ATI Mobility Radeon HD 3470

The ATI Mobility Radeon HD 3470 is an AMD mobile graphics part built on the M82 chip with TeraScale architecture. TSMC fabricated the die on a 55 nm process, and the die contains 181 million transistors on an area of 67 mm², for a transistor density of 2.7M per square millimeter. It was released on 2008-01-06 as part of the M8x (Mobility HD 3400) generation, with the M7x line as predecessor and the M9x line as successor. The specification record contains no benchmark entries, no aggregate score beyond zero, no nearest rivals, and no RT or tensor core details, so this analysis is anchored to the listed hardware and API facts.

Power and Cooling

The power documentation is sparse. The fact pack lists no TDP, no suggested PSU, and no power connectors. Instead of a desktop-style connector requirement, the form factor is identified as “MXM Module” and the bus interface as MXM-II. That combination points to a modular mobile GPU whose power and thermal design belong to the host portable computer rather than to an external power supply.

The physical construction is defined by the 55 nm TSMC process and the 67 mm² die with 181 million transistors. Those figures describe how densely the chip is built — 2.7M transistors per square millimeter — but they do not quantify heat output. Without a TDP value, the data cannot support any wattage estimate, nor can it support a PSU recommendation. The absence of a suggested PSU field is consistent with the MXM Module form factor: the host system is expected to supply whatever power the module needs through the MXM-II interface.

Output connectivity reinforces the same point. The displayOutputs field is “Portable Device Dependent,” so the actual display connectors and the power delivery to those outputs are determined by the portable device, not by the GPU module itself. This means thermal solution, power connector requirements, and PSU needs cannot be generalized from the product data; each system integration decides them. The fact pack therefore provides no basis for comparing this card to desktop GPUs on thermal load or power draw.

Ray Tracing and Feature Set

The dataset contains no ray tracing core count and no tensor core count. There is no evidence in the fact pack of dedicated RT acceleration or tensor-based compute hardware. The programmable core is described instead by 40 shading units, 4 texture mapping units, and 4 render output units. The architecture is TeraScale, the chip is M82, and the generation is M8x (Mobility HD 3400).

API support is the clearest feature boundary: DirectX 10.1 and OpenGL 3.3 are listed. Vulkan is not listed, and FP16 performance is not listed. The only compute rate present is FP32, at 54.40 GFLOPS. For feature-set purposes, the absence of Vulkan and the presence of DirectX 10.1 / OpenGL 3.3 mean the GPU is defined by that API generation. The listed throughput rates are 2.720 GPixel/s and 2.720 GTexel/s, which are fill-rate limits rather than feature flags.

What remains absent is just as important: no RT cores, no tensor cores, no Vulkan, no FP16 value. A user relying on this fact pack can verify compatibility only up to DirectX 10.1 / OpenGL 3.3 applications. Anything beyond those APIs is unsupported by this listing. Feature-set analysis therefore centers entirely on the 40 shading units, 4 TMUs, 4 ROPs, and the legacy API set.

Memory Subsystem

The memory subsystem is small and narrow. The card has 256 MB of DDR2, a 64-bit bus, and a memory clock of 400 MHz, rated at 800 Mbps effective. Those specifications combine to a bandwidth of 6.400 GB/s.

For high resolutions, the key constraints are the 256 MB capacity and the 6.400 GB/s bandwidth. A frame buffer of 256 MB limits the amount of rendered geometry and texture data that can reside on the GPU. The 64-bit interface restricts how much data can cross between memory and the processor each cycle. At high-resolution settings, where frame buffer demands rise with pixel count and texture data, these limits matter.

The list includes no larger memory option, so 256 MB is the only capacity recorded. No alternative memory type or wider bus is provided. The 800 Mbps effective data rate is the transfer rate used to reach 6.400 GB/s. In short, the memory subsystem is built for lower resolutions and modest texture loads, not for high-resolution frame buffers. The combination of DDR2, a 64-bit path, and 256 MB capacity is the entire memory story in this record.

Who Should Consider It

There is no benchmark score to anchor a recommendation. The benchmarks array is empty, the average benchmark score is 0, and nearestRivals is empty. The only database-wide field is percentileVsAllGpus, which reports 50. Without observed scores, that 50th percentile position is not supported by measured data. Consequently, any resolution or settings recommendation must be inferred from specifications, not from scores.

The specification profile points to older, lighter workloads. It is a TeraScale-based M8x part released on 2008-01-06, placed between M7x and M9x. It is end-of-life, so it is not a current production product. It uses 256 MB DDR2 memory, a 64-bit bus, and supports DirectX 10.1 / OpenGL 3.3. Because the display outputs are portable-device dependent and the form factor is MXM-II, it is intended for integration in a portable system rather than a desktop.

Potential use cases that fit this data include running legacy applications that expect DirectX 10.1 or OpenGL 3.3 support, using a portable system where the card is already built into the chassis, or operating at resolutions and settings that respect 256 MB of memory and 6.400 GB/s of bandwidth. There is no evidence in the fact pack that this card can drive modern high-fidelity workloads or high-resolution displays. The data simply cannot justify such a recommendation. No nearest-rival data exists to suggest where this part would sit relative to other mobile GPUs of its era.

Benchmark Performance

The benchmark performance record is empty. There are no entries in the benchmarks array, no nearestRivals list, and no average score other than a literal zero. As a result, no observed performance percentiles or exact percentage deltas versus rival products can be reported.

The available performance-related numbers are theoretical hardware limits. The ATI Mobility Radeon HD 3470 is rated for a pixel rate of 2.720 GPixel/s, a texture rate of 2.720 GTexel/s, and an FP32 rate of 54.40 GFLOPS. These values derive from the 40 shading units, 4 TMUs, and 4 ROPs in the design. They are useful for understanding the ceiling of the hardware, but they are not a substitute for benchmark results.

Because nearestRivals is empty, this analysis contains no percentage-delta statements against competitor cards. The percentileVsAllGpus value of 50 in the database cannot be checked against any named GPU. Without benchmark entries, any percentage comparison would be unsupported. The data supports only the specification-level throughput rates above. The conclusion from the benchmark section is therefore negative: no measured performance data exists for this GPU in this dataset, and any competitive comparison would require a populated benchmarks table and a nearestRivals list.