AMD Radeon RX 6950 XT vs NVIDIA GeForce RTX 5090 Mobile Comparison

FAQ

Q: Which GPU wins in the 3DMark Steel Nomad DX12 test?

A: The NVIDIA GeForce RTX 5090 Mobile wins decisively, scoring 5871 versus the AMD Radeon RX 6950 XT's 4235. This is a 27.9% advantage for the NVIDIA part.

Q: Does the AMD Radeon RX 6950 XT win any benchmark categories?

A: Yes, the AMD card wins 4 out of 10 head-to-head tests. It takes Geekbench OpenCL (205998 vs 201834, a 2.1% lead), Passmark DirectX 11 (300 vs 269, an 11.5% lead), Passmark G2D (1063 vs 1057, a 0.6% lead), and Passmark GPU Compute (14199 vs 13401, a 6% lead).

Q: How do the two GPUs rank against all other GPUs in the database?

A: The AMD Radeon RX 6950 XT sits in the 88th percentile, while the NVIDIA GeForce RTX 5090 Mobile sits in the 84th percentile. Their average benchmark scores are 58392 and 45152 respectively.

Q: What are the closest rivals to each GPU based on average benchmark score?

A: For the AMD card, the nearest rival is the NVIDIA P102-100 at a delta of -0.2%. For the NVIDIA mobile card, the nearest rival is the AMD Radeon Pro 5500 XT at a delta of -0.5%.

Q: Which GPU has the higher memory bandwidth?

A: The NVIDIA GeForce RTX 5090 Mobile has 896.0 GB/s of bandwidth, while the AMD Radeon RX 6950 XT has 576.0 GB/s. The NVIDIA part also has more memory (24 GB GDDR7) versus 16 GB GDDR6 for the AMD card.

Q: Are both GPUs using the same DirectX and Vulkan feature levels?

A: Yes, both support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The differences show up in raw performance, not API compatibility.

The Verdict

The data splits this comparison into two distinct use cases. The NVIDIA GeForce RTX 5090 Mobile is the clear winner for modern, API-heavy workloads: it takes 6 of 10 benchmarks, including a dominant 27.9% lead in 3DMark Steel Nomad DX12 and a 16.7% lead in Geekbench Vulkan. It also wins Passmark DirectX 10, DirectX 12, DirectX 9, and Passmark G3D. For anyone prioritizing current-generation rasterization and compute-heavy game engines, the NVIDIA part is the stronger pick.

The AMD Radeon RX 6950 XT, however, wins in specific legacy and compute scenarios. It leads by 11.5% in Passmark DirectX 11, by 6% in Passmark GPU Compute, and by 2.1% in Geekbench OpenCL. It also edges out the NVIDIA part in Passmark G2D, albeit by a marginal 0.6%. If the workload relies on older DirectX paths or OpenCL compute, the AMD card holds the advantage.

The broader database picture complicates the simple win count. Despite losing the head-to-head tally, the AMD Radeon RX 6950 XT has a higher average benchmark score (58392) than the NVIDIA GeForce RTX 5090 Mobile (45152). The AMD card also ranks higher in percentile vs all GPUs (88th versus 84th). This suggests that across all recorded tests, the AMD part is more consistently strong, even though the NVIDIA part wins the specific tests compared here. The NVIDIA mobile GPU's nearest rival is the AMD Radeon Pro 5500 XT (delta -0.5%), while the AMD desktop card's nearest rival is the NVIDIA P102-100 (delta -0.2%), indicating that the two GPUs occupy different tiers of the database.

Choose the NVIDIA GeForce RTX 5090 Mobile for modern DX12 and Vulkan gaming, and for scenarios where memory size (24 GB) and bandwidth (896.0 GB/s) matter. Choose the AMD Radeon RX 6950 XT for legacy DirectX 11 applications, OpenCL compute tasks, and for a higher overall average benchmark score in the database.

