AMD Radeon PRO V620 vs NVIDIA PG506-232 Comparison

The NVIDIA PG506-232 and AMD Radeon PRO V620 are both dual-slot, no-display accelerators aimed at the server market, but they target very different performance and feature profiles. The PG506-232, built on the Ampere architecture, delivers a dominant OpenCL score, while the V620, using RDNA 2.0, counters with a higher raw FP32 throughput and more memory capacity. The data reveals a clear split: the NVIDIA card excels in compute workloads as measured, while the AMD card offers distinct architectural advantages in shading and memory size.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The NVIDIA PG506-232 has a significantly higher average benchmark score of 225,124, placing it in the 99th percentile of all GPUs. The AMD Radeon PRO V620 scores 136,472 on average, which puts it in the 96th percentile.

Q: How large is the performance gap in the head-to-head OpenCL benchmark?

A: In the Geekbench OpenCL test, the NVIDIA PG506-232 scores 225,124, which is 75.1% higher than the AMD Radeon PRO V620's score of 128,580. This is a decisive win for the NVIDIA card.

Q: Which GPU offers more memory capacity?

A: The AMD Radeon PRO V620 comes with 32 GB of GDDR6 memory, which is larger than the 24 GB of HBM2 memory found on the NVIDIA PG506-232.

Q: What are the key differences in memory technology?

A: The NVIDIA PG506-232 uses HBM2 memory with a 3072-bit bus, providing a bandwidth of 933.1 GB/s. The AMD Radeon PRO V620 uses GDDR6 on a 256-bit bus, offering a lower bandwidth of 512.0 GB/s.

Q: Which GPU has a higher boost clock speed?

A: The AMD Radeon PRO V620 has a much higher boost clock of 2200 MHz, compared to the NVIDIA PG506-232's boost clock of 1440 MHz. The AMD card also has a higher base clock of 1825 MHz versus 930 MHz.

Q: Do these cards support modern graphics APIs?

A: The AMD Radeon PRO V620 supports DirectX 12 Ultimate (12_2), OpenGL 4.6, and Vulkan 1.4. The NVIDIA PG506-232 has no listed API support in the data, reflecting its compute-focused design.

Architecture Differences

The NVIDIA PG506-232 is built on the Ampere architecture using the GA100 chip, fabricated on a 7 nm process at TSMC. This is a massive chip with 54,200 million transistors on an 826 mm² die, yielding a transistor density of 65.6M per mm². It is part of the Server Ampere (Axx) generation. The AMD Radeon PRO V620 uses the RDNA 2.0 architecture with the Navi 21 chip, also on a 7 nm TSMC process, but with a smaller 26,800 million transistors on a 520 mm² die, resulting in a lower density of 51.5M per mm². It belongs to the Radeon Pro Navi (Navi II Series) generation.

The core configurations diverge significantly. The NVIDIA card has 3584 shading units, 224 TMUs, and 96 ROPs, supplemented by 224 tensor cores. It has no dedicated ray tracing cores. In contrast, the AMD card has 4608 shading units, 288 TMUs, and 128 ROPs, along with 72 ray tracing cores, but lacks tensor cores. This points to the AMD card being more oriented toward graphics and ray tracing workloads, while the NVIDIA card is designed for AI and compute tasks with its tensor core array.

Memory architecture is another major difference. The PG506-232 uses 24 GB of HBM2 with a 3072-bit bus, delivering 933.1 GB/s of bandwidth. The V620 uses 32 GB of GDDR6 with a 256-bit bus, providing 512.0 GB/s. The NVIDIA card’s higher bandwidth is a clear advantage for memory-intensive compute, while the AMD card offers more capacity for large datasets. The transistor density difference (65.6M vs 51.5M per mm²) also suggests the GA100 die is more efficiently packed, reflecting its server-grade compute focus versus the more graphics-oriented Navi 21.

Head-to-Head Benchmarks

The only direct benchmark comparison available is a single Geekbench OpenCL test, and the result is a landslide. The NVIDIA PG506-232 scores 225,124, while the AMD Radeon PRO V620 scores 128,580. This represents a 75.1% advantage for the NVIDIA card in raw compute performance as measured by OpenCL. The data shows a clear winner in this metric, with the PG506-232 delivering a score that is nearly double that of its rival. This result is consistent with the NVIDIA card’s positioning in the 99th percentile of all GPUs, compared to the V620’s 96th percentile.

