AMD Radeon HD 6670

The AMD Radeon HD 6670 is a TeraScale 2–based graphics card from the Northern Islands generation, built on TSMC's 40 nm process with 716 million transistors on a 118 mm² die. It occupies the 10th percentile of all GPUs in the benchmark database, placing it firmly in the low-end segment, yet its Geekbench OpenCL score of 1894 edges out every one of its four nearest rivals by a small margin, as detailed below.

Benchmark Performance

The HD 6670’s only recorded benchmark is Geekbench OpenCL, where it scores 1894 points. This places it at the 10th percentile of all GPUs, meaning roughly 90% of tested graphics cards outperform it. Within its immediate competitive set, however, the card is a leader: it beats the NVIDIA GeForce GT 425M (1869) by 1.3%, the ATI Radeon HD 5870 (1848) by 2.5%, and both the NVIDIA GeForce GT 525M (1846) and the NVIDIA Quadro K420 (1846) by 2.6%. These deltas are small but consistent, suggesting the HD 6670 holds a narrow compute advantage over these specific contemporaries.

The score reflects raw OpenCL throughput, not gaming rasterization, but it still gives a sense of the card’s overall computational headroom. With 480 shading units, 24 texture units, and 8 ROPs, the HD 6670 delivers 768.0 GFLOPS of FP32 compute, 19.20 GTexel/s of texture fill, and 6.400 GPixel/s of pixel fill. Those numbers are modest by modern standards, which is why the percentile ranking is so low. Yet the fact that it outperforms the GT 425M and HD 5870 in OpenCL suggests that for compute-oriented tasks, the card can hold its own against slightly older or lower-tier parts.

How It Compares

NVIDIA GeForce GT 425M — The HD 6670 is 1.3% faster in Geekbench OpenCL. The GT 425M is a mobile GPU, so the comparison is cross-platform, but the margin is negligible in practice. Both cards sit in the same performance strata, with the HD 6670 offering a slight edge in raw compute.

ATI Radeon HD 5870 — Despite being a higher-end part from an earlier generation, the HD 5870 scores 2.5% lower than the HD 6670 in this OpenCL test. That is surprising given the HD 5870’s larger memory bus and higher transistor count, but the benchmark result shows the HD 6670’s TeraScale 2 architecture and GDDR5 memory are sufficient to outperform it in this specific workload.

NVIDIA GeForce GT 525M — Another mobile GPU, the GT 525M trails by 2.6%. The HD 6670’s advantage here is consistent with its higher shading unit count and faster memory clock (4 Gbps effective vs. the GT 525M’s unspecified but likely slower GDDR3). The delta is small enough that real-world differences would be imperceptible.

NVIDIA Quadro K420 — A professional workstation card, the Quadro K420 also scores 1846, matching the GT 525M. The HD 6670 leads by 2.6%. For compute tasks, the HD 6670 is marginally ahead, though the Quadro’s driver optimizations for professional applications might narrow that gap outside of synthetic benchmarks.

Ray Tracing and Feature Set

The HD 6670 has no dedicated ray tracing cores and no tensor cores—both fields are null in the specifications. This is expected for a card from 2011, as ray tracing hardware did not appear in consumer GPUs until much later. The architecture is TeraScale 2, which predates even the first-generation GCN (Graphics Core Next) that introduced unified shader improvements but still lacked RT acceleration.

API support is limited to DirectX 11.2 (feature level 11_0) and OpenGL 4.4. There is no Vulkan support, which means the card cannot run modern Vulkan-based titles or compute workloads. DirectX 11.2 is sufficient for many games released up to the mid-2010s, but it lacks the newer features like mesh shaders or variable rate shading. The absence of Vulkan also rules out compatibility with a growing number of current engines that have dropped DirectX 11 support. In short, the HD 6670 is a legacy product with no modern feature set, making it unsuitable for contemporary AAA gaming or compute applications that rely on Vulkan or ray tracing.

FAQ

Q: How does the HD 6670 compare to the NVIDIA GeForce GT 425M in Geekbench OpenCL?

