R500 PCIe series and is succeeded by the Radeon R700. The 50th percentile rank among all GPUs in the database indicates that the HD 3470 sits exactly at the median of the performance distribution. This means half of all GPUs are slower and half are faster. Given its 40 shading units, 4 TMUs, and 4 ROPs, it is a very low-end part. The 64-bit memory bus and 256 MB VRAM place it at the entry level. Compared to its successor, the Radeon R700 series, the HD 3470 lacks the architectural improvements that came with that generation. The lack of Vulkan support and the absence of RT cores put it behind any modern GPU. The 30 W TDP is a strength, but it comes at the cost of raw performance. The data shows that the HD 3470 is a historical piece, relevant only for legacy systems or basic 2D acceleration. Without nearest rival data, we cannot state percentage deltas, but the percentile provides a global context. The 50th percentile is a surprising rank for such a low-spec card, but it reflects the fact that the database includes many equally old and low-powered parts. The predecessor, Radeon R500 PCIe, was an older architecture, so the HD 3470 represents a step forward in terms of DirectX 10.1 support and process node. The successor, Radeon R700, brought significant improvements, but the HD 3470 remains a baseline for entry-level performance. The empty nearestRivals field means we cannot benchmark against specific competitors, but the overall position is that of a median performer in a legacy context.

Benchmark Performance

The benchmark array for the HD 3470 is empty, and the average benchmark score is listed as 0. This indicates that no standardized benchmark scores have been recorded for this card in the database. However, the percentile vs all GPUs is 50, which places it at the median. This is a curious combination: an average score of 0 with a 50th percentile suggests that the percentile is derived from a distribution that includes many cards with zero scores, or that the percentile is based on a different metric. The theoretical compute capability is 64.00 GFLOPS of FP32. This is a very low number, roughly 1/1000th of a modern flagship. The pixel rate of 3.200 GPixel/s and texture rate of 3.200 GTexel/s are similarly low. Without rival scores, we cannot compute exact delta percentages. The data simply shows that the card is at the 50th percentile, meaning it is an average performer in the overall database, but given the era of its release (January 2008), that average is relative to a much older pool of GPUs. In practice, the HD 3470 would be severely outperformed by any card from the last decade. The absence of benchmark scores prevents a precise performance analysis, but the theoretical peak of 64 GFLOPS gives a baseline. The 15.20 GB/s bandwidth is a bottleneck, and the 256 MB VRAM is a hard limit for modern workloads. The data indicates that this card is not suitable for any current gaming or compute tasks. Its 50th percentile rank is likely an artifact of the database's inclusion of many similarly low-powered legacy cards. For users seeking a comparison, the lack of nearest rivals means we cannot offer specific deltas, but the overall picture is one of a very low-performance part. The 64.00 GFLOPS figure is the only compute metric available, and it is dwarfed by even entry-level integrated graphics from the past decade. The 3.200 GPixel/s and 3.200 GTexel/s rates are equally unimpressive, indicating that the card can only handle very light 2D workloads. The empty benchmark array means that no real-world tests have been run, so the theoretical numbers are the only guide. The 50th percentile is a statistical marker, but without actual scores, it is difficult to interpret its practical significance. The card's performance is best described as adequate for its intended role as a basic display adapter, not as a gaming or compute solution.