AMD Radeon RX 6700

The AMD Radeon RX 6700 is a mid-range offering from the Radeon RX 6000 series, built on the RDNA 2.0 architecture with the Navi 22 chip. It sits at the 74th percentile among all GPUs in the database, with an average benchmark score of 31112, placing it in a competitive position against several professional and consumer cards. This analysis examines its memory, performance, power, and feature set using only the available benchmark data.

Memory Subsystem

The RX 6700 ships with 10 GB of GDDR6 memory on a 160-bit bus, delivering a memory bandwidth of 320.0 GB/s. The memory clock runs at 2000 MHz, which translates to 16 Gbps effective data rate. This configuration is notable for its capacity-to-bandwidth ratio — 10 GB is generous for 1080p and 1440p gaming, but the 160-bit bus is narrower than what some rivals use, which could limit performance at very high resolutions where bandwidth demand scales sharply.

At 4K, texture-heavy scenes often require more than 320.0 GB/s to avoid frame pacing issues, but the 10 GB frame buffer helps with caching. The data indicates the card’s pixel rate is 156.8 GPixel/s and its texture rate is 352.8 GTexel/s, which suggests that fill-rate-bound workloads may be less of a bottleneck than memory throughput. For high-resolution gaming, the practical implication is that the RX 6700 can handle 1440p comfortably, but 4K performance will depend heavily on game optimization and whether the 320.0 GB/s bandwidth becomes a limiting factor. The 10 GB capacity is also relevant for modern titles with high-resolution texture packs, though it falls short of the 16 GB found on some higher-tier cards.

Who Should Consider It

Benchmark results indicate this GPU is best suited for gamers targeting 1440p at high settings. Its average benchmark score of 31112 places it just below the NVIDIA Quadro M5000 (31142) by -0.1%, but slightly above the GeForce RTX 3070 Ti (30849) by 0.9%. For 1080p, the card is overkill in most scenarios, delivering ample headroom for high refresh rates. At 1440p, the 10 GB VRAM and 320.0 GB/s bandwidth are sufficient for current titles, and the card’s 11.29 TFLOPS of FP32 performance supports demanding shader workloads.

For 4K, the data suggests caution. The card’s 74th percentile ranking implies it outperforms three-quarters of all GPUs, but the narrow memory bus and modest bandwidth compared to top-tier cards mean that 4K ultra settings may require reduced texture quality or upscaling. Users with high-refresh-rate 1440p monitors or those who prefer maxed-out settings in esports titles will find this card well matched. Conversely, users seeking a future-proof 4K solution should look elsewhere, as the benchmark deltas against faster rivals are small.

Benchmark Performance

The RX 6700’s benchmark scores vary significantly across different APIs and workloads. In Geekbench Metal, it scores 122223, which is its strongest showing, while Geekbench Vulkan scores 96619 and Geekbench OpenCL scores 92672. These numbers indicate strong compute performance in metal-based applications, but the Vulkan and OpenCL results are notably lower, suggesting driver optimizations or architectural preferences for certain APIs.

In Passmark tests, the card scores 250 in DirectX 9, 166 in DirectX 11, and 71 in DirectX 12, with a DirectX 10 score of 115. The DirectX 12 score is surprisingly low compared to the others, which may reflect workload-specific inefficiencies. The Passmark G3D score is 18931, while G2D is 936, and GPU compute is 8240. The 3DMark Steel Nomad DX12 test yields a score of 2014, which is a more modern and demanding benchmark.

Relative to rivals, the RX 6700’s average score of 31112 is nearly identical to the NVIDIA Quadro M5000’s 31142, a delta of just -0.1%. Against the NVIDIA GRID M60-1Q (31220), it trails by -0.3%, but against the GeForce RTX 3070 Ti (30849), it leads by 0.9%. The largest gap is versus the Quadro RTX 8000 (31401), where the RX 6700 is -0.9% behind. These deltas are all within 1 percentage point, indicating that the RX 6700 is firmly in a tight performance cluster. However, the real-world implications depend on which tests are prioritized — the Geekbench Metal score suggests Apple-centric workloads may favor this card, while the low Passmark DX12 score could be a concern for future DX12-heavy games.

How It Compares

vs. NVIDIA Quadro M5000: The RX 6700 is effectively tied with the M5000, trailing by just -0.1% in average score (31112 vs. 31142). Both cards share similar compute capabilities, but the RX 6700 offers modern features like RT cores and DirectX 12 Ultimate, which the M5000 lacks. For gaming, the RX 6700 is clearly the better choice, but for professional compute, the M5000’s driver support may give it an edge in specific workloads.

vs. NVIDIA GRID M60-1Q: The GRID M60-1Q scores 31220, putting it 0.3% ahead of the RX 6700. This is a virtualized GPU for cloud gaming, so the comparison is mostly academic. The RX 6700 has dedicated display outputs (1x HDMI 2.1 and 3x DisplayPort 1.4a), whereas the GRID card is designed for datacenter use. In raw performance, the difference is negligible, but the RX 6700’s consumer features make it more versatile.

