ATI Radeon HD 3870 Mac Edition

The ATI Radeon HD 3870 Mac Edition is a 55 nm TeraScale graphics card from AMD, built on the RV670 chip with 666 million transistors on a 192 mm² die at TSMC. It was released June 11, 2008, belongs to the Radeon R600 (HD 3800) generation, and is currently listed as end-of-life. The database record contains no entries in its benchmarks array and no entries in nearestRivals, so its only positional metric is a percentileVsAllGpus of 50. This places the card exactly at the middle of all GPUs in the database, rather than at a high or low extreme.

How It Compares

The nearestRivals field is empty. There are therefore no rival names, scores, or deltaPct values to cite for this product. Because the benchmarks array is also empty, no average benchmark score separates it from other cards. The only comparison data is the percentileVsAllGpus field, which is 50: in the database's overall GPU ranking, half of all GPUs are below this card and half are above. That is a median placement, not a performance outlier. No per-rival paragraphs can be written because the FACT PACK does not provide any rival records for the HD 3870 Mac Edition.

The absence of rival records is itself a notable data condition. This is an end-of-life card with no launch price listed in the pack, and its position in the database is defined by its specification set rather than by measured benchmark comparisons. Without nearestRivals, any relative statement such as "ahead of" or "behind" a specific card would be unsupported by the data.

Ray Tracing and Feature Set

The rtCores field is null, and the tensorCores field is null. This means the FACT PACK lists no ray tracing cores and no tensor cores on the RV670 chip. Ray tracing acceleration and tensor-based processing are not part of the feature set. Instead, the card relies on its TeraScale shader architecture with 320 shading units, 16 texture mapping units, and 16 raster output units. Its texture rate is 12.43 GTexel/s and pixel rate is 12.43 GPixel/s, giving balanced texel and pixel throughput at the listed clocks. FP32 compute is 497.3 GFLOPS.

On the API side, the card supports DirectX 10.1 (10_1), OpenGL 3.3 as full support, and OpenGL 4.0 as partial support. There is no Vulkan API listed. This makes the card a DirectX 10.1-class part with partial modern OpenGL coverage but no Vulkan path. Applications that rely on Vulkan, or on a higher DirectX feature level than 10.1, are not supported according to the FACT PACK. The absence of RT and tensor cores means any workload expecting dedicated ray tracing or tensor hardware will find no such blocks in this GPU.

Who Should Consider It

With 512 MB of GDDR4 memory and a 50th-percentile overall placement, this card is not aimed at high-resolution, max-texture workloads. Its memory clock is 1126 MHz, or 2.3 Gbps effective, across a 256-bit bus for 72.06 GB/s of bandwidth. That bandwidth can feed a moderate frame buffer, but 512 MB is the hard capacity ceiling. Users should consider it for lower resolutions and conservative texture settings, where the 12.43 GPixel/s pixel fill rate and 12.43 GTexel/s texture fill rate are less likely to be overwhelmed.

The card is a Mac Edition variant, and its display output set is 2x DVI and 1x S-Video. It connects through a PCIe 2.0 x16 bus interface. Because it is end-of-life and was released June 11, 2008, it is not a candidate for current-generation software expectations. It is most plausible as a legacy addition to a system that needs those specific display connectors and a single-slot card. The 512 MB memory capacity, rather than the 72.06 GB/s bandwidth, is the more likely limitation at higher resolutions and heavier texture loads.

FAQ

Q: What is the core chip and manufacturing information?

A: The card uses the RV670 chip from AMD, on a 55 nm TSMC process, with 666 million transistors on a 192 mm² die, for a transistor density of 3.5M transistors per mm².

Q: Does it support Vulkan?

A: No. The Vulkan field is null, so the FACT PACK lists no Vulkan support.

Q: What are the memory specifications?

A: It has 512 MB of GDDR4 memory on a 256-bit bus, with a memory clock of 1126 MHz, or 2.3 Gbps effective, yielding 72.06 GB/s of bandwidth.

Q: What power connector is needed?

A: The card uses a 1x 6-pin PCIe power connector, has a 106 W TDP, and a suggested PSU of 300 W.

Q: What are the physical size and slot requirements?

A: It is a single-slot card with a length of 241 mm, which is 9.5 inches.

Q: What is the direct predecessor and successor of this card?

A: Its predecessor is Radeon R500 PCIe, and its successor is Radeon R700.

Benchmark Performance

The benchmarks array in the FACT PACK is empty, and the avgBenchmarkScore field is 0. There are no benchmark scores to break down, no measured entries to average, and no per-rival deltaPct values to calculate. The only score-like value is percentileVsAllGpus: 50. This places the card at the midpoint of the database's GPU distribution. In practical terms, the data says nothing about superiority or inferiority to any specific card; it only records a median rank.

The FP32 throughput of 497.3 GFLOPS and the 12.43 GPixel/s pixel rate / 12.43 GTexel/s texture rate provide theoretical throughput values, but these are not benchmark results. They describe the chip's peak capacities according to its listed clock and unit counts. The 320 shading units, 16 TMUs, and 16 ROPs are the architectural resources behind those rates. Without benchmark scores, any relative performance statements would require the missing nearestRivals data, and none are available. The percentile of 50 is therefore the only quantitative performance position in the database entry.

Power and Cooling

The card's TDP is 106 W, and the suggested PSU is 300 W. It requires one 6-pin PCIe power connector. The single-slot width means it takes one bracket. Its length of 241 mm, or 9.5 inches, must fit inside the target chassis. The FACT PACK does not include a cooler model or noise measurement, so no judgement on cooling quality is possible from the data. What the data does establish is an electrical budget: 106 W TDP, one 6-pin connector, and a 300 W system power recommendation. The single-slot design also implies a thinner cooling solution than a multi-slot card, but the pack does not specify fan dimensions, heatsink material, or thermal behavior.

Memory Subsystem

Memory capacity is 512 MB of GDDR4, controlled by a 256-bit bus. The memory clock is 1126 MHz, which the FACT PACK cites as 2.3 Gbps effective, producing 72.06 GB/s of bandwidth. The 256-bit bus is a wide pathway for its generation, but the 512 MB capacity limits how much texture data and image data can be stored on the card at once. For high-resolution rendering, a larger frame buffer would be needed; the data lists no capacity beyond 512 MB.

The 12.43 GTexel/s texture rate and 12.43 GPixel/s pixel rate define the fill ceiling. Because the two rates are identical, neither texture throughput nor pixel throughput dominates. The 72.06 GB/s memory bandwidth and 256-bit bus are the supporting path for those fill rates, while the 512 MB capacity is the primary restraint on resolution-heavy work. For workloads that fit within 512 MB, the bandwidth and fill rates may be sufficient; for workloads that exceed that capacity, no amount of bandwidth compensates for the missing memory.