Head-to-Head Benchmarks

The largest single-test gap belongs to the NVIDIA GeForce RTX 5090 Mobile in 3DMark Steel Nomad DX12. The NVIDIA part scores 5871, while the AMD card scores 4235, a delta of -27.9% from the AMD perspective. This is not a close contest; the NVIDIA GPU is nearly 39% faster in absolute score terms (5871 vs 4235). For a mobile part running at 95 W TDP to outperform a desktop card with a 335 W TDP in this test is a significant architectural result.

The second-largest gap goes to the NVIDIA card in Geekbench Vulkan, where it scores 198405 versus 165212, a 16.7% lead. This shows that the NVIDIA part's Blackwell architecture handles Vulkan compute more efficiently than AMD's RDNA 2.0 implementation. The opposite is true in Geekbench OpenCL, where the AMD card wins by a slim 2.1% margin (205998 vs 201834), suggesting the AMD part retains an edge in OpenCL-specific workloads.

In the Passmark suite, the results split along generation lines. The NVIDIA card wins DirectX 10 (183 vs 164, a 10.4% lead), DirectX 12 (138 vs 114, a 17.4% lead), and DirectX 9 (324 vs 303, a 6.5% lead). The AMD card wins DirectX 11 (300 vs 269, an 11.5% lead) and G2D (1063 vs 1057, a 0.6% lead). The DirectX 11 win for AMD is notable because it is a legacy API that many older games and enterprise applications still rely on. The DirectX 12 win for NVIDIA is more relevant for modern titles.

The Passmark G3D test, which approximates general 3D rendering performance, goes to the NVIDIA card with 30034 versus 28070, a 6.5% lead. This aligns with the 3DMark result and reinforces that the NVIDIA part is the stronger overall 3D performer. However, the Passmark GPU Compute test goes the other way: the AMD card scores 14199 versus 13401, a 6% lead. This suggests AMD's compute shader implementation is more efficient in this particular benchmark, even though the NVIDIA card has more shading units (10496 vs 5120) and higher FP32 throughput (31.80 TFLOPS vs 23.65 TFLOPS).

Specification Differences

The two GPUs differ in nearly every physical and electrical specification. The AMD Radeon RX 6950 XT is a desktop card built on TSMC's 7 nm process, with a die size of 520 mm² and 26,800 million transistors. The NVIDIA GeForce RTX 5090 Mobile is an integrated GPU package (IGP) built on TSMC's 5 nm process, with a die size of 378 mm² and 45,600 million transistors. The transistor density tells the story: the NVIDIA part packs 120.6M transistors per mm², while the AMD card manages 51.5M per mm².

Memory configurations diverge sharply. The AMD card has 16 GB of GDDR6 on a 256-bit bus, yielding 576.0 GB/s of bandwidth. The NVIDIA card has 24 GB of GDDR7 on the same 256-bit bus, yielding 896.0 GB/s. The clock speeds also differ: the AMD card runs at a base of 1860 MHz and a boost of 2310 MHz, while the NVIDIA card runs at a base of 990 MHz and a boost of 1515 MHz. Despite lower clocks, the NVIDIA card achieves higher absolute performance due to its newer architecture and higher transistor count.

Power and physical requirements are opposite extremes. The AMD card has a 335 W TDP, uses two 8-pin power connectors, requires a 700 W suggested PSU, and is a triple-slot card measuring 267 mm in length. The NVIDIA card has a 95 W TDP, requires no power connectors, has no suggested PSU, and is an IGP with no dimensions listed. The bus interface also differs: the AMD card uses PCIe 4.0 x16, while the NVIDIA card uses PCIe 5.0 x16.

The AMD card has 5120 shading units, 320 TMUs, and 128 ROPs. The NVIDIA card has 10496 shading units, 328 TMUs, and 112 ROPs. In terms of ray tracing, the NVIDIA card has 82 RT cores versus 80 for the AMD card. The NVIDIA card also includes 328 tensor cores, while the AMD card has none listed. Pixel and texture rates favor the AMD card: 295.7 GPixel/s and 739.2 GTexel/s versus 169.7 GPixel/s and 496.9 GTexel/s for the NVIDIA part. FP32 throughput favors the NVIDIA card: 31.80 TFLOPS versus 23.65 TFLOPS. FP16 throughput is also higher on the NVIDIA card at 31.80 TFLOPS (1:1) versus 47.31 TFLOPS (2:1) for AMD, though the AMD figure relies on a 2:1 ratio.