A: The HD 6670 scores 1894, which is 1.3% higher than the GT 425M’s 1869.

Q: What is the memory configuration of the HD 6670?

A: It has 1024 MB of GDDR5 memory on a 128-bit bus, providing 64.00 GB/s of bandwidth.

Q: Does the HD 6670 support Vulkan?

A: No. The API list includes DirectX 11.2 (11_0) and OpenGL 4.4, but Vulkan is not supported.

Q: What is the transistor count and die size?

A: The chip, codenamed Turks, contains 716 million transistors on a 118 mm² die, fabricated on a 40 nm process at TSMC.

Q: What is the TDP and power connector requirement?

A: The TDP is 66 W, and it requires no external power connectors, drawing all power from the PCIe slot. A 250 W PSU is suggested.

Q: When was the HD 6670 released and what is its production status?

A: It was released on April 18, 2011, and is now end-of-life. Its predecessor is Evergreen and its successor is Southern Islands.

Who Should Consider It

Given its 10th percentile ranking and 1894 OpenCL score, the HD 6670 is not a card for high-end gaming or heavy compute. Its 768.0 GFLOPS FP32 throughput and 64.00 GB/s memory bandwidth are sufficient for 720p or low-detail 1080p gaming in titles from its era, but modern games will struggle. The 1024 MB VRAM is also a limiting factor for high-resolution textures—at 1080p with high detail, the frame buffer can easily overflow, causing stutter or texture pop-in.

This card is best suited for users who need a basic display output for office work, legacy software, or as a backup GPU. Its single-slot design and lack of power connectors make it easy to install in small form factor systems. It also supports DirectX 11.2, so older games that use that API will run acceptably at reduced settings. However, anyone expecting to play current titles or run compute workloads that require Vulkan or ray tracing will be disappointed. The HD 6670 is a relic of a bygone era, and its benchmark performance confirms that it should be considered only for very light, non-demanding tasks.

Memory Subsystem

The HD 6670 is equipped with 1024 MB of GDDR5 memory, a 128-bit bus, and a bandwidth of 64.00 GB/s. The memory clock runs at 1000 MHz, which translates to 4 Gbps effective due to the GDDR5 double data rate. This configuration was standard for entry-level cards in 2011, but it is severely limited by modern standards. The 64 GB/s bandwidth is roughly one-tenth of what a contemporary mid-range card offers, and the 1 GB capacity is insufficient for high-resolution gaming with large texture packs.

At 1080p, the HD 6670 can handle low to medium settings in games from its release era, but the memory bandwidth will become a bottleneck in scenes with heavy texture streaming or high anti-aliasing. At 1440p or 4K, the card would be completely overwhelmed, as the frame buffer and bandwidth cannot sustain the data throughput required. For compute tasks, the 64 GB/s bandwidth also limits performance in memory-bound workloads, even though the OpenCL score shows a slight edge over its direct rivals. In summary, the memory subsystem is a clear weak point, and any usage scenario that demands high resolutions or large datasets will expose this limitation.

Power and Cooling

The HD 6670 has a TDP of 66 W, which is modest and allows for a single-slot cooling solution. It requires no external power connectors—all power is supplied through the PCIe 2.0 x16 slot. The suggested PSU is a 250 W unit, which is very low by today’s standards and indicates that the card can run in almost any system with a functional PCIe slot. The physical dimensions are 168 mm (6.6 inches) in length, making it compatible with most cases, including compact ones.

The absence of power connectors simplifies installation, but it also means the card is strictly limited to the 66 W envelope. This is fine for a card of this performance class, but it also means there is no headroom for overclocking beyond what the board’s power delivery can handle. The single-slot design keeps the card thin, but the cooler must be efficient enough to dissipate 66 W—typically a small fan and aluminum heatsink suffice. For a system builder looking for a low-power, low-noise GPU for basic tasks, the HD 6670’s power profile is a positive attribute. However, for any heavy load, the thermal and power constraints will quickly become apparent. The card’s end-of-life status and 2011 release date further suggest that it is not a practical choice for modern systems, but its low power draw and simple installation requirements make it a viable option for retro builds or troubleshooting.