AMD Radeon Pro Vega 20 vs NVIDIA GeForce RTX 3090 Comparison

The AMD Radeon Pro Vega 20 and the NVIDIA GeForce RTX 3090 occupy opposite ends of the GPU spectrum, yet the aggregate benchmark data places them in a surprisingly close overall percentile ranking. Both cards share a 73rd percentile vs. all GPUs, and their average benchmark scores are separated by only 1% (27,839 vs. 27,565). However, this statistical near-tie is misleading, as the head-to-head benchmark results reveal a stark performance gap that is masked by the very different workloads each card was designed to handle.

Head-to-Head Benchmarks

The data provides two direct head-to-head comparisons, and in both, the NVIDIA GeForce RTX 3090 dominates decisively. In the Geekbench OpenCL test, the RTX 3090 scores 172,758 against the Radeon Pro Vega 20’s 26,679, a delta of -84.6% for the AMD card. This is not a marginal difference; the RTX 3090 delivers roughly 6.5 times the raw compute throughput in this OpenCL workload. The Geekbench Vulkan result tells a similar story, with the RTX 3090 scoring 53,927 versus 26,410 for the Radeon Pro Vega 20, representing a -51% delta. In Vulkan, the NVIDIA card is more than twice as fast, indicating a substantial advantage in graphics API performance that scales with the card’s much larger hardware budget.

These two wins give the RTX 3090 a clean 2-0 sweep in the head-to-head section. Notably, the OpenCL margin is far larger than the Vulkan margin, suggesting that the RTX 3090’s advantage is most pronounced in compute-heavy, general-purpose GPU tasks. The Radeon Pro Vega 20’s nearest rival data confirms its position: it sits within 0.6% of the NVIDIA GeForce GTX 980 Ti and 0.5% of the AMD Radeon Pro W5500X, placing it in a performance tier that is contemporaneous with 2015-era enthusiast silicon. Conversely, the RTX 3090’s nearest rivals include the NVIDIA GeForce RTX 4070 Mobile and AMD Radeon RX 6700 XT, both within 0.5%, which shows its aggregate score is competitive with newer mid-range parts, even if its raw compute is far ahead of the Vega 20.

Architecture Differences

The two cards are built on fundamentally different architectural philosophies separated by two process generations. The Radeon Pro Vega 20 uses the Vega 12 chip on the GCN 5.0 architecture, fabricated on a 14 nm process at GlobalFoundries. The RTX 3090 uses the GA102 chip on the Ampere architecture, built on an 8 nm process at Samsung. This process shrink allows the RTX 3090 to pack 28,300 million transistors onto a 628 mm² die, yielding a transistor density of 45.1M per mm²; the Vega 20’s transistor count and die size are not listed in the data, but its 14 nm node and much lower shading unit count indicate a far smaller and simpler chip.

The compute resources diverge wildly. The Radeon Pro Vega 20 has 1,280 shading units, 80 TMUs, and 32 ROPs, while the RTX 3090 has 10,496 shading units, 328 TMUs, and 112 ROPs. The NVIDIA card also includes 82 ray tracing cores and 328 tensor cores, features that are entirely absent from the AMD part. Memory configurations are equally disparate: the Vega 20 uses 4 GB of HBM2 on a 1024-bit bus, delivering 189.4 GB/s of bandwidth, whereas the RTX 3090 has 24 GB of GDDR6X on a 384-bit bus, achieving 936.2 GB/s. The RTX 3090’s memory bandwidth is nearly 5 times higher, which is critical for its large frame buffer and compute workloads.

Clock speeds also differ, with the Vega 20 running a base of 815 MHz and a boost of 1,283 MHz, while the RTX 3090 boosts to 1,695 MHz from a 1,395 MHz base. The resulting pixel rate is 41.06 GPixel/s for the Vega 20 versus 189.8 GPixel/s for the RTX 3090, and texture rates are 102.6 GTexel/s versus 556.0 GTexel/s. FP32 throughput is 3.284 TFLOPS for the AMD card and 35.58 TFLOPS for the NVIDIA card, a 10.8x difference. The RTX 3090 also offers FP16 at a 1:1 ratio (35.58 TFLOPS), while the Vega 20’s FP16 is 6.569 TFLOPS via a 2:1 ratio.

Where Each One Wins

Based on the benchmark data, the RTX 3090 wins in every measured workload where the two are directly compared. Its OpenCL and Vulkan victories are emphatic, and its higher FP32, texture rate, and pixel rate all point to superior performance in 3D rendering, video editing, and GPU-accelerated compute tasks. The 24 GB memory capacity and 936.2 GB/s bandwidth make it suited for large datasets, high-resolution textures, and machine learning training, where memory capacity is a hard constraint. The presence of ray tracing and tensor cores further extends its utility to real-time ray-traced graphics and AI inference.

