AMD Radeon RX 470 vs AMD Radeon RX 6800 Comparison

The AMD Radeon RX 6800 and AMD Radeon RX 470 represent two distinct eras of AMD graphics architecture, separated by four years of silicon evolution. Benchmark data shows the RX 6800 wins 3 of 4 head-to-head tests, including a dominant 278.6% lead in 3DMark Steel Nomad DX12, yet the RX 470 actually wins in Geekbench OpenCL by 27%, a surprising result given the generational gap. The RX 6800 commands the higher average benchmark score at 30095 versus 28996 for the RX 470, though the latter sits at the 74th percentile of all GPUs compared to the RX 6800's 75th, indicating the older card remains surprisingly competitive in aggregate metrics despite its age.

The Verdict

The data paints a clear picture: the AMD Radeon RX 6800 is the decisive choice for modern DirectX 12 and Vulkan workloads, while the RX 470 retains a narrow niche in OpenCL compute. The RX 6800 delivers a 278.6% advantage in 3DMark Steel Nomad DX12 (3188 vs 842) and a 188.6% lead in Geekbench Vulkan (115107 vs 39884), making it the only sensible pick for contemporary gaming and ray-traced titles. Its Geekbench Metal score of 153635 crushes the RX 470's 41690 by 268.5%, confirming superiority in Apple ecosystem compute scenarios.

However, the RX 470's 27% win in Geekbench OpenCL (33568 vs 24508) is not a fluke — it suggests the older GCN architecture handles certain OpenCL workloads more efficiently than RDNA 2.0, possibly due to driver optimizations or compute unit scheduling. Buyers with OpenCL-specific legacy applications might find the RX 470 unexpectedly adequate, but this is a narrow use case. The RX 6800's nearest rivals include the RTX 3070 Ti (0.5% higher average score) and RTX 2080 Ti (1% higher), placing it in upper-midrange territory, while the RX 470's rivals — RX 6800M, Arc A370M, and Vega M GH — all sit within 0.7% of its score, showing it clusters with mobile and older discrete parts. For anyone building a system today, the RX 6800 is the data-backed winner; the RX 470 is a curiosity for OpenCL enthusiasts or budget retro builds where its 120W TDP and 4GB GDDR5 suffice.

FAQ

Q: Which card wins the most head-to-head benchmarks?

A: The AMD Radeon RX 6800 wins 3 of 4 tests, with victories in 3DMark Steel Nomad DX12, Geekbench Metal, and Geekbench Vulkan, while the RX 470 wins only Geekbench OpenCL.

Q: How large is the performance gap in DirectX 12 gaming?

A: In 3DMark Steel Nomad DX12, the RX 6800 scores 3188 versus the RX 470's 842, a 278.6% difference favoring the newer card.

Q: Is the RX 470 competitive in any compute workload?

A: Yes — the RX 470 beats the RX 6800 in Geekbench OpenCL by 27% (33568 vs 24508), making it the better choice for certain OpenCL compute tasks.

Q: What are the memory specifications of each card?

A: The RX 6800 has 16 GB of GDDR6 on a 256-bit bus with 512.0 GB/s bandwidth, while the RX 470 has 4 GB of GDDR5 on a 256-bit bus with 211.2 GB/s bandwidth.

Q: How do the cards compare in transistor count and process node?

A: The RX 6800 uses 26,800 million transistors on a 7 nm TSMC process with a 520 mm² die, whereas the RX 470 uses 5,700 million transistors on a 14 nm GlobalFoundries process with a 232 mm² die.

Q: What is the percentile ranking of each card among all GPUs?

A: The RX 6800 ranks at the 75th percentile, while the RX 470 sits at the 74th percentile, a surprisingly narrow gap given the architectural generation difference.

Architecture Differences

The RX 6800 is built on the Navi 21 chip using RDNA 2.0 architecture, a ground-up redesign that introduced hardware ray tracing with 60 dedicated RT cores. Its 7 nm process from TSMC packs 26,800 million transistors into a 520 mm² die, achieving a density of 51.5M transistors per square millimeter. The RX 470, in contrast, uses the Ellesmere chip based on GCN 4.0 architecture, fabricated on GlobalFoundries' 14 nm node with just 5,700 million transistors across a 232 mm² die (24.6M / mm² density). This represents a 4.7x transistor count advantage for the RX 6800, enabling 3840 shading units versus 2048, 240 TMUs versus 128, and 96 ROPs versus 32.

