ATI Radeon HD 3870

The ATI Radeon HD 3870, built by AMD on the RV670 chip, arrived in November 2007 as part of the Radeon R600 generation. It uses the TeraScale architecture on a 55 nm TSMC process, packing 666 million transistors into a 192 mm² die. With 320 shading units, 16 TMUs, and 16 ROPs, it was positioned as a mainstream offering at launch, carrying a launch MSRP of 269 USD. The card sits at the 50th percentile among all GPUs in the database, though its benchmark array is empty.

Benchmark Performance

The benchmark database records no scores for the HD 3870; its average benchmark score is 0, and its percentile rank among all GPUs is 50. That percentile places it exactly at the median of the database's tracked graphics cards, meaning half of all GPUs score higher and half score lower. However, without concrete benchmark entries, the card's practical performance must be inferred from its architectural throughput figures.

The compute throughput is 497.3 GFLOPS of FP32. The pixel fill rate is 12.43 GPixel/s, and the texture fill rate is 12.43 GTexel/s. These matching pixel and texture rates suggest a balanced design where the 16 ROPs and 16 TMUs are fed by the 320 shading units at a consistent rate. Memory bandwidth is 72.06 GB/s, delivered over a 256-bit bus using GDDR4 memory clocked at 1126 MHz (2.3 Gbps effective). That bandwidth is the ceiling for how much data the shading units can pull from the 512 MB frame buffer.

In the context of its generation, the 50th percentile ranking indicates a card that would have been competitive for mainstream gaming at the time of its release, though the absence of recorded benchmarks means the database cannot quantify its lead or deficit against any specific rival. The 12.43 GPixel/s pixel rate and 12.43 GTexel/s texture rate are moderate figures by the standards of the database's full range, reinforcing the median position. The FP32 throughput of 497.3 GFLOPS is similarly a middle-ground figure.

The 512 MB memory capacity, combined with the 256-bit bus, would have been adequate for the resolution and texture detail levels common in the late 2000s, but the card's performance ceiling is defined by that 72.06 GB/s of bandwidth. When the shading units demand more data than the bus can deliver, fill rate and compute throughput will stall. The data suggests a card that performs in line with its percentile: neither a flagship nor an entry-level part, but a solid midpoint.

Power and Cooling

The HD 3870 has a thermal design power of 106 W. That figure places it below the power-hungry flagships of its era, but it still requires dedicated power delivery. The card uses a single 6-pin PCIe power connector, and AMD recommends a 300 W power supply. This is a modest recommendation, indicating that the card was designed to fit into mainstream system builds without requiring an oversized PSU.

The card is a dual-slot design, meaning it occupies two expansion slots in a chassis. Its length is 241 mm, or 9.5 inches, which is a compact footprint for a dual-slot card and would fit in most mid-tower cases of the period. The 55 nm process node from TSMC contributes to the 106 W TDP; the smaller process geometry allows the 666 million transistors to operate at that power level. The die size is 192 mm², and the transistor density is 3.5M per mm², which is a direct result of the 55 nm fabrication process.

Cooling requirements are straightforward: the dual-slot cooler dissipates the 106 W of heat, and the 300 W PSU recommendation accounts for the rest of a typical system's draw. The single 6-pin connector is a single point of power delivery, and the card's power draw is consistent with its mid-range positioning. Users building a system around this card would need a power supply with at least one 6-pin PCIe connector, but no more than that.

Ray Tracing and Feature Set

The HD 3870 has no ray tracing cores and no tensor cores; these fields are null in the hardware data. The card's feature set is defined by its API support rather than dedicated RT hardware. This is a card from an era before such accelerators were part of the graphics pipeline.

The card supports DirectX 10.1, specifically the 10_1 feature level. This is the highest DirectX feature level exposed by the card. On the OpenGL side, the card offers full OpenGL 3.3 support and partial OpenGL 4.0 support. There is no Vulkan support listed in the data.

The display outputs are 2x DVI and 1x S-Video, which is a conventional output set for the period. The card connects to a system via PCIe 2.0 x16, providing a modern (for its time) bus interface. The absence of RT and tensor cores means the card relies entirely on its 320 shading units for all graphics compute, with no dedicated acceleration for ray tracing or machine learning workloads. The 16 TMUs and 16 ROPs handle texturing and rasterization, respectively, and the 12.43 GTexel/s texture rate and 12.43 GPixel/s pixel rate define those limits.

Who Should Consider It

The HD 3870, with its 50th percentile ranking and no recorded benchmark scores, is not a card for modern high-end gaming. Its 512 MB of GDDR4 memory and 72.06 GB/s bandwidth would be quickly saturated by contemporary titles. However, for users building a period-correct system or seeking a basic display adapter for legacy software, the card's 320 shading units and 16 ROPs provide adequate throughput for older games at lower resolutions and reduced detail settings.

The 12.43 GPixel/s pixel rate and 12.43 GTexel/s texture rate suggest the card can handle 2D desktop workloads and early 3D games with ease. The 497.3 GFLOPS of FP32 compute is sufficient for the DirectX 10.1 era titles that were contemporary with its release. The 256-bit memory bus and 72.06 GB/s bandwidth are well-matched to the 512 MB frame buffer, meaning the card will not be bottlenecked by memory in workloads that fit within its capacity.

The card's 106 W TDP and 300 W PSU recommendation make it easy to integrate into a vintage build. The 241 mm length and dual-slot design are standard, and the single 6-pin connector is widely compatible with period power supplies. The display outputs, 2x DVI and 1x S-Video, support the monitors and TVs of the era. This is a card for enthusiasts of the late-2000s era, not for users seeking modern features like ray tracing or Vulkan support.

How It Compares

The nearestRivals data for the HD 3870 is empty, so the database provides no direct competitor scores or delta percentages. The card's position must be understood through its own metrics and its place in the AMD lineup. Its predecessor is the Radeon R500 PCIe series, and its successor is the Radeon R700 series. The HD 3870 bridges these two generations, inheriting the R500's PCIe interface while introducing the TeraScale architecture that would be refined in the R700.

Within the R600 generation, the HD 3870 sits below the higher-end parts of that lineup, but the database does not list those specific rivals. The 50th percentile rank is the only comparative metric available, and it places the card at the median of all GPUs tracked by the database. This is a meaningful data point: the card is neither a standout performer nor a laggard. Its 497.3 GFLOPS of FP32 compute and 72.06 GB/s of memory bandwidth are the numbers that define its capability, and they are consistent with a median ranking.

Without rival scores, the HD 3870's performance can only be assessed in absolute terms. The 12.43 GPixel/s fill rate and 12.43 GTexel/s texture rate are the card's raw throughput limits. The 512 MB memory capacity is a constraint that would become more limiting as texture sizes grew. The card's 106 W TDP is modest, and its 300 W PSU recommendation is accessible. In the absence of competitor data, the HD 3870 stands as a well-balanced mid-range card whose legacy is defined by its architectural efficiency rather than by any dominance over rivals.