AMD Radeon PRO W7900 vs NVIDIA GeForce RTX 5090 D Comparison

Head-to-Head Benchmarks

The benchmark data records two direct comparisons between the AMD Radeon PRO W7900 and the NVIDIA GeForce RTX 5090 D, and in both cases the NVIDIA card takes a commanding lead. In Geekbench OpenCL, the RTX 5090 D scores 310,674 against the Radeon PRO W7900's 84,379, a delta of 72.8 percent in NVIDIA's favor. The Vulkan test shows a similar pattern: the RTX 5090 D reaches 376,915 while the AMD card manages 137,070, a 63.6 percent advantage for NVIDIA. These are not marginal wins; they are decisive margins that appear consistently across both compute APIs.

The OpenCL result is particularly telling because it reflects general compute throughput, the kind of workload that matters for rendering, simulation, and data processing. A 72.8 percent gap is substantial, and it aligns with the raw specifications: the RTX 5090 D carries 21,760 shading units and 680 tensor cores, while the Radeon PRO W7900 has 6,144 shading units and no tensor core field. The FP32 throughput tells the same story, 104.8 TFLOPS for NVIDIA versus 61.32 TFLOPS for AMD, a 1.71x difference. The Vulkan result, while narrower, still leaves AMD trailing by more than half.

It is importantly the head-to-head table records zero wins for the AMD card. No benchmark in the shared dataset favors the Radeon PRO W7900. The average benchmark score across all recorded tests also reflects this: the AMD card averages 110,725, while the RTX 5090 D averages 77,712. That average, however, is pulled down by the NVIDIA card's inclusion of several legacy DirectX and Passmark tests, which score very low: Passmark DirectX 9 at 434, DirectX 10 at 231, DirectX 11 at 371, and DirectX 12 at 219. These legacy tests are not part of the direct comparison, but they do affect the aggregate. The Geekbench scores, which are the common ground, are unambiguous.

Where Each One Wins

The recorded data shows no benchmark category where the AMD Radeon PRO W7900 outperforms the NVIDIA GeForce RTX 5090 D. Every direct comparison, both OpenCL and Vulkan, goes to NVIDIA. That said, the two cards occupy different design targets, and the data supports distinct use cases even if the raw scores favor one side.

The RTX 5090 D excels in raw compute and modern API performance. Its Vulkan score of 376,915 is the highest recorded value in the comparison, and its OpenCL score of 310,674 is more than 3.6 times the AMD card's OpenCL result. For workloads that lean heavily on FP32 compute, shader throughput, or tensor operations, the NVIDIA card is the clear choice. The 680 tensor cores and 170 RT cores provide dedicated hardware for AI inference and ray tracing, and the 1.79 TB/s memory bandwidth ensures that data movement does not bottleneck those cores.

The Radeon PRO W7900, on the other hand, offers a larger memory pool: 48 GB of GDDR6 versus 32 GB of GDDR7. For datasets that exceed 32 GB, the AMD card can hold more working data on the GPU, avoiding host-device transfers. The memory bus is narrower at 384-bit versus 512-bit, but the capacity advantage remains real. The card also draws less power, 295 W versus 575 W, and uses two 8-pin connectors instead of a single 16-pin connector, which may simplify integration into existing workstations with older power supplies. The suggested PSU is 600 W for AMD and 950 W for NVIDIA, a meaningful difference in system-level power planning.

In terms of form factor, the Radeon PRO W7900 is a triple-slot card at 280 mm long, 110 mm high, and 51 mm wide. The RTX 5090 D is dual-slot at 304 mm long, 137 mm high, and 48 mm wide. The NVIDIA card is longer and taller but thinner, which matters for chassis clearance and adjacent PCIe slot access. The AMD card is shorter and lower but wider, so it occupies more vertical space above the slot.

Architecture Differences

The two GPUs come from different architectural lineages. The AMD Radeon PRO W7900 uses the Navi 31 chip built on RDNA 3.0, with the codename Plum Bonito. It belongs to the Radeon Pro Navi (Navi III Series) generation and was released on May 25, 2023. The NVIDIA GeForce RTX 5090 D uses the GB202 chip built on Blackwell 2.0, part of the GeForce 50 series, and was released on January 29, 2025.

