ATI Mobility Radeon HD 5470

The ATI Mobility Radeon HD 5470 is an AMD mobile GPU built for the Manhattan (Mobility HD 5400) generation, using the Park chip and the TeraScale 2 architecture. It was manufactured by TSMC on a 40 nm process, with 292 million transistors on a 59 mm² die, and released on 2010-01-06. The product is marked end-of-life, and the database entry is unusual: its benchmark array is empty, its average benchmark score is 0, and its nearest-rival list is empty. That leaves specification analysis as the primary way to interpret the hardware.

Memory Subsystem — VRAM size/type, bus width, bandwidth and what it means for high resolutions

The memory subsystem is a 512 MB GDDR5 frame buffer connected over a 64-bit bus. The memory clock is 800 MHz, which translates to 3.2 Gbps effective, and the resulting bandwidth is 25.60 GB/s. A 64-bit interface is narrow for a discrete-class GPU, and 25.60 GB/s is a modest figure even within the context of this mobile part. The 512 MB capacity is the more immediate constraint for high-resolution work: larger frame buffers require more capacity than 512 MB can provide, and the bandwidth figure determines how quickly the GPU can move texture and geometry data once that capacity is filled.

For high-resolution rendering, the combination of 512 MB and 25.60 GB/s means the GPU must keep working sets small. A 64-bit memory bus generally favors lower-resolution workloads, where the smaller bus width is less likely to become the limiting factor. At higher resolutions, both capacity and bandwidth become more stressed, and the data shows no headroom in the memory path beyond what the 25.60 GB/s figure allows. The pixel rate of 3.000 GPixel/s and texture rate of 6.000 GTexel/s are also modest, so the GPU’s fill rates could be exhausted before the memory subsystem is fully utilized in some scenes, while in data-heavy workloads the 64-bit bus could become the bottleneck first.

The memory clock is the only clock listed in the data; base clock and boost clock fields are null. That makes the memory timing the only concrete frequency reference. GDDR5 at 3.2 Gbps effective is reasonable for the era, but the 64-bit bus halves the effective transfer width compared to wider-memory designs. The result is a memory subsystem that is balanced toward low-power, compact mobile designs rather than high-resolution performance. The bus interface is PCIe 2.0 x16, which provides the system connection, but the 64-bit memory interface is the important constraint for frame-buffer throughput.

Ray Tracing and Feature Set

The RT core field is null and the tensor core field is null. There is no stated hardware ray tracing capability and no stated tensor-core capability in this product record. The feature set is therefore defined by the API support and the underlying TeraScale 2 architecture, not by dedicated acceleration blocks. The GPU supports DirectX 11.2 (11_0) and OpenGL 4.4. Vulkan is listed as null, so no Vulkan version is recorded. DirectX 11.2 (11_0) gives the part a feature level of 11_0, which is the relevant API baseline for this GPU.

The shading core consists of 80 shading units, 8 texture mapping units, and 4 raster output units. Those counts are modest, and they set expectations for the GPU’s compute and fill-rate profile. Single-precision floating-point throughput is 120.0 GFLOPS, a number that indicates a low-power part rather than a high-throughput accelerator. With no RT cores and no tensor cores, any ray tracing or machine-learning-style workload would have to run on the general-purpose shader units, and the 120.0 GFLOPS figure is not a strong foundation for such workloads.

The absence of Vulkan support also narrows the modern API surface. DirectX 11.2 (11_0) and OpenGL 4.4 are the supported paths, and neither points to hardware-accelerated ray tracing. The feature set is essentially a DirectX 11-class pipeline built around a small number of shader units, texture units, and ROPs. For the 2010-01-06 release date, this is an expected feature set, but the data does not include any newer acceleration features.

Benchmark Performance

The benchmark data for this GPU is empty. There are no entries in the benchmarks array, and the average benchmark score is 0. The percentile vs all GPUs is 50, which places this part exactly at the midpoint of the database distribution. That midpoint ranking carries less weight than a measured score because the average score of 0 indicates there are no populated benchmark results behind the percentile figure. In other words, the 50th-percentile placement appears to be a positional default rather than a performance measurement.

Without benchmark scores, the concrete performance indicators are the compute and fill-rate numbers. The GPU delivers 120.0 GFLOPS of FP32 throughput, a 3.000 GPixel/s pixel rate, and a 6.000 GTexel/s texture rate. Those figures are the upper bounds on raw throughput, assuming no other bottleneck. The 80 shading units and 8 texture mapping units are aligned with those rates, and the 4 ROPs cap the pixel throughput. A GPU with this configuration is likely to show low frame rates in demanding modern titles, but the database record does not contain a measured score to confirm or refute that expectation.

The nearestRivals list is empty. There are no rival names, no rival scores, and no deltaPct values to analyze. Because of that, it is not possible to state from the data that this GPU is a certain percentage faster or slower than any specific competitor. The only comparative clue is the 50th-percentile rank, which is a database-wide position rather than a rival-to-rival delta. Any attempt to construct exact percentage comparisons would require nearestRivals entries, and those entries are absent.

How It Compares

The nearestRivals field contains no entries. That means the database does not provide a direct comparison to any named rival GPU. There are no deltaPct values, no rival scores, and no rival names to work from. The positional data is limited to the percentileVsAllGpus value of 50, which puts the ATI Mobility Radeon HD 5470 in the middle of all GPUs in the database, and to the product lineage fields that list M9x as the predecessor and Vancouver as the successor.

Against its immediate predecessor, M9x, the data does not list any performance score or percentage delta. The comparison can only be made at the level of product succession: the HD 5470 comes after M9x in the chronological lineup. Similarly, the successor, Vancouver, is listed as the next product but with no benchmark comparison. The absence of rival entries means no paragraph can be written for a named nearest rival. The GPU’s position is defined by its specification and its database-wide percentile, not by side-by-side measurements.

The 50th-percentile rank is the one quantitative placement available. It suggests that, in the database’s overall distribution, this GPU sits exactly at the median. That is not the same as saying it outperforms half of all other GPUs, because no benchmark scores are stored for this part. It is simply the recorded percentile. The product is end-of-life, so its competitive context is historical, but the data set does not include the rival relationships needed to sharpen that context.

Power and Cooling

The TDP is 15 W. That is the only power figure in the data, and it indicates a low-power mobile GPU. A 15 W TDP is consistent with a compact laptop part that does not require the elaborate cooling solutions needed by higher-power desktop or workstation GPUs. The data does not list a slot width, so there is no physical dimension for how much space the card occupies in an expansion sense. More importantly for system integration, the power connector field is null and the suggested PSU field is null. No connector requirement is recorded, and no power supply recommendation is available from the data.

The display outputs are listed as "Portable Device Dependent," which means the physical output configuration is determined by the laptop or mobile device rather than by the GPU itself. This reinforces the mobile nature of the part. Cooling is likely dependent on the host system’s thermal design, since the data gives no cooler specification, no dimensions, and no slot width. The bus interface is PCIe 2.0 x16, which is the system connection, but the power delivery details are not specified.

The 15 W TDP is useful for estimating thermal impact. It is a modest number, which suggests that a quiet, small cooling solution can handle the heat output. However, the absence of a suggested PSU and power connector data means the record does not state what system power supply is required. For a laptop GPU, that is unsurprising: power comes from the portable platform’s internal power system, not from an external PSU. The combination of a 15 W TDP, no power connector data, and portable-device-dependent outputs points to a GPU designed for integration into a sealed mobile chassis rather than for user-upgradeable desktop use.