AMD Radeon Pro Vega II vs NVIDIA GeForce RTX 5090 D Comparison

FAQ

Q: Which GPU has the higher average benchmark score, and what does that mean for overall performance?

A: The AMD Radeon Pro Vega II has a higher average benchmark score of 109,617 compared to the NVIDIA GeForce RTX 5090 D's 77,712. However, this is misleading because the AMD card's average is derived from only three Geekbench tests, while the NVIDIA card's average includes ten tests spanning multiple benchmark suites, including several Passmark tests that score much lower numerically.

Q: How do the two cards compare in raw compute performance?

A: The NVIDIA GeForce RTX 5090 D delivers 104.8 TFLOPS of FP32 compute, while the AMD Radeon Pro Vega II delivers 14.09 TFLOPS. This represents a massive difference in raw floating-point throughput, with the NVIDIA card offering roughly 7.4 times the FP32 performance.

Q: What is the difference in memory bandwidth between these two cards?

A: The AMD Radeon Pro Vega II uses 32 GB of HBM2 memory on a 4096-bit bus, providing 825.3 GB/s of bandwidth. The NVIDIA GeForce RTX 5090 D uses 32 GB of GDDR7 memory on a 512-bit bus, providing 1.79 TB/s of bandwidth, which is more than double the AMD card's bandwidth.

Q: Which card is more power-hungry according to the specifications?

A: The NVIDIA GeForce RTX 5090 D has a TDP of 575 W and a suggested PSU of 950 W, while the AMD Radeon Pro Vega II has a TDP of 475 W and a suggested PSU of 850 W. The NVIDIA card consumes more power and requires a stronger power supply.

Q: Are these cards in the same performance percentile?

A: The AMD Radeon Pro Vega II sits in the 94th percentile of all GPUs, while the NVIDIA GeForce RTX 5090 D sits in the 92nd percentile. Despite the NVIDIA card's far higher raw performance in head-to-head tests, both are near the top of the overall GPU distribution.

Q: What are the physical dimensions and slot requirements for each card?

A: The AMD Radeon Pro Vega II is a quad-slot card with an Apple MPX bus interface, while the NVIDIA GeForce RTX 5090 D is a dual-slot card measuring 304 mm in length, 137 mm in height, and 48 mm in width, using a PCIe 5.0 x16 interface and a single 16-pin power connector.

Architecture Differences

The AMD Radeon Pro Vega II is built on the Vega 20 chip using the GCN 5.1 architecture, manufactured on a 7 nm process at TSMC. It packs 13,230 million transistors into a 331 mm² die, giving a transistor density of 40.0 million transistors per square millimeter. This is a mature architecture from the Radeon Pro Mac generation, designed for professional Mac workstations, and it is now marked as end-of-life with a release date of June 2, 2019.

The NVIDIA GeForce RTX 5090 D, in contrast, is built on the GB202 chip using the Blackwell 2.0 architecture, manufactured on a 5 nm process, also at TSMC. It contains 92,200 million transistors on a 750 mm² die, achieving a transistor density of 122.9 million transistors per square millimeter. This is a current-generation part from the GeForce 50 series, released on January 29, 2025, and it remains in active production.

The architectural differences are stark. The AMD card features 4,096 shading units, 256 texture mapping units, and 64 ROPs. It has no dedicated ray tracing cores or tensor cores. The NVIDIA card features 21,760 shading units, 680 TMUs, and 176 ROPs, along with 170 ray tracing cores and 680 tensor cores. The NVIDIA card supports DirectX 12 Ultimate (12_2) and Vulkan 1.4, while the AMD card supports DirectX 12 (12_1) and Vulkan 1.3. Both cards support OpenGL 4.6.

The AMD card's FP16 performance is listed at 28.18 TFLOPS (2:1 ratio), while the NVIDIA card's FP16 performance matches its FP32 at 104.8 TFLOPS (1:1 ratio). This means the NVIDIA card does not halve its throughput for half-precision work, which is significant for compute workloads that can utilize FP16.

Head-to-Head Benchmarks

The database contains two direct head-to-head benchmark comparisons between these cards, and the NVIDIA GeForce RTX 5090 D wins both decisively.

In Geekbench OpenCL, the AMD Radeon Pro Vega II scores 99,048, while the NVIDIA GeForce RTX 5090 D scores 310,674. The NVIDIA card wins by a delta of -68.1 percent, meaning the AMD card achieves only about 32 percent of the NVIDIA card's score. This is a 3.1x advantage for the NVIDIA card in OpenCL compute.

In Geekbench Vulkan, the AMD Radeon Pro Vega II scores 99,621, while the NVIDIA GeForce RTX 5090 D scores 376,915. The NVIDIA card wins by a delta of -73.6 percent, meaning the AMD card achieves only about 26 percent of the NVIDIA card's score. This is a 3.8x advantage for the NVIDIA card in Vulkan graphics and compute.

These results align with the raw specification differences. The NVIDIA card's FP32 throughput of 104.8 TFLOPS versus 14.09 TFLOPS for the AMD card, combined with more than double the memory bandwidth, explains the large performance gap in these synthetic benchmarks.

The NVIDIA card also has a broader benchmark footprint in the database, with scores across 3DMark Steel Nomad DX12 (14,326), Passmark DirectX 9 (434), DirectX 10 (231), DirectX 11 (371), DirectX 12 (219), G2D (1,487), G3D (44,065), and GPU Compute (28,396). These additional tests provide a more complete picture of its performance across different API generations and workload types.

Specification Differences

The two cards differ across nearly every major specification category.