Both are fabricated on a 5 nm process at TSMC, but the physical characteristics differ. The AMD chip contains 57,700 million transistors on a 529 mm² die, giving a transistor density of 109.1 million per square millimeter. The NVIDIA chip packs 92,200 million transistors on a 750 mm² die, a density of 122.9 million per square millimeter. The NVIDIA die is 41.8 percent larger by area and holds 59.8 percent more transistors.

Memory technology also diverges. The Radeon PRO W7900 uses GDDR6 with 48 GB capacity, a 384-bit bus, and 864.0 GB/s of bandwidth. The RTX 5090 D uses GDDR7 with 32 GB capacity, a 512-bit bus, and 1.79 TB/s of bandwidth. The NVIDIA card delivers more than double the bandwidth, 1.79 TB/s versus 864 GB/s, even though it has less capacity. Effective memory clock is 18 Gbps on the AMD card and 28 Gbps on the NVIDIA card.

Compute resources differ sharply. The AMD card has 6,144 shading units, 384 texture mapping units, 192 ROPs, and 96 RT cores. The NVIDIA card has 21,760 shading units, 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores. The pixel rate is 479.0 GPixel/s on AMD versus 423.6 GPixel/s on NVIDIA, a rare specification where AMD leads, though the texture rate is far higher on NVIDIA: 1,636.8 GTexel/s versus 958.1 GTexel/s. FP16 and FP32 both run at 1:1 ratios on both cards, but the absolute numbers favor NVIDIA: 104.8 TFLOPS versus 61.32 TFLOPS for both precision levels.

Clock behavior also differs. The AMD card has a base clock of 1760 MHz and a boost clock of 2495 MHz. The NVIDIA card has a base clock of 2017 MHz and a boost clock of 2407 MHz. AMD has a higher boost frequency, but NVIDIA starts from a higher base, and the overall throughput is determined by the combination of clock speed and shader count.

Interface and connectivity show a generational gap. The Radeon PRO W7900 uses PCIe 4.0 x16, while the RTX 5090 D uses PCIe 5.0 x16. Display outputs are three DisplayPort 2.1 and one mini-DisplayPort 2.1 on AMD, versus one HDMI 2.1b and three DisplayPort 2.1b on NVIDIA. Both support DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4.

The launch MSRP for the AMD Radeon PRO W7900 is 3,999 USD. The launch MSRP for the NVIDIA GeForce RTX 5090 D is 2,299 USD.

FAQ

Q: Which card has higher raw compute throughput?

A: The NVIDIA GeForce RTX 5090 D records 104.8 TFLOPS in both FP32 and FP16, while the AMD Radeon PRO W7900 records 61.32 TFLOPS in both. The Geekbench OpenCL scores reflect this, with NVIDIA at 310,674 versus AMD at 84,379.

Q: Does the AMD card have any advantage in memory capacity?

A: Yes, the Radeon PRO W7900 has 48 GB of GDDR6 memory, while the RTX 5090 D has 32 GB of GDDR7. The NVIDIA card has a wider 512-bit bus and higher bandwidth at 1.79 TB/s, versus 864.0 GB/s on AMD.

Q: Which card is more power-hungry?

A: The RTX 5090 D has a TDP of 575 W and requires a 950 W suggested PSU, while the Radeon PRO W7900 has a TDP of 295 W and a 600 W suggested PSU. The NVIDIA card uses a single 16-pin connector; the AMD card uses two 8-pin connectors.

Q: How do the cards compare in Vulkan performance?

A: The RTX 5090 D scores 376,915 in Geekbench Vulkan, compared to 137,070 for the Radeon PRO W7900, a 63.6 percent difference in NVIDIA's favor.

Q: What are the physical dimensions?

A: The Radeon PRO W7900 measures 280 mm by 110 mm by 51 mm and is triple-slot. The RTX 5090 D measures 304 mm by 137 mm by 48 mm and is dual-slot.

