AMD Radeon RX 6900 XT vs NVIDIA GeForce RTX 5090 Comparison

Head-to-Head Benchmarks

The recorded data shows a complete sweep for the NVIDIA GeForce RTX 5090 across all ten head-to-head benchmark comparisons. The most dramatic gap appears in 3DMark Steel Nomad DX12, where the RTX 5090 scores 18,355 against the Radeon RX 6900 XT's 4,079, a 350% advantage. This is the single largest delta in the comparison and reflects a generational leap in raw rendering throughput.

In compute-oriented workloads, the RTX 5090 continues its dominance. Geekbench OpenCL results show 334,370 versus 187,673, a 78.2% lead. The Vulkan test widens further: 376,728 against 148,525, representing a 153.6% margin. These results indicate that the NVIDIA card maintains its advantage across both API environments, with the Vulkan gap being particularly pronounced.

PassMark benchmarks provide a nuanced view of different rendering paths. The DirectX 12 test shows 185 for the RTX 5090 against 113 for the RX 6900 XT, a 63.7% difference. DirectX 11 narrows slightly to 22.2% (341 vs 279), while DirectX 10 shows a 35.3% gap (226 vs 167). Interestingly, the DirectX 9 test produces a 47.4% difference (395 vs 268), suggesting that even legacy APIs favor the newer architecture.

The 2D performance test yields 1,413 for the RTX 5090 versus 1,055 for the RX 6900 XT, a 33.9% delta. This is the smallest relative gap in the suite, yet still a clear win. The G3D test shows 39,650 against 26,732, a 48.3% advantage, while GPU Compute delivers 26,756 versus 14,547, an 83.9% lead. Across all tests, the RTX 5090 wins by a minimum of 22.2% and a maximum of 350%, with zero losses recorded.

Architecture Differences

The fundamental divergence begins at the manufacturing process. The RTX 5090 uses a 5 nm node at TSMC, while the RX 6900 XT uses a 7 nm node, also at TSMC. This process advantage allows the NVIDIA chip, GB202, to pack 92,200 million transistors on a 750 mm² die, yielding a transistor density of 122.9M per mm². The AMD Navi 21, by contrast, contains 26,800 million transistors on a 520 mm² die, for a density of 51.5M per mm². The RTX 5090's density is more than double that of its rival.

Architecturally, the RTX 5090 is built on Blackwell 2.0, while the RX 6900 XT uses RDNA 2.0. This represents a full generational shift for NVIDIA, whose predecessor was GeForce 40, versus AMD's Navi II generation, whose predecessor was Navi. The Blackwell design introduces 21,760 shading units, 680 TMUs, and 176 ROPs. The RDNA 2.0 design offers 5,120 shading units, 320 TMUs, and 128 ROPs. The NVIDIA card also carries 170 ray tracing cores and 680 tensor cores, while the AMD card has 80 ray tracing cores and no tensor cores at all.

Memory subsystems differ substantially. The RTX 5090 features 32 GB of GDDR7 on a 512-bit bus, delivering 1.79 TB/s of bandwidth. The RX 6900 XT uses 16 GB of GDDR6 on a 256-bit bus, providing 512.0 GB/s. This is a 3.5x bandwidth advantage for the NVIDIA card. Clock speeds show the RTX 5090 boosting to 2,407 MHz from a base of 2,017 MHz, while the RX 6900 XT boosts to 2,250 MHz from a base of 1,825 MHz, with a game clock of 2,015 MHz. Memory clocks also differ: 1750 MHz (28 Gbps effective) for NVIDIA versus 2000 MHz (16 Gbps effective) for AMD.

Power and interface specifications reflect the performance gap. The RTX 5090 has a TDP of 575 W, uses a single 16-pin connector, and requires a 950 W suggested PSU. The RX 6900 XT draws 300 W, uses two 8-pin connectors, and needs a 700 W PSU. The bus interface is PCIe 5.0 x16 for the NVIDIA card versus PCIe 4.0 x16 for AMD. Display outputs differ as well: the RTX 5090 offers 1x HDMI 2.1b and 3x DisplayPort 2.1b, while the RX 6900 XT provides 1x HDMI 2.1, 2x DisplayPort 1.4a, and 1x USB Type-C.

Where Each One Wins

The RTX 5090 wins every benchmark category, so the "where each one wins" split is largely about degrees of dominance rather than alternative strengths. In DirectX 11, the smallest margin (22.2%) suggests that older, single-threaded CPU-bound workloads reduce the gap. Yet even there, the NVIDIA card prevails. For gamers targeting the highest frame rates in DX12 titles, the 63.7% advantage in that specific test is the most relevant metric.

Compute workloads show the RTX 5090's tensor core advantage. The 83.9% lead in GPU Compute and 78.2% in OpenCL point to significant headroom for AI, scientific, and productivity tasks. The Vulkan result, at 153.6%, is particularly strong for cross-platform engines and modern game APIs. The 2D test, at 33.9%, indicates that desktop and UI acceleration is also faster, though the gap is smaller.

