AMD Radeon RX 6950 XT vs NVIDIA GeForce RTX 5090 Comparison

Where Each One Wins

The benchmark record splits decisively in one direction: the NVIDIA GeForce RTX 5090 wins all ten recorded head-to-head tests against the AMD Radeon RX 6950 XT. That is not a narrow sweep. Across the database, the RTX 5090 holds an average benchmark score of 79,842, while the RX 6950 XT averages 58,392, a difference of roughly 37% in aggregate performance.

Yet the shape of that advantage changes depending on the workload. The largest margin appears in the 3DMark Steel Nomad DX12 test, a modern DirectX 12 scene, where the RTX 5090 scores 18,355 against 4,235 for the RX 6950 XT, a 333.4% delta. That is the single biggest gap in the recorded data, and it suggests the RTX 5090 is built for current-generation graphics loads that stress geometry, ray tracing, and memory bandwidth simultaneously.

Compute-oriented tests tell a different story in scale but not in direction. In Geekbench OpenCL, the RTX 5090 posts 334,370 versus 205,998, a 62.3% lead. In Geekbench Vulkan, the margin widens to 128%, with scores of 376,728 and 165,212. These results imply that the RTX 5090's architecture scales better under parallel API workloads, likely due to its much larger shader array and memory subsystem.

Legacy DirectX tests show the narrowest margins. In PassMark DirectX 11, the RTX 5090 leads by only 13.7% (341 versus 300). In DirectX 10, the gap is 37.8% (226 versus 164), and in DirectX 9 it is 30.4% (395 versus 303). These older APIs do not fully exercise the newer hardware features, so both cards fall back on raw rasterization throughput. The RX 6950 XT is comparatively closer here, but still loses every test.

The 2D and compute segments reinforce the pattern. PassMark G2D shows a 32.9% lead for the RTX 5090 (1,413 versus 1,063), and PassMark GPU Compute shows an 88.4% advantage (26,756 versus 14,199). The compute result is particularly telling: the RTX 5090's FP32 rating of 104.8 TFLOPS versus the RX 6950 XT's 23.65 TFLOPS aligns with the recorded compute benchmark delta.

In short, the RTX 5090 wins everywhere, but its dominance is most pronounced in modern DX12 and Vulkan scenarios, moderate in compute and OpenCL, and least dramatic in legacy DX9-11 tests where the RX 6950 XT remains competitive in relative terms.

The Verdict

The database points to a clear conclusion: the NVIDIA GeForce RTX 5090 is the superior card in every measured benchmark. There is no recorded test where the AMD Radeon RX 6950 XT comes out ahead. If the question is raw performance, the answer is unambiguous.

For users who prioritize modern game workloads, particularly DirectX 12 or Vulkan titles, the RTX 5090 is the only rational choice from these two. Its 333.4% lead in 3DMark Steel Nomad DX12 and 128% lead in Geekbench Vulkan mean that current and near-future titles will see massive frame rate advantages. The RX 6950 XT, while still a capable GPU in absolute terms, sits far behind in the scenarios that matter most for new releases.

For legacy software compatibility, the RX 6950 XT closes the gap somewhat. In PassMark DirectX 11, the delta shrinks to 13.7%. Older games that rely on DX9 or DX10 pipelines will not show the same dramatic divide. Still, the RTX 5090 wins those tests too, so there is no workload category from the recorded data where the AMD card should be preferred on performance grounds.

The alternative consideration is physical and thermal footprint. The RTX 5090 is a dual-slot card with a 575 W TDP and a suggested 950 W PSU, while the RX 6950 XT is triple-slot, draws 335 W, and suggests a 700 W PSU. The RX 6950 XT is also shorter at 267 mm versus 304 mm. For builds constrained by case space or power delivery, the AMD card is the lighter ask. But the benchmark data does not support choosing it for speed.

The RTX 5090 also sits at the 92nd percentile among all GPUs in the database, versus 88th for the RX 6950 XT. Its nearest rivals average around 79,605 to 80,959, placing it in elite company. The RX 6950 XT's nearest rivals cluster near 58,000 to 58,657, which is a lower performance tier entirely.

