AMD Radeon PRO V620 vs AMD Radeon Pro W6900X Comparison

FAQ

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

A: The AMD Radeon Pro W6900X leads in both available benchmark tests. In Geekbench OpenCL it scores 130,035 versus 128,580 for the Radeon PRO V620, a delta of 1.1%. In Geekbench Vulkan, the W6900X scores 148,865 against 144,364, a 3.1% advantage.

Q: What is the core configuration difference between the two GPUs?

A: The W6900X has 5,120 shading units, 320 texture mapping units, and 80 ray tracing cores. The PRO V620 has 4,608 shading units, 288 TMUs, and 72 RT cores. Both share 128 ROPs.

Q: Are the memory subsystems identical?

A: Yes. Both cards feature 32 GB of GDDR6 on a 256-bit bus, with 512.0 GB/s bandwidth and 2000 MHz memory clock (16 Gbps effective).

Q: Which card has a higher boost clock?

A: The PRO V620 boosts to 2200 MHz, while the W6900X boosts to 2171 MHz. Interestingly, the lower-clock PRO V620 still trails in both recorded benchmarks.

Q: What is the form factor difference?

A: The W6900X uses an Apple MPX bus interface with display outputs (1x HDMI 2.1, 4x Thunderbolt). The PRO V620 is a dual-slot PCIe 4.0 x16 card with no display outputs and requires 2x 8-pin power connectors.

Q: How do these cards rank among all GPUs?

A: The W6900X sits at the 97th percentile with an average benchmark score of 168,574. The PRO V620 is at the 96th percentile with 136,472 average. The W6900X's nearest rival is the NVIDIA RTX 4500 Ada Generation at 166,094 (1.5% behind), while the PRO V620's closest competitor is the AMD Radeon Pro W6800X Duo at 135,774 (0.5% behind).

Architecture Differences

Both GPUs are built on the same Navi 21 chip, using RDNA 2.0 architecture on TSMC's 7 nm process. The die is identical: 26,800 million transistors on a 520 mm² package, yielding a transistor density of 51.5M per mm². The architectural lineage differs only in generation naming — the W6900X belongs to "Radeon Pro Mac (Navi II Series)" while the PRO V620 is "Radeon Pro Navi (Navi II Series)".

The core counts diverge meaningfully. The W6900X enables 5,120 shading units, 320 TMUs, and 80 RT cores. The PRO V620 ships with 4,608 shading units, 288 TMUs, and 72 RT cores — a reduction of 512 shaders, 32 TMUs, and 8 RT cores. Both retain 128 ROPs, so pixel throughput differences are minimal. The pixel rate for the W6900X is 277.9 GPixel/s versus 281.6 GPixel/s for the PRO V620, reflecting the latter's higher boost clock.

Texture throughput tells a different story. The W6900X achieves 694.7 GTexel/s compared to 633.6 GTexel/s for the PRO V620. This 61.1 GTexel/s gap stems from the W6900X's additional TMUs outweighing the PRO V620's clock advantage. FP32 compute follows the same pattern: 22.23 TFLOPS versus 20.28 TFLOPS, a difference of 1.95 TFLOPS. FP16 rates are 44.46 TFLOPS and 40.55 TFLOPS respectively, both at 2:1 ratio.

Memory architecture is identical — same capacity, type, bus width, and bandwidth. This means memory-bound workloads should perform similarly, pushing performance differences toward compute-bound scenarios. The interface and physical design diverge sharply: the W6900X uses Apple MPX with integrated display outputs, while the PRO V620 is a standard PCIe 4.0 x16 dual-slot card with no outputs and dual 8-pin power. The PRO V620 adds 50 mm width (2 inches) versus the W6900X's unspecified width. Both share 267 mm length and 120 mm height, and both carry a 300 W TDP with 700 W suggested PSU.

The Verdict

The data presents a clear but modest hierarchy. The W6900X wins both head-to-head benchmarks, taking 2 wins to 0. In OpenCL, the margin is 1.1% — nearly negligible for real-world workloads. In Vulkan, the gap widens to 3.1%, still not dramatic but consistent. The W6900X's additional 512 shading units and 8 RT cores translate into a measurable, if not overwhelming, performance edge.

Who should pick the W6900X? The data suggests Mac ecosystem users. Its Apple MPX interface and Thunderbolt outputs make it purpose-built for Mac Pro systems. The 97th percentile ranking and 168,574 average score place it among top-tier workstation GPUs, slightly ahead of the NVIDIA RTX 4500 Ada Generation (166,094, 1.5% behind). Users needing display outputs directly from the card have no choice but the W6900X.

Who should pick the PRO V620? The benchmark data alone doesn't favor it — it loses both tests. But its form factor tells a different story. The dual-slot PCIe 4.0 x16 layout with 2x 8-pin power and no display outputs suggests a server or cloud compute role. The 96th percentile ranking (136,472 average) still places it in elite company, nearly matching the AMD Radeon Pro W6800X Duo (135,774, 0.5% behind). For compute farms prioritizing standard PCIe compatibility over Mac integration, the PRO V620's lower average score may be an acceptable trade for deployment flexibility. Its predecessor was the Radeon Pro Vega, indicating a lineage aimed at professional compute rather than desktop display.

The verdict hinges on ecosystem, not raw performance. Both are end-of-life products. The W6900X is the faster card by every measured metric. The PRO V620 is the more conventional card for standard PC systems. Neither offers a compelling performance advantage over the other — the differences are within 3.1% across both benchmarks.

