AMD Radeon PRO V620 vs NVIDIA GeForce RTX 5090 D Comparison

Head-to-Head Benchmarks

The recorded data shows a decisive advantage for the NVIDIA GeForce RTX 5090 D in the two shared benchmark tests. In Geekbench OpenCL, the RTX 5090 D scores 310,674 against the AMD Radeon PRO V620's 128,580. That is a 58.6% deficit for the AMD card, meaning NVIDIA's result is roughly 2.4 times higher. The gap widens slightly in Geekbench Vulkan, where the RTX 5090 D posts 376,915 versus 144,364 for the Radeon PRO V620, a 61.7% difference.

The head-to-head table records zero wins for the AMD Radeon PRO V620 and two wins for the NVIDIA GeForce RTX 5090 D. This is not a close contest in raw compute performance. The NVIDIA card dominates both OpenCL and Vulkan workloads by wide margins. The Vulkan gap is particularly telling, as it often reflects driver efficiency and architectural scaling rather than just raw shader count.

For context, the AMD Radeon PRO V620's average benchmark score across its two entries is 136,472, which places it in the 96th percentile of all GPUs in the database. The NVIDIA GeForce RTX 5090 D's average score is 77,112, but that figure is skewed by its inclusion of several Passmark tests with relatively low scores (DX9: 434, DX10: 231, DX11: 371, DX12: 219, G2D: 1487). Its strongest Passmark result is G3D at 44,065, and its GPU compute score is 28,396. The Geekbench scores are the more representative comparison here, and they strongly favor NVIDIA.

Architecture Differences

The two cards come from very different design philosophies and generations. The AMD Radeon PRO V620 uses the Navi 21 chip built on RDNA 2.0 architecture, fabricated on a 7 nm process at TSMC. It packs 26,800 million transistors into a 520 mm² die, giving a transistor density of 51.5 million per square millimeter. The NVIDIA GeForce RTX 5090 D uses the GB202 chip on Blackwell 2.0 architecture, fabricated on a 5nm process at TSMC. It contains 92,200 million transistors across a 750 mm² die, with a density of 122.9 million per square millimeter. NVIDIA's chip has more than three times the transistor count and more than double the density.

Core counts tell a similar story. The Radeon PRO V620 has 4,608 shading units, 288 texture mapping units, 128 raster operation pipelines, and 72 ray tracing cores. The RTX 5090 D has 21,760 shading units, 680 TMUs, 176 ROPs, 170 RT cores, and 680 tensor cores. The NVIDIA card has no direct tensor core equivalent on the AMD side, as the Radeon PRO V620 lists no tensor cores. The shading unit advantage is massive: NVIDIA has 4.7 times more.

Clock speeds differ as well. The AMD card runs a base of 1825 MHz and a boost of 2200 MHz. The NVIDIA card has a base of 2017 MHz and a boost of 2407 MHz. So NVIDIA is both higher clocked and much wider, a combination that drives its benchmark lead.

Memory architecture is another major divergence. Both cards have 32 GB of memory, but the type and configuration differ. The Radeon PRO V620 uses GDDR6 on a 256-bit bus, achieving 512.0 GB/s bandwidth. The RTX 5090 D uses GDDR7 on a 512-bit bus, achieving 1.79 TB/s. That is over triple the bandwidth: 1.79 TB/s versus 0.512 TB/s. The memory chips run at 2000 MHz (16 Gbps effective) on the AMD side and 1750 MHz (28 Gbps effective) on the NVIDIA side. NVIDIA's memory bus width is double, and its effective speed is higher, resulting in a massive bandwidth advantage.

Compute rates follow the same pattern. The Radeon PRO V620 delivers 20.28 TFLOPS FP32 and 40.55 TFLOPS FP16 (2:1 ratio). The RTX 5090 D delivers 104.8 TFLOPS FP32 and 104.8 TFLOPS FP16 (1:1 ratio). The FP16 ratio difference is notable: AMD halves its FP16 rate relative to FP32, while NVIDIA maintains parity. Pixel and texture rates also favor NVIDIA: 423.6 GPixel/s and 1,636.8 GTexel/s versus 281.6 GPixel/s and 633.6 GTexel/s.

There are physical differences too. The AMD card is 267mm long, 120mm tall, and 50mm wide, requiring dual-slot space and two 8-pin power connectors, with a suggested PSU of 700W and a TDP of 300W. The NVIDIA card is 304mm long, 137mm tall, and 48mm wide, also dual-slot, using a single 16-pin connector, a suggested PSU of 950W, and a TDP of 575W. The NVIDIA card is longer and taller but slightly thinner, and it demands nearly double the power budget.

