AMD Radeon RX 6900 XT vs NVIDIA GeForce RTX 4070 Comparison

Head-to-Head Benchmarks

The benchmark data presents a surprisingly balanced picture, with each card taking five wins across the ten recorded tests. The AMD Radeon RX 6900 XT claims the larger individual victories, while the NVIDIA GeForce RTX 4070 edges ahead in several close contests.

The most significant single win for the RX 6900 XT comes in Geekbench OpenCL, where it scores 187,673 against the RTX 4070's 154,858, a commanding 21.2% lead. This is a substantial margin in a compute-oriented workload, suggesting the older RDNA 2 architecture retains real strength in general-purpose GPU compute. The AMD card also wins decisively in PassMark DirectX 10, posting 167 versus 139, a 20.1% advantage, and in PassMark DirectX 11 with 279 against 244, a 14.3% lead. In the modern 3DMark Steel Nomad DX12 test, the RX 6900 XT scores 4079 to the RTX 4070's 3854, a 5.8% win that indicates the AMD card holds an edge in at least one current-generation DirectX 12 workload. PassMark DirectX 12 also favors AMD, 113 to 103, a 9.7% difference.

The RTX 4070's wins are generally narrower. Its largest victory is in PassMark DirectX 9, where it scores 320 versus 268, a 16.2% lead, which is notable given how much older that API is. In Geekbench Vulkan, the NVIDIA card scores 174,152 against 148,525, a 14.7% advantage, showing the Ada Lovelace architecture has a clear edge in Vulkan compute. The remaining three wins are close: PassMark G2D at 1164 versus 1055 (9.4% lead), PassMark GPU Compute at 14,720 versus 14,547 (1.2% lead), and PassMark G3D at 26,927 versus 26,732 (a razor-thin 0.7% lead). The overall average benchmark scores reflect this split differently: the RX 6900 XT averages 50,951, while the RTX 4070 averages 37,648. The RX 6900 XT sits at the 86th percentile of all GPUs, while the RTX 4070 is at the 81st percentile.

Looking at the nearest rivals for context, the RX 6900 XT's average score of 50,951 places it 1.6% behind the NVIDIA CMP 50HX (51,790), 1.9% ahead of the AMD Radeon RX Vega 64 (50,001), 2% ahead of the NVIDIA GeForce RTX 5070 Ti (49,957), and 2.4% ahead of the Intel Arc A550M (49,737). The RTX 4070's average of 37,648 is essentially tied with the NVIDIA Tesla P4 (37,628, a 0.1% delta), 0.4% ahead of the AMD Radeon RX Vega 56 (37,507), 1.3% behind the NVIDIA GeForce RTX 4080 Mobile (38,135), and 1.3% ahead of the AMD Radeon PRO W6400 (37,157). The gap between the two cards' average scores is large, but the head-to-head tests show that the RTX 4070 is competitive in several specific workloads, particularly Vulkan and legacy DirectX 9 scenarios.

FAQ

Q: Which card wins in DirectX 12 Ultimate workloads?

A: The AMD Radeon RX 6900 XT wins the 3DMark Steel Nomad DX12 test with a score of 4079 against the RTX 4070's 3854, a 5.8% lead. It also wins PassMark DirectX 12, 113 to 103, a 9.7% advantage. However, the RTX 4070 wins PassMark DirectX 11 by a 0.7% margin in G3D, so the picture is mixed across APIs.

Q: How do the two compare in compute performance?

A: The RX 6900 XT has a strong lead in Geekbench OpenCL, scoring 187,673 versus 154,858, a 21.2% difference. The RTX 4070 counters with a 174,152 to 148,525 win in Geekbench Vulkan, a 14.7% lead. In PassMark GPU Compute, the RTX 4070 wins narrowly, 14,720 to 14,547, a 1.2% margin.

Q: Which card has better legacy DirectX 9 performance?

A: The NVIDIA GeForce RTX 4070 wins PassMark DirectX 9 with a score of 320 versus 268, a 16.2% lead. This is the RTX 4070's largest single win in the head-to-head data.

Q: What is the difference in average benchmark scores?

A: The RX 6900 XT has an average benchmark score of 50,951, while the RTX 4070 averages 37,648. The RX 6900 XT sits at the 86th percentile of all GPUs, compared to the RTX 4070's 81st percentile.

Q: How close are the two in overall 3D graphics performance?

A: In PassMark G3D, the RTX 4070 wins by a minimal margin, scoring 26,927 against 26,732, a 0.7% difference. This is the closest result in the entire head-to-head set.

Q: Which card wins in 2D performance?

A: The RTX 4070 wins PassMark G2D with a score of 1164 versus 1055, a 9.4% lead. This is one of the NVIDIA card's four wins outside of Vulkan and DirectX 9.

Architecture Differences

The two cards come from fundamentally different architectural generations. The AMD Radeon RX 6900 XT uses the Navi 21 chip built on RDNA 2.0, fabricated on a 7 nm process at TSMC. The NVIDIA GeForce RTX 4070 uses the AD104 chip built on Ada Lovelace, fabricated on a 5 nm process, also at TSMC. The process node difference is significant: 7 nm versus 5 nm, which partly explains the transistor density disparity. The RX 6900 XT packs 26,800 million transistors across a 520 mm² die, yielding a density of 51.5 million transistors per square millimeter. The RTX 4070 contains 35,800 million transistors on a much smaller 294 mm² die, achieving 121.8 million transistors per square millimeter, more than double the density of the AMD part.