Q: Which card has tensor cores?

A: Only the RTX 5090 D has tensor cores, with 680 of them, along with 170 RT cores. The Radeon PRO W7900 has no tensor core field listed and 96 RT cores.

The Verdict

The data points to a clear winner for compute-heavy workloads. The NVIDIA GeForce RTX 5090 D beats the AMD Radeon PRO W7900 by 72.8 percent in OpenCL and 63.6 percent in Vulkan. It offers more than 3.5 times the shading units, more than double the FP32 throughput, and more than double the memory bandwidth. It also arrives with a lower launch MSRP of 2,299 USD against AMD's 3,999 USD. For any workload that depends on raw shader throughput, tensor operations, or memory bandwidth, the RTX 5090 D is the stronger choice.

The AMD Radeon PRO W7900 still has a role. Its 48 GB memory pool is 50 percent larger than the NVIDIA card's 32 GB, which matters for very large datasets that must reside on the GPU. Its 295 W TDP and 600 W suggested PSU make it easier to integrate into systems with modest power budgets, and its triple-slot cooler may be preferable in cases where acoustic or thermal behavior under sustained load is a concern. The pixel rate of 479.0 GPixel/s also edges out the NVIDIA card's 423.6 GPixel/s.

But the head-to-head results are decisive. Zero wins for AMD in the recorded comparisons, and the NVIDIA card dominates every shared benchmark. The RTX 5090 D is the pick for users who prioritize performance, while the Radeon PRO W7900 suits those who need maximum memory capacity and lower power draw. The percentile ranking versus all GPUs places AMD at 94 and NVIDIA at 92, but that aggregate includes the NVIDIA card's legacy DirectX scores, which drag down its average. On the tests that matter for modern workloads, NVIDIA leads by a wide margin.

Specification Differences

| Specification | AMD Radeon PRO W7900 | NVIDIA GeForce RTX 5090 D |

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

| Architecture | RDNA 3.0 | Blackwell 2.0 |

| Chip | Navi 31 | GB202 |

| Generation | Radeon Pro Navi (Navi III Series) | GeForce 50 |

| Process node | 5 nm | 5 nm |

| Transistors | 57,700 million | 92,200 million |

| Die size | 529 mm² | 750 mm² |

| Transistor density | 109.1M / mm² | 122.9M / mm² |

| Base clock | 1760 MHz | 2017 MHz |

| Boost clock | 2495 MHz | 2407 MHz |

| Memory clock | 2250 MHz 18 Gbps effective | 1750 MHz 28 Gbps effective |

| Memory size | 48 GB | 32 GB |

| Memory type | GDDR6 | GDDR7 |

| Memory bus width | 384 bit | 512 bit |

| Memory bandwidth | 864.0 GB/s | 1.79 TB/s |

| Shading units | 6144 | 21760 |

| TMUs | 384 | 680 |

| ROPs | 192 | 176 |

| RT cores | 96 | 170 |

| Tensor cores | null | 680 |

| Pixel rate | 479.0 GPixel/s | 423.6 GPixel/s |

| Texture rate | 958.1 GTexel/s | 1,636.8 GTexel/s |

| FP32 | 61.32 TFLOPS | 104.8 TFLOPS |

| FP16 | 61.32 TFLOPS (1:1) | 104.8 TFLOPS (1:1) |

| TDP | 295 W | 575 W |

| Slot width | Triple-slot | Dual-slot |

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

| Suggested PSU | 600 W | 950 W |

| Bus interface | PCIe 4.0 x16 | PCIe 5.0 x16 |

| Display outputs | 3x DisplayPort 2.1, 1x mini-DisplayPort 2.1 | 1x HDMI 2.1b, 3x DisplayPort 2.1b |

| Length | 280 mm 11 inches | 304 mm 12 inches |

| Height | 110 mm 4.3 inches | 137 mm 5.4 inches |

| Width | 51 mm 2 inches | 48 mm 1.9 inches |

| Release date | 2023-05-25 | 2025-01-29 |

| Launch MSRP | 3,999 USD | 2,299 USD |