For the RX 6900 XT, the data offers no outright wins. Its best relative performance is in DirectX 11, where it trails by only 22.2%, and in 2D, where it lags by 33.9%. Users already invested in the Radeon ecosystem might still find it adequate for legacy DX11 titles, but the recorded benchmarks show no scenario where it outperforms the RTX 5090. The RX 6900 XT also carries a lower TDP at 300 W, which suggests an efficiency advantage per watt, though no efficiency ratio is recorded in the data.

The Verdict

The database measurements are unambiguous: the NVIDIA GeForce RTX 5090 is the superior product, winning 10 of 10 head-to-head comparisons. Its average benchmark score of 79,842 places it at the 92nd percentile among all GPUs, while the RX 6900 XT averages 50,951 at the 86th percentile. The NVIDIA card's nearest rivals are Tesla P100 variants and AMD's RX 6850M XT, all within 1.1%, indicating it sits in a distinct performance tier above the Radeon.

The RX 6900 XT's nearest rivals include the NVIDIA CMP 50HX (1.6% below), the RX Vega 64 (1.9% above), the RTX 5070 Ti (2% above), and the Intel Arc A550M (2.4% above). This places the AMD card in a crowded mid-to-high range, whereas the RTX 5090 operates at the top of the measured spectrum.

Users should choose the RTX 5090 for any workload where maximum performance is non-negotiable: DX12 gaming, Vulkan applications, compute tasks, or ray tracing. The 350% lead in Steel Nomad and the 153.6% Vulkan advantage are decisive. The RX 6900 XT remains a capable card for DX11 legacy titles or for users constrained by power limits, but the data shows no performance scenario where it exceeds the RTX 5090. The RTX 5090 was released on 2025-01-29 with a launch MSRP of 1,999 USD; the RX 6900 XT launched on 2020-10-27 with a launch MSRP of 999 USD.

FAQ

Q: Which card has the higher average benchmark score?

A: The NVIDIA GeForce RTX 5090 scores 79,842 on average, while the AMD Radeon RX 6900 XT scores 50,951, a difference of 28,891 points.

Q: What is the largest performance gap between the two cards?

A: In 3DMark Steel Nomad DX12, the RTX 5090 outscores the RX 6900 XT by 350% (18,355 vs 4,079).

Q: Does the RX 6900 XT win any benchmark in the head-to-head data?

A: No. The RTX 5090 wins all 10 recorded head-to-head tests, with the RX 6900 XT recording zero wins.

Q: How do the memory sizes compare?

A: The RTX 5090 has 32 GB of GDDR7 on a 512-bit bus, while the RX 6900 XT has 16 GB of GDDR6 on a 256-bit bus.

Q: Which card has more ray tracing cores?

A: The RTX 5090 has 170 ray tracing cores, compared to 80 on the RX 6900 XT.

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

A: The RTX 5090 ranks at the 92nd percentile, and the RX 6900 XT ranks at the 86th percentile.

Specification Differences

| Field | NVIDIA GeForce RTX 5090 | AMD Radeon RX 6900 XT |

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

| Process Node | 5 nm | 7 nm |

| Transistors | 92,200 million | 26,800 million |

| Die Size | 750 mm² | 520 mm² |

| Transistor Density | 122.9M / mm² | 51.5M / mm² |

| Architecture | Blackwell 2.0 | RDNA 2.0 |

| Shading Units | 21,760 | 5,120 |

| TMUs | 680 | 320 |

| ROPs | 176 | 128 |

| RT Cores | 170 | 80 |

| Tensor Cores | 680 | null |

| Base Clock | 2017 MHz | 1825 MHz |

| Boost Clock | 2407 MHz | 2250 MHz |

| Game Clock | null | 2015 MHz |

| Memory Clock | 1750 MHz (28 Gbps effective) | 2000 MHz (16 Gbps effective) |

| Memory Size | 32 GB | 16 GB |

| Memory Type | GDDR7 | GDDR6 |

| Memory Bus Width | 512 bit | 256 bit |

| Memory Bandwidth | 1.79 TB/s | 512.0 GB/s |

| Pixel Rate | 423.6 GPixel/s | 288.0 GPixel/s |

| Texture Rate | 1,636.8 GTexel/s | 720.0 GTexel/s |

| FP32 Performance | 104.8 TFLOPS | 23.04 TFLOPS |

| FP16 Performance | 104.8 TFLOPS (1:1) | 46.08 TFLOPS (2:1) |

| TDP | 575 W | 300 W |

| Slot Width | Dual-slot | Triple-slot |

| Power Connectors | 1x 16-pin | 2x 8-pin |

| Suggested PSU | 950 W | 700 W |

| Bus Interface | PCIe 5.0 x16 | PCIe 4.0 x16 |

| Display Outputs | 1x HDMI 2.1b, 3x DisplayPort 2.1b | 1x HDMI 2.1, 2x DisplayPort 1.4a, 1x USB Type-C |

| Dimensions (L/H/W) | 304 mm / 137 mm / 40 mm | 267 mm / 120 mm / 50 mm |

| Production Status | Active | End-of-life |

| Release Date | 2025-01-29 | 2020-10-27 |

| Predecessor | GeForce 40 | Navi |

| Successor | GeForce 60 | Navi III |