To put the scores in context, the PG506-232’s nearest rivals include the NVIDIA L20 with a score of 251,147, which is 10.4% higher, and the NVIDIA A100 PCIe 80 GB at 207,124, which is 8.7% lower. The AMD V620’s closest competitors are all within a tight 1% band: the AMD Radeon Pro W6800X Duo at 135,774 (0.5% higher), the AMD Radeon PRO W6800 at 135,396 (0.8% higher), the NVIDIA A10M at 135,230 (0.9% higher), and the NVIDIA RTX 4000 Ada Generation at 135,218 (0.9% higher). This suggests the V620 is firmly in a mid-tier compute bracket, while the PG506-232 sits at the high end, far ahead of most rivals except the L20.

The V620, however, counters with a higher peak FP32 performance of 20.28 TFLOPS versus the PG506-232’s 10.32 TFLOPS. In FP16, the AMD card also leads with 40.55 TFLOPS (2:1 rate), while the NVIDIA card achieves 10.32 TFLOPS (1:1). These theoretical peak figures are not reflected in the OpenCL benchmark, which favors the NVIDIA card, but they indicate the AMD card’s raw shading capability is substantially higher on paper. The benchmark results indicate that real-world OpenCL performance favors NVIDIA, but the AMD architecture has untapped potential in other workloads.

The Verdict

The data is unequivocal for compute workloads: the NVIDIA PG506-232 is the superior choice. Its Geekbench OpenCL score of 225,124 is 75.1% higher than the AMD Radeon PRO V620’s 128,580, placing it in the 99th percentile of all GPUs versus the V620’s 96th. For any task that relies on OpenCL compute, the NVIDIA card is the clear winner. Its higher memory bandwidth (933.1 GB/s vs 512.0 GB/s) and tensor cores further reinforce its suitability for AI and data-center compute roles, despite having less memory (24 GB vs 32 GB).

On the other hand, the AMD Radeon PRO V620 should be considered by users who prioritize theoretical shading performance and memory capacity. It offers 32 GB of memory, 46.2% more than the NVIDIA card, and a much higher FP32 throughput of 20.28 TFLOPS, which is 96.5% higher. The inclusion of 72 ray tracing cores and support for DirectX 12 Ultimate, OpenGL 4.6, and Vulkan 1.4 makes it a more flexible option for graphics-centric tasks, even though it lacks the compute dominance of the PG506-232 in OpenCL tests. The V620’s nearest rivals are all within 1% of its score, indicating it is a solid mid-range performer, but it cannot match the PG506-232’s top-tier compute status.

Ultimately, the choice depends on the workload. For pure compute acceleration where OpenCL is the standard, the NVIDIA PG506-232 is unmatched in this comparison. For users needing more memory and higher peak FP32 rates, potentially for rendering or simulation, the AMD Radeon PRO V620 offers a compelling alternative, albeit with a significant performance deficit in the measured benchmark.

Specification Differences

| Specification | NVIDIA PG506-232 | AMD Radeon PRO V620 |

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

| Architecture | Ampere | RDNA 2.0 |

| Chip | GA100 | Navi 21 |

| Generation | Server Ampere (Axx) | Radeon Pro Navi (Navi II Series) |

| Transistors | 54,200 million | 26,800 million |

| Die Size | 826 mm² | 520 mm² |

| Transistor Density | 65.6M / mm² | 51.5M / mm² |

| Base Clock | 930 MHz | 1825 MHz |

| Boost Clock | 1440 MHz | 2200 MHz |

| Memory Clock | 1215 MHz (2.4 Gbps effective) | 2000 MHz (16 Gbps effective) |

| Memory Size | 24 GB | 32 GB |

| Memory Type | HBM2 | GDDR6 |

| Memory Bus Width | 3072 bit | 256 bit |

| Memory Bandwidth | 933.1 GB/s | 512.0 GB/s |

| Shading Units | 3584 | 4608 |

| TMUs | 224 | 288 |

| ROPs | 96 | 128 |

| RT Cores | null | 72 |

| Tensor Cores | 224 | null |

| Pixel Rate | 138.2 GPixel/s | 281.6 GPixel/s |

| Texture Rate | 322.6 GTexel/s | 633.6 GTexel/s |

| FP32 Performance | 10.32 TFLOPS | 20.28 TFLOPS |

| FP16 Performance | 10.32 TFLOPS (1:1) | 40.55 TFLOPS (2:1) |

| TDP | 165 W | 300 W |

| Power Connectors | 8-pin EPS | 2x 8-pin |

| Suggested PSU | 450 W | 700 W |

| Height | 112 mm (4.4 inches) | 120 mm (4.7 inches) |

| Width | null | 50 mm (2 inches) |

| Release Date | 2021-04-11 | 2021-11-03 |

| Predecessor | Tesla Turing | Radeon Pro Vega |

| Successor | Server Ada | null |

| DirectX | null | 12 Ultimate (12_2) |

| OpenGL | null | 4.6 |

| Vulkan | null | 1.4 |

| Avg Benchmark Score | 225,124 | 136,472 |

| Percentile | 99 | 96 |