• Process node: AMD uses 7 nm, NVIDIA uses 5 nm.
• Transistors: AMD has 13,230 million, NVIDIA has 92,200 million.
• Die size: AMD is 331 mm², NVIDIA is 750 mm².
• Transistor density: AMD is 40.0 million per mm², NVIDIA is 122.9 million per mm².
• Base clock: AMD is 1574 MHz, NVIDIA is 2017 MHz.
• Boost clock: AMD is 1720 MHz, NVIDIA is 2407 MHz.
• Memory speed: AMD is 806 MHz (1612 Mbps effective), NVIDIA is 1750 MHz (28 Gbps effective).
• Memory type: AMD uses HBM2, NVIDIA uses GDDR7.
• Memory bus width: AMD is 4096 bit, NVIDIA is 512 bit.
• Memory bandwidth: AMD is 825.3 GB/s, NVIDIA is 1.79 TB/s.
• Shading units: AMD has 4,096, NVIDIA has 21,760.
• TMUs: AMD has 256, NVIDIA has 680.
• ROPs: AMD has 64, NVIDIA has 176.
• Ray tracing cores: AMD has none, NVIDIA has 170.
• Tensor cores: AMD has none, NVIDIA has 680.
• Pixel rate: AMD is 110.1 GPixel/s, NVIDIA is 423.6 GPixel/s.
• Texture rate: AMD is 440.3 GTexel/s, NVIDIA is 1,636.8 GTexel/s.
• FP32 performance: AMD is 14.09 TFLOPS, NVIDIA is 104.8 TFLOPS.
• FP16 performance: AMD is 28.18 TFLOPS (2:1), NVIDIA is 104.8 TFLOPS (1:1).
• TDP: AMD is 475 W, NVIDIA is 575 W.
• Slot width: AMD is quad-slot, NVIDIA is dual-slot.
• Power connectors: AMD has none listed, NVIDIA uses 1x 16-pin.
• Suggested PSU: AMD is 850 W, NVIDIA is 950 W.
• Bus interface: AMD uses Apple MPX, NVIDIA uses PCIe 5.0 x16.
• Display outputs: AMD has 1x HDMI 2.0b and 4x Thunderbolt, NVIDIA has 1x HDMI 2.1b and 3x DisplayPort 2.1b.
• DirectX support: AMD is 12 (12_1), NVIDIA is 12 Ultimate (12_2).
• Vulkan support: AMD is 1.3, NVIDIA is 1.4.
• Release date: AMD is June 2, 2019, NVIDIA is January 29, 2025.
• Production status: AMD is end-of-life, NVIDIA is active.

Where Each One Wins

The NVIDIA GeForce RTX 5090 D wins every head-to-head benchmark in the database. It has a 3.1x advantage in OpenCL and a 3.8x advantage in Vulkan. Its raw compute specifications are overwhelmingly superior: 7.4x more FP32 throughput, 4.7x more shading units, 3.9x more TMUs, 2.75x more ROPs, and double the memory bandwidth. The NVIDIA card also brings dedicated ray tracing and tensor cores, which the AMD card lacks entirely.

The AMD Radeon Pro Vega II's advantages are narrower. It has a higher average benchmark score (109,617 versus 77,712), though this is due to the different test sets used. It also sits one percentile higher in the overall GPU distribution (94th versus 92nd). The AMD card has a wider memory bus (4096 bit versus 512 bit) and uses HBM2 memory, which may be preferable in certain professional memory-bound workloads. It also has a lower TDP at 475 W versus 575 W, and its quad-slot design with Apple MPX interface is tailored for specific Mac Pro chassis configurations.

The AMD card's nearest rivals in the database are the AMD Radeon PRO W7900 (average score 110,725, delta -1 percent), the AMD Radeon Pro W6600X (107,342, delta 2.1 percent), and the AMD Radeon Pro Vega II Duo (106,750, delta 2.7 percent). This shows the Pro Vega II is closely matched with other professional AMD cards. The NVIDIA RTX 5090 D's nearest rivals include the AMD Radeon RX 6650M XT (76,904, delta 1.1 percent) and the AMD Radeon RX 6850M XT (78,940, delta -1.6 percent), indicating its average score is pulled down by the low Passmark sub-scores despite its massive compute advantage.

The Verdict

The data makes the choice clear for compute-heavy workloads. The NVIDIA GeForce RTX 5090 D is in a different performance class entirely, with roughly 3 to 4 times the benchmark scores in direct comparisons and 7.4 times the raw FP32 throughput. Anyone needing maximum OpenCL or Vulkan performance should choose the NVIDIA card without hesitation.

The AMD Radeon Pro Vega II's only defensible use case is within the Apple MPX ecosystem. Its bus interface, quad-slot form factor, and 4x Thunderbolt outputs are designed for specific Mac Pro configurations. For users locked into that platform, the Pro Vega II remains a competent card, and its 94th percentile ranking shows it is still competitive against the broader GPU field.

However, the NVIDIA card is the superior general-purpose GPU. It is dual-slot, uses a standard PCIe 5.0 x16 interface, supports newer DirectX and Vulkan versions, includes ray tracing and tensor cores, and delivers substantially higher performance in every measured test. Its 92nd percentile ranking understates its capability because the average includes multiple low-scoring Passmark DirectX tests that drag the mean down.

The verdict is straightforward: the NVIDIA GeForce RTX 5090 D wins on raw performance, modern features, and efficiency of execution. The AMD Radeon Pro Vega II wins only on platform compatibility for Apple MPX systems and a slightly higher percentile ranking. For any new build or upgrade not constrained to Apple hardware, the NVIDIA card is the data-backed choice.