NVIDIA GeForce RTX 5090 vs NVIDIA GeForce RTX 5090 D Comparison

The NVIDIA GeForce RTX 5090 and RTX 5090 D are, on paper, near-identical twins. Both use the GB202 chip on TSMC's 5 nm process, pack 92,200 million transistors into a 750 mm² die, and share the same base clock of 2017 MHz, boost clock of 2407 MHz, 32 GB of GDDR7 memory on a 512-bit bus, and a 575 W TDP. The benchmark data, however, reveals a clear split in behavior. The standard RTX 5090 holds a razor-thin 0.1% lead in average benchmark score (84306 vs 84241) and wins 3 of the 10 head-to-head tests, while the RTX 5090 D wins 7. The D is also physically wider (48 mm vs 40 mm) and carries a different launch MSRP (2,299 USD vs 1,999 USD). This is not a story of a cut-down chip; it is a story of two cards that trade blows across different API generations, and the right pick depends entirely on which benchmark suite matches your workload.

The Verdict

From the data alone, the standard RTX 5090 is the choice for modern DirectX 12 workloads and cross-platform compute. It wins 3DMark Steel Nomad DX12 (14411 vs 14326, 0.6% ahead), Geekbench OpenCL (380114 vs 375966, 1.1% ahead), and Geekbench Vulkan (379571 vs 376915, 0.7% ahead). If your primary concern is the latest DX12 games or OpenCL/Vulkan compute tasks, the standard card is the safer bet. The RTX 5090 D, on the other hand, dominates the Passmark suite: it wins all five DirectX tests (DX9, DX10, DX11, DX12) plus Passmark G2D, G3D, and GPU Compute. Its largest margin is a 15.5% victory in Passmark DirectX 12 (219 vs 185) and a 10% win in Passmark G3D (44065 vs 39650). The D also has a higher launch MSRP (2,299 USD vs 1,999 USD) and a wider 48 mm profile. If your application relies on Passmark's specific rendering paths or legacy DirectX APIs, the D is the data-backed winner. Both cards sit at the 94th percentile of all GPUs, so neither is a weak option—but the D wins more individual tests, while the standard card wins the aggregate average by a hair.

Where Each One Wins

The standard RTX 5090 wins exclusively in modern, cross-API compute and the newest DX12 test. Its three victories are: 3DMark Steel Nomad DX12 (0.6% ahead), Geekbench OpenCL (1.1% ahead), and Geekbench Vulkan (0.7% ahead). These are tests that stress raw shader throughput and modern API efficiency, and the standard card's 104.8 TFLOPS FP32 and 104.8 TFLOPS FP16 (1:1) clearly translate to a slight edge here. The RTX 5090 D wins every Passmark test, which is a broader sweep. Its margins range from 2.2% in DirectX 10 (231 vs 226) to 15.5% in DirectX 12 (219 vs 185). It also wins Passmark G2D (1487 vs 1413, 5% ahead), G3D (44065 vs 39650, 10% ahead), and GPU Compute (28396 vs 26756, 5.8% ahead). The D's Passmark DirectX 11 win is 8.1% (371 vs 341), and its DirectX 9 win is 9% (434 vs 395). The pattern is unmistakable: the D is tuned for Passmark's legacy and compute-heavy workloads, while the standard card is better aligned with modern DX12 and Geekbench's OpenCL/Vulkan paths. If you rely on Passmark for stability or legacy compatibility, the D is the one; if you use modern game engines or compute frameworks, the standard card edges ahead.

Architecture Differences

The FACT PACK shows zero architectural differences between the two cards in every core specification. Both use the GB202 chip, Blackwell 2.0 architecture, and the same 5 nm TSMC process. Transistor count is identical at 92,200 million, die size is 750 mm², and transistor density is 122.9M / mm². Shading units are 21760 on both, with 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores. Clock speeds are the same: base 2017 MHz, boost 2407 MHz, and memory at 1750 MHz (28 Gbps effective). Memory configuration is identical: 32 GB GDDR7, 512-bit bus, 1.79 TB/s bandwidth. Pixel rate (423.6 GPixel/s), texture rate (1,636.8 GTexel/s), FP32 (104.8 TFLOPS), and FP16 (104.8 TFLOPS 1:1) are all the same. TDP is 575 W for both, with the same dual-slot width, 1x 16-pin power connector, and 950 W suggested PSU. Both use PCIe 5.0 x16, have the same display outputs (1x HDMI 2.1b, 3x DisplayPort 2.1b), and support the same APIs (DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4). The only physical difference is the card width: the standard RTX 5090 is 40 mm (1.6 inches) wide, while the RTX 5090 D is 48 mm (1.9 inches) wide. Length and height are identical (304 mm / 12 inches and 137 mm / 5.4 inches). The launch MSRP also differs (1,999 USD vs 2,299 USD), but no internal specification differs. This means the benchmark differences must stem from firmware or driver-level tuning, not hardware configuration.