Architecture Differences

The architectural divide is generational. The AMD Radeon RX 6950 XT uses RDNA 2.0, AMD's second-generation graphics architecture, built on the Navi 21 chip. It belongs to the Navi II (RX 6000) generation, released on May 9, 2022, and is now end-of-life. The NVIDIA GeForce RTX 5090 Mobile uses Blackwell 2.0, NVIDIA's latest architecture, built on the GB203 chip. It belongs to the GeForce 50 Mobile generation, released on March 26, 2025, and is still active in production.

The process node difference is significant: 7 nm for AMD versus 5 nm for NVIDIA, both from TSMC. This allows the NVIDIA chip to pack 45,600 million transistors into a smaller die (378 mm²) than the AMD chip's 26,800 million transistors in a larger die (520 mm²). The transistor density difference (120.6M / mm² versus 51.5M / mm²) is a direct consequence of the process shrink.

Cache and memory architecture are not directly comparable from the data, but the memory types tell the story: GDDR6 for AMD versus GDDR7 for NVIDIA. The GDDR7 memory runs at an effective 28 Gbps, while the GDDR6 runs at 18 Gbps effective. This contributes to the NVIDIA card's higher bandwidth despite the same 256-bit bus width.

The ray tracing and tensor core counts differ. The NVIDIA card has 82 RT cores and 328 tensor cores, while the AMD card has 80 RT cores and no tensor cores listed. The tensor cores are critical for AI-accelerated workloads, and their presence gives the NVIDIA card a feature advantage that the AMD card cannot match. The shading unit count also differs dramatically: 10496 for NVIDIA versus 5120 for AMD, which explains the higher FP32 throughput despite lower clock speeds.

The API support is identical (DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4), so the architectural differences manifest entirely in performance and efficiency, not in feature parity. The NVIDIA card achieves its results at a 95 W TDP, while the AMD card requires 335 W, a 3.5x power differential that underscores the efficiency gain of the 5 nm process and Blackwell architecture.

Where Each One Wins

The NVIDIA GeForce RTX 5090 Mobile wins in modern 3D rendering and API-heavy workloads. The 3DMark Steel Nomad DX12 result (5871 vs 4235) is the strongest single indicator: this is a DX12 benchmark that reflects current game engine requirements. The Geekbench Vulkan win (198405 vs 165212) reinforces this, as Vulkan is increasingly used in cross-platform engines. The Passmark DirectX 10, DirectX 12, and DirectX 9 wins (183 vs 164, 138 vs 114, and 324 vs 303 respectively) show that the NVIDIA card is faster across the board in legacy DirectX paths, with the DirectX 12 lead being particularly large at 17.4%. The Passmark G3D win (30034 vs 28070) confirms the NVIDIA card is the better general-purpose 3D performer.

The AMD Radeon RX 6950 XT wins in specific compute and legacy API scenarios. The Passmark GPU Compute win (14199 vs 13401) is the most significant, a 6% lead that suggests AMD's compute shader pipeline is more efficient in this test. The Passmark DirectX 11 win (300 vs 269, an 11.5% lead) is important for older games and enterprise applications that still use DX11. The Geekbench OpenCL win (205998 vs 201834) is small but real, and the Passmark G2D win (1063 vs 1057) is negligible but consistent.

The broader database picture favors the AMD card in terms of average score: 58392 versus 45152, a 29% gap. The AMD card also ranks higher in percentile vs all GPUs (88th versus 84th). This means that if you average across all recorded benchmarks, the AMD Radeon RX 6950 XT is the more consistently strong performer, even though it loses the specific head-to-head tests that favor the NVIDIA card's modern architecture. For users who prioritize legacy DirectX 11 compatibility, OpenCL compute, or overall average benchmark performance, the AMD card is the safer choice. For users who prioritize DX12 and Vulkan gaming, the NVIDIA card is the clear winner.