Specification Differences

| Field | AMD Radeon Pro W6900X | AMD Radeon PRO V620 |

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

| Generation | Radeon Pro Mac (Navi II Series) | Radeon Pro Navi (Navi II Series) |

| Boost Clock | 2171 MHz | 2200 MHz |

| Shading Units | 5120 | 4608 |

| TMUs | 320 | 288 |

| RT Cores | 80 | 72 |

| Pixel Rate | 277.9 GPixel/s | 281.6 GPixel/s |

| Texture Rate | 694.7 GTexel/s | 633.6 GTexel/s |

| FP32 | 22.23 TFLOPS | 20.28 TFLOPS |

| FP16 | 44.46 TFLOPS (2:1) | 40.55 TFLOPS (2:1) |

| Slot Width | Not specified | Dual-slot |

| Power Connectors | Not specified | 2x 8-pin |

| Bus Interface | Apple MPX | PCIe 4.0 x16 |

| Display Outputs | 1x HDMI 2.1, 4x Thunderbolt | No outputs |

| Width | Not specified | 50 mm (2 inches) |

| Release Date | 2021-08-02 | 2021-11-03 |

| Predecessor | None | Radeon Pro Vega |

| Launch MSRP | 5,999 USD | None |

| Geekbench Metal | 226,821 | Not tested |

| Geekbench OpenCL | 130,035 | 128,580 |

| Geekbench Vulkan | 148,865 | 144,364 |

| Percentile | 97 | 96 |

| Avg Benchmark Score | 168,574 | 136,472 |

Identical specifications include: 7 nm process, TSMC foundry, 26,800 million transistors, 520 mm² die size, 51.5M/mm² density, 1825 MHz base clock, 2000 MHz memory (16 Gbps effective), 32 GB GDDR6, 256-bit bus, 512.0 GB/s bandwidth, 128 ROPs, 300 W TDP, 700 W suggested PSU, DirectX 12 Ultimate (12_2), OpenGL 4.6, Vulkan 1.4, 267 mm length, 120 mm height, and end-of-life production status.

Head-to-Head Benchmarks

The head-to-head results are limited to two tests, both favoring the W6900X. In Geekbench OpenCL, the W6900X scores 130,035 against 128,580 for the PRO V620 — a 1.1% delta. This is a narrow margin that could fall within run-to-run variance. In Geekbench Vulkan, the W6900X extends its lead to 3.1%, scoring 148,865 versus 144,364.

The Vulkan gap is more instructive. Vulkan typically exposes lower-level hardware capabilities, so the W6900X's extra 512 shading units and 8 RT cores likely contribute to the larger margin. The PRO V620's higher boost clock (2200 MHz vs 2171 MHz) is not enough to overcome the core deficit. The W6900X also posted a Geekbench Metal score of 226,821 — a test the PRO V620 did not take, likely reflecting the W6900X's Mac-oriented design and Apple MPX interface.

Contextualizing these scores against rivals adds perspective. The W6900X's average of 168,574 is 1.5% ahead of the NVIDIA RTX 4500 Ada Generation (166,094), 2% ahead of the NVIDIA RTX A5500 (165,217), 2.2% ahead of the AMD Radeon PRO W7800 (164,894), and 3.7% ahead of the NVIDIA A100 PCIe 40 GB (162,504). The PRO V620's average of 136,472 is 0.5% ahead of the AMD Radeon Pro W6800X Duo (135,774), 0.8% ahead of the AMD Radeon PRO W6800 (135,396), 0.9% ahead of the NVIDIA A10M (135,230), and 0.9% ahead of the NVIDIA RTX 4000 Ada Generation (135,218).

The striking observation is the gap between the two cards' averages. The W6900X's 168,574 is roughly 23.5% higher than the PRO V620's 136,472 — far larger than the 1.1-3.1% deltas in the shared tests. This discrepancy suggests the W6900X benefits from its Metal benchmark score, which the PRO V620 lacks. In shared workloads, these cards are close. In Metal-specific tasks, the W6900X pulls decisively ahead.

Where Each One Wins

The W6900X wins in every head-to-head benchmark, but the margins tell a nuanced story. Its 1.1% OpenCL advantage is minor; workloads using OpenCL should see nearly identical performance from either card. The 3.1% Vulkan advantage is more substantial, favoring the W6900X for Vulkan-based rendering or compute pipelines. The W6900X's Metal benchmark (226,821) represents its strongest domain, though no direct comparison exists. For Mac-centric workflows leveraging Metal, the W6900X is the clear choice.

The PRO V620 wins in deployment flexibility. Its PCIe 4.0 x16 interface is universal across modern PC servers and workstations, while the Apple MPX bus restricts the W6900X to Apple hardware. The PRO V620's dual-slot form factor and 2x 8-pin power connectors are standard, simplifying integration into existing systems. Its lack of display outputs, while seemingly a drawback, is actually an advantage for headless compute nodes where display capabilities are unused overhead.

In terms of raw specs, the PRO V620's higher boost clock (2200 MHz) gives it a theoretical peak-rate advantage in pixel throughput (281.6 GPixel/s vs 277.9 GPixel/s). This could marginally favor the PRO V620 in fill-rate-bound scenarios, though the 1.3% advantage is unlikely to be perceptible. The W6900X counters with superior texture rate (694.7 GTexel/s vs 633.6 GTexel/s), a 9.6% advantage that benefits textured workloads.

The W6900X's launch MSRP of 5,999 USD positions it as a premium Mac workstation card. The PRO V620 lacks a launch MSRP, suggesting a different sales channel — likely OEM or cloud provider procurement rather than retail. This aligns with its server-oriented design. For users choosing between the two, the decision is less about performance — which favors the W6900X by narrow margins — and more about platform compatibility. Mac systems require the W6900X. Standard PC or server systems should use the PRO V620. The performance gap in shared benchmarks is too small to override platform constraints.