FAQ

Q: Which card has the higher average benchmark score?

A: The standard NVIDIA GeForce RTX 5090 has a slightly higher average benchmark score of 84306, compared to 84241 for the RTX 5090 D—a 0.1% lead.

Q: How many head-to-head tests does each card win?

A: The standard RTX 5090 wins 3 tests, while the RTX 5090 D wins 7 tests out of the 10 head-to-head benchmarks.

Q: What is the biggest single-test margin between the two?

A: The RTX 5090 D wins Passmark DirectX 12 by 15.5% (219 vs 185). The standard card's largest win is in Geekbench OpenCL at 1.1% (380114 vs 375966).

Q: Are the core counts different?

A: No. Both cards have 21760 shading units, 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores.

Q: What are the physical dimensions?

A: Both are 304 mm long and 137 mm high. The standard RTX 5090 is 40 mm wide, while the RTX 5090 D is 48 mm wide.

Q: What is the launch MSRP of each card?

A: The RTX 5090 has a launch MSRP of 1,999 USD, and the RTX 5090 D has a launch MSRP of 2,299 USD.

Head-to-Head Benchmarks

The biggest win for the standard RTX 5090 is in Geekbench OpenCL, where it scores 380114 against 375966, a 1.1% advantage. Its Geekbench Vulkan result is close behind: 379571 vs 376915, a 0.7% lead. In 3DMark Steel Nomad DX12, the standard card scores 14411 vs 14326, a 0.6% win. These three victories are all modern-API compute or the latest DX12 path. The RTX 5090 D's largest win is Passmark DirectX 12: 219 vs 185, a massive 15.5% delta. Its Passmark G3D result is 44065 vs 39650, a 10% lead, and its DirectX 9 win is 434 vs 395, a 9% margin. The D also wins Passmark DirectX 11 (371 vs 341, 8.1%), Passmark GPU Compute (28396 vs 26756, 5.8%), Passmark G2D (1487 vs 1413, 5%), and Passmark DirectX 10 (231 vs 226, 2.2%). The standard card never wins a Passmark test, and the D never wins a Geekbench or 3DMark test. The deltaPct values confirm the direction: positive for the standard card (0.6, 1.1, 0.7) and negative for the D (ranging from -2.2 to -15.5). This is a clean split—modern API workloads favor the standard card, while legacy and Passmark-specific paths favor the D.

Specification Differences

The only fields where the two cards differ are the physical width and the launch MSRP. The standard NVIDIA GeForce RTX 5090 has a width of 40 mm (1.6 inches), while the NVIDIA GeForce RTX 5090 D is 48 mm (1.9 inches) wide. All other dimensions—length (304 mm / 12 inches) and height (137 mm / 5.4 inches)—are identical. The launch MSRP is 1,999 USD for the RTX 5090 and 2,299 USD for the RTX 5090 D. Every other listed specification is exactly the same: chip (GB202), architecture (Blackwell 2.0), process node (5 nm), foundry (TSMC), transistors (92,200 million), die size (750 mm²), transistor density (122.9M / mm²), clocks (2017 MHz base, 2407 MHz boost, 1750 MHz memory / 28 Gbps effective), memory (32 GB GDDR7, 512-bit, 1.79 TB/s), shading units (21760), TMUs (680), ROPs (176), RT cores (170), tensor cores (680), pixel rate (423.6 GPixel/s), texture rate (1,636.8 GTexel/s), FP32 (104.8 TFLOPS), FP16 (104.8 TFLOPS 1:1), TDP (575 W), slot width (dual-slot), power connectors (1x 16-pin), suggested PSU (950 W), bus interface (PCIe 5.0 x16), display outputs (1x HDMI 2.1b, 3x DisplayPort 2.1b), APIs (DirectX 12 Ultimate, OpenGL 4.6, Vulkan 1.4), release date (2025-01-29), predecessor (GeForce 40), and production status (Active). The benchmark differences therefore cannot be attributed to any hardware spec listed; they are a product of firmware or driver behavior, and the wider chassis of the D suggests a different cooler design, though no thermal data is provided to confirm.