vs. NVIDIA GeForce RTX 3070 Ti: The RX 6700 leads the RTX 3070 Ti by 0.9% (31112 vs. 30849), which is a surprising result given the RTX 3070 Ti’s higher tier in NVIDIA’s lineup. However, this average includes multiple API tests, and the RX 6700’s strong Metal score may skew the average. In DX12-specific tests, the RTX 3070 Ti likely performs better, but the data does not break down per-test rival scores. The RX 6700’s 10 GB VRAM matches the RTX 3070 Ti’s capacity, but the latter has a wider memory bus.

vs. NVIDIA Quadro RTX 8000: The RTX 8000 scores 31401, which is 0.9% higher than the RX 6700. The RTX 8000 is a professional card with significantly more VRAM (though not specified here), but the benchmark average shows they are closely matched. The RX 6700’s power efficiency (175 W TDP) is likely superior, but for professional ray tracing workloads, the RTX 8000’s dedicated RT cores may offer advantages that the average score does not capture.

Power and Cooling

The RX 6700 has a TDP of 175 W, which is modest for its performance class. AMD recommends a 450 W power supply, and the card requires a single 8-pin power connector. This makes it suitable for a wide range of systems, including smaller form factor builds, as long as the PSU meets the 450 W requirement. The card is dual-slot in width, with dimensions of 267 mm (10.5 inches) in length, 110 mm (4.3 inches) in height, and 40 mm (1.6 inches) in width. These dimensions are relatively compact, allowing for compatibility with most mid-tower cases.

The 175 W TDP is lower than many competing cards with similar performance, which suggests efficient power delivery. The 7 nm process node from TSMC, with 17,200 million transistors on a 335 mm² die, contributes to this efficiency. The transistor density is 51.3M per mm², indicating a dense design that balances performance and power draw. For users with existing 450 W PSUs, this card is a drop-in upgrade without needing to replace the power supply. The single 8-pin connector is also a plus, as it avoids the need for multiple cables or adapters.

FAQ

Q: What is the RX 6700’s memory bandwidth?

A: The card has a memory bandwidth of 320.0 GB/s, derived from 10 GB of GDDR6 memory on a 160-bit bus running at 2000 MHz (16 Gbps effective).

Q: How does the RX 6700 compare to the GeForce RTX 3070 Ti?

A: In average benchmark scores, the RX 6700 leads by 0.9% (31112 vs. 30849), though this aggregate includes tests like Geekbench Metal where the RX 6700 excels. Individual game performance may vary.

Q: Does the RX 6700 support DirectX 12 Ultimate?

A: Yes, the card supports DirectX 12 Ultimate (12_2), along with OpenGL 4.6 and Vulkan 1.4, as listed in the API specifications.

Q: What is the recommended power supply for this card?

A: AMD suggests a 450 W PSU, and the card uses a single 8-pin power connector. The TDP is 175 W.

Q: Is the RX 6700 good for 4K gaming?

A: The 10 GB VRAM and 320.0 GB/s bandwidth are adequate, but the card’s performance percentile (74th) and narrow memory bus suggest 1440p is the sweet spot. 4K may require lower settings.

Q: What is the card’s production status?

A: The RX 6700 is marked as end-of-life, having been released on 2021-06-08. It is part of the Radeon RX 6000 series, succeeding the Navi architecture and preceding Navi III.

Ray Tracing and Feature Set

The RX 6700 includes 36 dedicated RT cores, which is notably fewer than some higher-end RDNA 2 cards, but it still enables hardware-accelerated ray tracing. The card does not list tensor cores, so AI-accelerated features like DLSS are not available; however, AMD’s FSR (FidelityFX Super Resolution) can be used as a spatial upscaler, though it is not mentioned in the fact pack. The architecture is RDNA 2.0, which supports DirectX 12 Ultimate (12_2), ensuring compatibility with modern ray-traced titles and mesh shaders.

The API support includes Vulkan 1.4, which is the latest version, and OpenGL 4.6, providing broad compatibility across older and newer applications. The display outputs are 1x HDMI 2.1 and 3x DisplayPort 1.4a, allowing for high refresh rates at 1440p and 4K on compatible monitors. The absence of tensor cores means that any AI-based features must rely on compute shaders, which the 36 RT cores do not accelerate. For users prioritizing ray tracing, the RT core count (36) is modest, and the card’s performance in ray-traced workloads will likely be lower than cards with more RT cores, such as the RTX 3070 Ti, though the average score shows the RX 6700 is slightly ahead overall.

The 11.29 TFLOPS of FP32 performance and 22.58 TFLOPS of FP16 (2:1 ratio) provide solid compute throughput, which benefits both gaming and content creation. The pixel rate of 156.8 GPixel/s and texture rate of 352.8 GTexel/s round out the feature set, indicating balanced rasterization capabilities. Overall, the RX 6700 is a well-rounded card for its class, with modern API support and a reasonable RT core count, though its legacy as an end-of-life product means future driver optimizations may be limited.