The Radeon Pro Vega 20, despite losing both head-to-head tests, still has a defined use case. Its 100 W TDP and IGP slot width indicate it is designed for integrated use in portable devices, specifically Apple Mac systems, as its generation is listed as "Radeon Pro Mac (Vega Series)." Its performance, while lower, is sufficient for tasks that do not require the RTX 3090’s massive compute resources. The Vega 20’s 4 GB of HBM2 memory is small but fast for its era, and its 73rd percentile ranking shows it outperforms a significant majority of all GPUs. It wins in efficiency per watt, though the data does not list a direct efficiency metric; its 100 W TDP is 3.5 times lower than the RTX 3090’s 350 W, and it requires no power connectors or suggested PSU, unlike the RTX 3090’s 1x 12-pin connector and 750 W suggested PSU.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The AMD Radeon Pro Vega 20 has a slightly higher average benchmark score of 27,839, compared to the NVIDIA GeForce RTX 3090’s 27,565, a difference of 1%.

Q: How much faster is the RTX 3090 in OpenCL compute?

A: The RTX 3090 scores 172,758 in Geekbench OpenCL, which is 84.6% higher than the Radeon Pro Vega 20’s 26,679, representing a roughly 6.5x performance advantage.

Q: Does the RTX 3090 support ray tracing?

A: Yes, the RTX 3090 includes 82 ray tracing cores, a feature not present on the Radeon Pro Vega 20.

Q: What are the memory capacities of each card?

A: The Radeon Pro Vega 20 has 4 GB of HBM2 memory, while the RTX 3090 has 24 GB of GDDR6X memory.

Q: What is the process node difference between the two?

A: The Radeon Pro Vega 20 is built on a 14 nm process at GlobalFoundries, while the RTX 3090 uses an 8 nm process at Samsung.

Q: Which card has a higher boost clock?

A: The RTX 3090 has a boost clock of 1,695 MHz, which is higher than the Radeon Pro Vega 20’s boost clock of 1,283 MHz.

The Verdict

The data is unambiguous for raw performance: the NVIDIA GeForce RTX 3090 is the superior card in every measured head-to-head benchmark, with a 84.6% lead in OpenCL and a 51% lead in Vulkan. Its 10.8x higher FP32 throughput, 5x higher memory bandwidth, and 6x more memory capacity make it the only choice for demanding compute, 4K gaming, or professional 3D workloads. The RTX 3090’s 73rd percentile ranking understates its capability because its aggregate score is dragged down by a mix of diverse benchmarks, but in the tests where it is directly compared, it is overwhelmingly faster.

The AMD Radeon Pro Vega 20 is not without merit, but its role is narrow. As an IGP with a 100 W TDP, it is designed for mobile Mac systems where power and space are constrained. Its 73rd percentile ranking shows it is a capable part, but its 3.284 TFLOPS FP32 and 4 GB memory limit it to lighter tasks. It wins only in power efficiency, using 71% less power than the RTX 3090, and in its integrated form factor. For a user who requires the RTX 3090’s features—ray tracing, tensor cores, 24 GB VRAM—the Vega 20 is simply not an alternative. Conversely, for a portable system that cannot accommodate a 336 mm, triple-slot card requiring a 750 W PSU, the Vega 20 is the only viable option in this comparison.

Specification Differences

| Field | AMD Radeon Pro Vega 20 | NVIDIA GeForce RTX 3090 |

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

| Architecture | GCN 5.0 | Ampere |

| Process Node | 14 nm | 8 nm |

| Foundry | GlobalFoundries | Samsung |

| Transistors | Not listed | 28,300 million |

| Die Size | Not listed | 628 mm² |

| Base Clock | 815 MHz | 1,395 MHz |

| Boost Clock | 1,283 MHz | 1,695 MHz |

| Memory Size | 4 GB | 24 GB |

| Memory Type | HBM2 | GDDR6X |

| Memory Bus Width | 1024 bit | 384 bit |

| Memory Bandwidth | 189.4 GB/s | 936.2 GB/s |

| Shading Units | 1,280 | 10,496 |

| TMUs | 80 | 328 |

| ROPs | 32 | 112 |

| RT Cores | Not listed | 82 |

| Tensor Cores | Not listed | 328 |

| Pixel Rate | 41.06 GPixel/s | 189.8 GPixel/s |

| Texture Rate | 102.6 GTexel/s | 556.0 GTexel/s |

| FP32 | 3.284 TFLOPS | 35.58 TFLOPS |

| FP16 | 6.569 TFLOPS (2:1) | 35.58 TFLOPS (1:1) |

| TDP | 100 W | 350 W |

| Slot Width | IGP | Triple-slot |

| Power Connectors | Not listed | 1x 12-pin |

| Suggested PSU | Not listed | 750 W |

| Bus Interface | PCIe 3.0 x16 | PCIe 4.0 x16 |

| Display Outputs | Portable Device Dependent | 1x HDMI 2.13x DisplayPort 1.4a |

| DirectX | 12 (12_1) | 12 Ultimate (12_2) |

| Vulkan | 1.3 | 1.4 |

| Release Date | 2018-11-13 | 2020-08-31 |

| Launch MSRP | Not listed | 1,499 USD |