Head-to-Head Benchmarks

The largest recorded victory for the RTX 5090 comes in 3DMark Steel Nomad DX12, where it scores 18,355 against the RX 6950 XT's 4,235. That 333.4% delta is not a marginal improvement; it is a generational leap. This test stresses modern rendering features, and the RTX 5090's Blackwell architecture, with 21,760 shading units and 170 RT cores, overwhelms the RX 6950 XT's 5,120 shaders and 80 RT cores.

Geekbench Vulkan produces the second-largest gap at 128%. The RTX 5090 scores 376,728, while the RX 6950 XT manages 165,212. Vulkan is a low-overhead API that scales with raw hardware throughput, and the RTX 5090's 104.8 TFLOPS FP32 and 1.79 TB/s memory bandwidth give it a decisive edge over the RX 6950 XT's 23.65 TFLOPS and 576.0 GB/s.

PassMark GPU Compute shows an 88.4% lead, with scores of 26,756 and 14,199. This test measures general compute throughput, and the RTX 5090's tensor core count of 680 (versus none on the RX 6950 XT) and its 92,200 million transistors versus 26,800 million explain the scale of the advantage.

Geekbench OpenCL and PassMark DirectX 12 both show 62.3% deltas, though from different baselines. In OpenCL, the RTX 5090 scores 334,370 versus 205,998. In DX12, the scores are 185 and 114. The DX12 result is interesting: despite the massive hardware difference, the absolute numbers are low because PassMark's DX12 test is not as demanding as 3DMark's. But the relative gap remains consistent.

PassMark G3D shows a 41.3% lead (39,650 versus 28,070), which is a broad rasterization test. The RTX 5090's 176 ROPs and 1,636.8 GTexel/s texture rate outpace the RX 6950 XT's 128 ROPs and 739.2 GTexel/s.

The narrowest margins are in PassMark DirectX 11 (13.7%, 341 versus 300) and DirectX 10 (37.8%, 226 versus 164). DirectX 9 shows a 30.4% gap (395 versus 303). These legacy APIs do not use ray tracing or tensor cores, so the test reduces to basic shader and fill-rate performance. The RX 6950 XT's RDNA 2 architecture remains respectable here, but it still loses.

PassMark G2D rounds out the sweep with a 32.9% lead (1,413 versus 1,063), indicating that even 2D operations favor the newer card.

FAQ

Q: Which card wins in DirectX 12 performance?

A: The RTX 5090 wins in both recorded DX12 tests. In 3DMark Steel Nomad DX12, it scores 18,355 versus 4,235, a 333.4% lead. In PassMark DirectX 12, it scores 185 versus 114, a 62.3% lead.

Q: Is the RX 6950 XT competitive in any benchmark?

A: The RX 6950 XT does not win any of the ten recorded head-to-head tests. Its closest relative performance is in PassMark DirectX 11, where it trails by 13.7% (300 versus 341).

Q: How do the cards compare in compute workloads?

A: The RTX 5090 leads PassMark GPU Compute by 88.4% (26,756 versus 14,199) and Geekbench OpenCL by 62.3% (334,370 versus 205,998). Its FP32 rating is 104.8 TFLOPS versus 23.65 TFLOPS for the RX 6950 XT.

Q: What memory specifications differ between the two?

A: The RTX 5090 has 32 GB of GDDR7 on a 512-bit bus with 1.79 TB/s bandwidth. The RX 6950 XT has 16 GB of GDDR6 on a 256-bit bus with 576.0 GB/s bandwidth.

Q: Which card has a higher percentile ranking among all GPUs?

A: The RTX 5090 sits at the 92nd percentile, while the RX 6950 XT sits at the 88th percentile.

Q: What are the power requirements?

A: The RTX 5090 has a TDP of 575 W and suggests a 950 W PSU. The RX 6950 XT has a TDP of 335 W and suggests a 700 W PSU.

Architecture Differences

The two cards come from different architectural generations. The RTX 5090 uses NVIDIA's Blackwell 2.0 architecture on a 5 nm process at TSMC, built around the GB202 chip with 92,200 million transistors on a 750 mm² die. The RX 6950 XT uses AMD's RDNA 2.0 on a 7 nm process, also at TSMC, with the Navi 21 chip containing 26,800 million transistors on a 520 mm² die. Transistor density tells the story: the RTX 5090 packs 122.9 million transistors per mm², while the RX 6950 XT manages 51.5 million per mm².