The compute architectures diverge in how they handle floating-point workloads. The RX 6900 XT delivers 23.04 TFLOPS of FP32 performance and 46.08 TFLOPS of FP16 using a 2:1 ratio, meaning FP16 throughput is double FP32. The RTX 4070 delivers 29.15 TFLOPS of FP32 and 29.15 TFLOPS of FP16 with a 1:1 ratio, meaning both precisions run at the same rate. This is a notable philosophical difference: RDNA 2 dedicates extra hardware to half-precision, while Ada Lovelace treats FP16 as a first-class citizen at the same throughput as FP32.

Ray tracing hardware also differs. The RX 6900 XT has 80 ray accelerators, while the RTX 4070 has 46 RT cores. The NVIDIA card also includes 184 tensor cores, which the AMD card lacks entirely. The RTX 4070's shading unit count is higher at 5,888 versus 5,120, but the AMD card has more texture mapping units (320 versus 184) and more raster operation units (128 versus 64). The pixel rate reflects this: the RX 6900 XT outputs 288.0 GPixel/s, while the RTX 4070 manages 158.4 GPixel/s. Texture rate similarly favors AMD at 720.0 GTexel/s versus 455.4 GTexel/s. Both cards support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

Specification Differences

The memory subsystems are markedly different. The RX 6900 XT features 16 GB of GDDR6 memory on a 256-bit bus, delivering 512.0 GB/s of bandwidth. The RTX 4070 has 12 GB of GDDR6X memory on a 192-bit bus, delivering 504.2 GB/s. The AMD card has slightly more bandwidth despite using slower memory, while the NVIDIA card uses faster memory on a narrower bus. The memory clocks differ: the RX 6900 XT runs at 2000 MHz (16 Gbps effective), while the RTX 4070 runs at 1313 MHz (21 Gbps effective).

Clock speeds favor NVIDIA. The RTX 4070 has a base clock of 1920 MHz and a boost clock of 2475 MHz. The RX 6900 XT has a base clock of 1825 MHz, a game clock of 2015 MHz, and a boost clock of 2250 MHz. Power consumption diverges significantly: the RX 6900 XT has a TDP of 300 W and requires a 700 W power supply, while the RTX 4070 has a TDP of 200 W and requires a 550 W supply. The physical footprint differs as well: the RX 6900 XT is a triple-slot card measuring 267 mm in length, 120 mm in height, and 50 mm in width, while the RTX 4070 is a dual-slot card at 240 mm long, 110 mm high, and 40 mm wide. The RX 6900 XT uses two 8-pin power connectors, while the RTX 4070 uses a single 16-pin connector.

Display outputs differ: the RX 6900 XT offers 1x HDMI 2.1, 2x DisplayPort 1.4a, and 1x USB Type-C, while the RTX 4070 offers 1x HDMI 2.1 and 3x DisplayPort 1.4a. Both use PCIe 4.0 x16. The release dates are separated by roughly two and a half years: the RX 6900 XT launched on 2020-10-27, and the RTX 4070 launched on 2023-04-11. Both are now listed as end-of-life products. The RX 6900 XT carries a launch MSRP of 999 USD, while the RTX 4070 has a launch MSRP of 599 USD.

Where Each One Wins

The AMD Radeon RX 6900 XT is the stronger choice for OpenCL compute workloads, where its 21.2% lead over the RTX 4070 is the largest margin in the entire comparison. It also dominates in DirectX 10 and DirectX 11 legacy titles, with wins of 20.1% and 14.3% respectively. For users running DirectX 12 games, the RX 6900 XT shows a 5.8% advantage in 3DMark Steel Nomad and a 9.7% lead in PassMark DirectX 12. The card's higher pixel rate of 288.0 GPixel/s and texture rate of 720.0 GTexel/s suggest it is well suited to rasterization-heavy workloads that stress fill rate. Its 16 GB memory capacity is also larger, which may benefit texture-heavy scenarios.

The NVIDIA GeForce RTX 4070 is the better option for Vulkan-based workloads, where it leads by 14.7% in Geekbench Vulkan. It also wins in legacy DirectX 9 applications with a 16.2% advantage. The RTX 4070 edges ahead in PassMark G3D by 0.7%, making it the nominal winner for general 3D performance as measured by that suite. Its PassMark G2D win of 9.4% indicates faster 2D operations. The card's 1:1 FP16 to FP32 ratio of 29.15 TFLOPS means it handles half-precision workloads at the same throughput as full precision, which may benefit certain compute tasks. The tensor cores present on the RTX 4070 are a feature the RX 6900 XT does not have, though the recorded benchmarks do not isolate tensor-specific workloads.

The Verdict

The data presents a nuanced picture. The AMD Radeon RX 6900 XT wins five of ten head-to-head tests, including the three largest margins in the comparison (OpenCL, DirectX 10, and DirectX 11). Its average benchmark score of 50,951 is substantially higher than the RTX 4070's 37,648, and it sits at the 86th percentile of all GPUs against the RTX 4070's 81st. For users prioritizing OpenCL compute, DirectX 10/11 legacy performance, or DirectX 12 workloads as measured by 3DMark Steel Nomad, the RX 6900 XT is the better pick.

The NVIDIA GeForce RTX 4070 counters with wins in Vulkan, DirectX 9, and the two PassMark 3D tests, though its largest victory is 16.2% and its narrowest is 0.7%. It achieves these results while consuming 200 W versus 300 W, requiring a 550 W power supply instead of 700 W, and occupying a dual-slot form factor instead of triple-slot. The RTX 4070 also has a lower launch MSRP of 599 USD against 999 USD. For users focused on Vulkan compute, legacy DirectX 9 titles, or who prioritize efficiency and physical footprint, the RTX 4070 is the data-backed choice. The choice depends on which workloads matter most, as the two cards split their wins almost evenly.