Interface differences: AMD uses PCIe 4.0 x16, NVIDIA uses PCIe 5.0 x16. Display outputs are a major split. The Radeon PRO V620 has no display outputs, which is typical for a compute-oriented server card. The RTX 5090 D has one HDMI 2.1b and three DisplayPort 2.1b outputs. The AMD card is end-of-life, released November 2021, while the NVIDIA card is active with a January 2025 release. The NVIDIA card lists a launch MSRP of 2,299 USD, which can be stated once.

Where Each One Wins

The benchmark data is unambiguous: the NVIDIA GeForce RTX 5090 D wins every shared test. In Geekbench OpenCL and Vulkan, the NVIDIA card leads by 58.6% and 61.7%, respectively. There is no test in the recorded head-to-head where the AMD Radeon PRO V620 comes out ahead. Therefore, in any compute workload that resembles those synthetic benchmarks, the NVIDIA card should deliver materially higher performance.

However, the AMD card does have a niche. Its 300W TDP is much lower than NVIDIA's 575W, which matters for power-constrained environments. The AMD card also uses two standardized 8-pin connectors, which are more common on older power supplies than the 16-pin connector required by the NVIDIA card. For a system with a 700W PSU, the AMD card fits; the NVIDIA card needs a 950W PSU. So in a small form factor or an existing server with limited power headroom, the Radeon PRO V620 is the only reasonable option despite its lower performance.

The AMD card also has a smaller physical footprint: 267mm versus 304mm in length. That can matter in compact chassis. Its lack of display outputs is a non-issue in compute workloads, and its PCIe 4.0 x16 interface is backward compatible with most systems, though the NVIDIA card's PCIe 5.0 offers more bandwidth.

For rendering, gaming, or any display-related workload, the NVIDIA card is the clear choice due to its display outputs. The Radeon PRO V620 cannot output video at all. If the use case is purely server-side compute (e.g., rendering without a monitor), the AMD card's lack of outputs is irrelevant.

One more point: the NVIDIA card has tensor cores, which the AMD card lacks. This matters for AI and machine learning workloads that rely on tensor operations. The AMD card has no tensor core support, so any tensor-heavy workload will be limited to FP32 or FP16.

FAQ

Q: Which card has the higher Geekbench Vulkan score?

A: The NVIDIA GeForce RTX 5090 D scores 376,915 in Geekbench Vulkan, versus 144,364 for the AMD Radeon PRO V620, a 61.7% advantage.

Q: What are the memory bandwidth figures?

A: The AMD card has 512.0 GB/s across a 256-bit GDDR6 bus. The NVIDIA card has 1.79 TB/s across a 512-bit GDDR7 bus.

Q: Does the AMD Radeon PRO V620 have any display outputs?

A: No, the AMD card has no display outputs. The NVIDIA card has one HDMI 2.1b and three DisplayPort 2.1b outputs.

Q: What is the transistor count and process node for each?

A: The AMD card uses 26,800 million transistors on a 7nm process. The NVIDIA card uses 92,200 million transistors on a 5nm process.

Q: Which card has a higher FP32 throughput?

A: The NVIDIA card achieves 104.8 TFLOPS FP32, while the AMD card achieves 20.28 TFLOPS FP32.

Q: Is the AMD card still in production?

A: No, the AMD Radeon PRO V620 is end-of-life. The NVIDIA GeForce RTX 5090 D is active.

The Verdict

Based on the recorded data, the NVIDIA GeForce RTX 5090 D is the superior product in raw performance. It wins both head-to-head tests with wide margins: 58.6% in OpenCL and 61.7% in Vulkan. It has more than triple the transistors, nearly five times the shading units, more than triple the memory bandwidth, and over five times the FP32 throughput. It also supports tensor cores, which the AMD card lacks entirely.

Users choosing between these two should pick the NVIDIA card unless there is a specific constraint. The AMD card only makes sense in a system with a 300W power ceiling, a 700W PSU, or a requirement for dual 8-pin power connectors. Its 32 GB memory is the same capacity as the NVIDIA card, but the bandwidth difference is enormous: 0.512 TB/s versus 1.79 TB/s.

The NVIDIA card is also the only one that can connect to a display, so any workstation requiring a monitor must use it. The AMD card is a compute-only accelerator with no video output. Its lower power draw (300W vs 575W) and smaller physical footprint (267mm vs 304mm) are the only advantages.

The data does not support a scenario where the AMD Radeon PRO V620 outperforms the NVIDIA GeForce RTX 5090 D. The NVIDIA card is faster in both compute APIs, and its architecture is significantly more advanced with better density and bandwidth. The only reason to choose the AMD card is a specific power or PSU constraint, not performance. For any user who can supply a 950W PSU and a 16-pin connector, the NVIDIA GeForce RTX 5090 D is the definitive choice.