The RX 6800 supports DirectX 12 Ultimate (12_2) and Vulkan 1.4, while the RX 470 is limited to DirectX 12 (12_0) and Vulkan 1.3. RDNA 2.0 also implements a 2:1 FP16 ratio, delivering 32.33 TFLOPS half-precision versus 16.17 TFLOPS FP32, whereas GCN 4.0 runs FP16 at 1:1, meaning 4.940 TFLOPS for both precision modes. The RX 6800's PCIe 4.0 x16 interface doubles bandwidth over the RX 470's PCIe 3.0 x16, and its display outputs include a USB Type-C port absent on the older card. Power delivery differs dramatically — the RX 6800 requires 2x 8-pin connectors and a 600 W PSU, while the RX 470 uses a single 6-pin with a 300 W PSU recommendation, reflecting the TDP gap of 250 W versus 120 W.

Specification Differences

| Specification | AMD Radeon RX 6800 | AMD Radeon RX 470 |

|---|---|---|

| Process Node | 7 nm (TSMC) | 14 nm (GlobalFoundries) |

| Transistors | 26,800 million | 5,700 million |

| Die Size | 520 mm² | 232 mm² |

| Base Clock | 1700 MHz | 926 MHz |

| Boost Clock | 2105 MHz | 1206 MHz |

| Memory Clock | 2000 MHz (16 Gbps effective) | 1650 MHz (6.6 Gbps effective) |

| Memory Size | 16 GB GDDR6 | 4 GB GDDR5 |

| Memory Bandwidth | 512.0 GB/s | 211.2 GB/s |

| Shading Units | 3840 | 2048 |

| TMUs | 240 | 128 |

| ROPs | 96 | 32 |

| RT Cores | 60 | None |

| Pixel Rate | 202.1 GPixel/s | 38.59 GPixel/s |

| Texture Rate | 505.2 GTexel/s | 154.4 GTexel/s |

| FP32 Performance | 16.17 TFLOPS | 4.940 TFLOPS |

| FP16 Performance | 32.33 TFLOPS (2:1) | 4.940 TFLOPS (1:1) |

| TDP | 250 W | 120 W |

| Power Connectors | 2x 8-pin | 1x 6-pin |

| Suggested PSU | 600 W | 300 W |

| Bus Interface | PCIe 4.0 x16 | PCIe 3.0 x16 |

| DirectX Support | 12 Ultimate (12_2) | 12 (12_0) |

| Vulkan Support | 1.4 | 1.3 |

| Dimensions (L×H×W) | 267×120×40 mm | 240×95×35 mm |

| Release Date | 2020-10-27 | 2016-08-03 |

| Launch MSRP | 579 USD | 179 USD |

The RX 6800 also offers a game clock of 1815 MHz, a feature absent from the RX 470's specifications. The display outputs differ: the RX 6800 provides 1x HDMI 2.1, 2x DisplayPort 1.4a, and 1x USB Type-C, whereas the RX 470 has 1x HDMI 2.0b and 3x DisplayPort 1.4a.

Head-to-Head Benchmarks

The most lopsided result appears in 3DMark Steel Nomad DX12, where the RX 6800 scores 3188 against the RX 470's 842 — a 278.6% margin that underscores how far GPU architecture advanced between 2016 and 2020. This test stresses modern DirectX 12 features like mesh shaders and variable rate shading, areas where RDNA 2.0's dedicated hardware excels over GCN 4.0's legacy design. The RX 470's 32 ROPs and 154.4 GTexel/s texture rate simply cannot keep pace with the RX 6800's 96 ROPs and 505.2 GTexel/s, translating to a 4.2x pixel throughput advantage.

Geekbench Metal shows a similarly wide gap: 153635 for the RX 6800 versus 41690 for the RX 470, a 268.5% difference. This benchmark is particularly relevant for macOS users via eGPU enclosures, and the RX 6800's 60 RT cores and 16 GB VRAM provide substantial compute headroom. Vulkan results follow the same trend — the RX 6800 hits 115107 versus 39884, a 188.6% lead that confirms its superiority in cross-platform graphics APIs. The RX 6800's nearest rival in average score, the RTX 3070 Ti at 29945, trails by just 0.5%, while the RX 470's closest competitor, the RX 6800M at 28874, is only 0.4% behind — showing the older card punches near its weight class.

The single anomaly appears in Geekbench OpenCL, where the RX 470 wins 33568 to 24508, a 27% margin in its favor. This result defies the generational trend and suggests GCN 4.0's compute unit organization handles certain OpenCL kernels more efficiently than RDNA 2.0's wavefront scheduling. It is worth noting the RX 470's FP16 performance matches its FP32 at 4.940 TFLOPS, whereas the RX 6800 halves its FP16 throughput to 32.33 TFLOPS via the 2:1 ratio — potentially explaining why OpenCL workloads that leverage FP16 see a relative advantage on the older architecture. Despite this win, the RX 470's overall average benchmark score of 28996 remains below the RX 6800's 30095, and the newer card's 75th percentile ranking versus 74th confirms its status as the better-rounded performer.