The shader configurations diverge sharply. The RTX 5090 has 21,760 shading units, 680 TMUs, and 176 ROPs. The RX 6950 XT has 5,120 shading units, 320 TMUs, and 128 ROPs. The RTX 5090 also carries 170 RT cores and 680 tensor cores, while the RX 6950 XT has 80 RT cores and no tensor cores. This means the RTX 5090 can handle ray tracing and AI-accelerated workloads natively, whereas the RX 6950 XT lacks tensor hardware entirely.

Memory architecture is another major split. The RTX 5090 uses 32 GB of GDDR7 on a 512-bit bus, achieving 1.79 TB/s bandwidth. The RX 6950 XT uses 16 GB of GDDR6 on a 256-bit bus, with 576.0 GB/s bandwidth. The memory clock differs as well: 1750 MHz (28 Gbps effective) for the RTX 5090 versus 2250 MHz (18 Gbps effective) for the RX 6950 XT. The RTX 5090's wider bus and newer memory type give it more than triple the bandwidth.

Compute rates reflect the architectural gap. The RTX 5090 achieves 104.8 TFLOPS FP32 and 104.8 TFLOPS FP16 (1:1 ratio). The RX 6950 XT achieves 23.65 TFLOPS FP32 and 47.31 TFLOPS FP16 (2:1 ratio). The pixel and texture rates also favor the RTX 5090: 423.6 GPixel/s versus 295.7 GPixel/s, and 1,636.8 GTexel/s versus 739.2 GTexel/s.

Both cards support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4, so API compatibility is equal. The production status differs: the RTX 5090 is active, while the RX 6950 XT is end-of-life.

Specification Differences

The RTX 5090 and RX 6950 XT differ across nearly every specification field. The process node is 5 nm for the RTX 5090 versus 7 nm for the RX 6950 XT. Transistor count is 92,200 million versus 26,800 million, and die size is 750 mm² versus 520 mm².

Clock speeds differ modestly. The RTX 5090 has a base clock of 2017 MHz and boost of 2407 MHz. The RX 6950 XT has a base of 1860 MHz, a game clock of 2100 MHz, and a boost of 2310 MHz. The memory clocks are 1750 MHz (28 Gbps effective) versus 2250 MHz (18 Gbps effective).

Memory is 32 GB GDDR7 on a 512-bit bus for the RTX 5090, versus 16 GB GDDR6 on a 256-bit bus for the RX 6950 XT. Bandwidth is 1.79 TB/s versus 576.0 GB/s.

Shading units are 21,760 versus 5,120. TMUs are 680 versus 320. ROPs are 176 versus 128. RT cores are 170 versus 80. Tensor cores are 680 versus null (the RX 6950 XT has none).

Pixel rate is 423.6 GPixel/s versus 295.7 GPixel/s. Texture rate is 1,636.8 GTexel/s versus 739.2 GTexel/s. FP32 is 104.8 TFLOPS versus 23.65 TFLOPS. FP16 is 104.8 TFLOPS (1:1) versus 47.31 TFLOPS (2:1).

TDP is 575 W versus 335 W. Slot width is dual-slot versus triple-slot. Power connectors are 1x 16-pin versus 2x 8-pin. Suggested PSU is 950 W versus 700 W. Bus interface is PCIe 5.0 x16 versus PCIe 4.0 x16.

Display outputs are 1x HDMI 2.1b and 3x DisplayPort 2.1b for the RTX 5090, versus 1x HDMI 2.1 and 2x DisplayPort 1.4a for the RX 6950 XT.

Dimensions: the RTX 5090 is 304 mm long, 137 mm high, and 40 mm wide. The RX 6950 XT is 267 mm long, 120 mm high, and 50 mm wide. The RTX 5090 is longer but thinner.

Release dates are 2025-01-29 for the RTX 5090 and 2022-05-09 for the RX 6950 XT. The RTX 5090 has a launch MSRP of 1,999 USD; the RX 6950 XT has a launch MSRP of 1,099 USD. Production status is active for the RTX 5090 and end-of-life for the RX 6950 XT.