GPU Comparison

The AMD Radeon Pro 570 and NVIDIA P104-100 represent two distinct approaches to GPU design from the same era, yet they deliver nearly identical average performance. The AMD part, built on GCN 4.0, is a mobile-oriented solution with a 150 W TDP and integrated form factor, while the NVIDIA part, based on Pascal, is a mining-focused card with no display outputs and a dual-slot cooler. Benchmark data shows the Radeon Pro 570 holds a razor-thin 0.7% lead in average score (33207 vs 32982), placing both cards within 1.1% of each other and the NVIDIA T600 Mobile. Despite this statistical tie in overall averages, their architectural and specification differences create a clear performance split in individual workloads, with the P104-100 dominating in compute-heavy tasks.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Pro 570 edges out the NVIDIA P104-100 with an average benchmark score of 33207 compared to 32982, a difference of only 0.7%. This places the AMD card at the 78th percentile among all GPUs, while the NVIDIA card sits at the 77th percentile.

Q: How do the two cards compare in Geekbench OpenCL performance?

A: The NVIDIA P104-100 is significantly faster, scoring 52368 versus the Radeon Pro 570's 27702. This represents a 47.1% advantage for the NVIDIA card, which is the largest performance gap between the two in any shared benchmark.

Q: What are the key architectural differences between the two GPUs?

A: The Radeon Pro 570 uses AMD's GCN 4.0 architecture on a 14 nm GlobalFoundries process, while the P104-100 uses NVIDIA's Pascal architecture on a 16 nm TSMC process. The NVIDIA chip has a larger die (314 mm² vs 232 mm²) and more transistors (7,200 million vs 5,700 million).

Q: Which card has higher memory bandwidth?

A: The NVIDIA P104-100 offers 320.3 GB/s of bandwidth using 4 GB of GDDR5X memory, while the Radeon Pro 570 provides 217.0 GB/s with 4 GB of GDDR5. Both use a 256-bit memory bus.

Q: What is the difference in FP16 performance?

A: The Radeon Pro 570 delivers 3.960 TFLOPS of FP16 performance at a 1:1 ratio with FP32, while the P104-100 offers only 104.0 GFLOPS at a 1:64 ratio. This makes the AMD card dramatically better suited for FP16 workloads.

Q: Do both cards support the same API levels?

A: No, they differ. The Radeon Pro 570 supports DirectX 12 (12_0) and Vulkan 1.3, while the P104-100 supports DirectX 12 (12_1) and Vulkan 1.4. Both support OpenGL 4.6.

Architecture Differences

The two GPUs are built on fundamentally different architectures and process technologies. The AMD Radeon Pro 570 uses the Ellesmere chip with GCN 4.0 architecture, manufactured by GlobalFoundries on a 14 nm process. It contains 5,700 million transistors on a 232 mm² die, yielding a transistor density of 24.6 million per square millimeter. In contrast, the NVIDIA P104-100 uses the GP104 chip with Pascal architecture, fabricated by TSMC on a 16 nm process. This larger chip houses 7,200 million transistors on a 314 mm² die, with a slightly lower transistor density of 22.9 million per square millimeter.

These architectural differences manifest in several key areas. The Radeon Pro 570 features 1792 shading units, 112 texture mapping units, and 32 ROPs. The P104-100 counters with 1920 shading units, 120 TMUs, and 64 ROPs, the latter being double the AMD card's count. Clock speeds also differ substantially: the AMD card runs at a 1000 MHz base and 1105 MHz boost, while the NVIDIA card operates at 1607 MHz base and 1733 MHz boost. This clock advantage, combined with higher shader counts, gives the P104-100 a theoretical FP32 throughput of 6.655 TFLOPS versus 3.960 TFLOPS for the Radeon Pro 570.

The memory subsystems diverge as well. Both cards have 4 GB of memory on a 256-bit bus, but the Radeon Pro 570 uses GDDR5 at 1695 MHz (6.8 Gbps effective), while the P104-100 uses GDDR5X at 1251 MHz (10 Gbps effective). This results in 217.0 GB/s of bandwidth for the AMD card and 320.3 GB/s for the NVIDIA card. The FP16 capabilities are starkly different: the Radeon Pro 570 achieves 3.960 TFLOPS at a 1:1 ratio with FP32, whereas the P104-100 manages only 104.0 GFLOPS at a 1:64 ratio, indicating the NVIDIA card is not designed for half-precision work.

Head-to-Head Benchmarks

The shared benchmark suite between these two cards includes Geekbench OpenCL and Geekbench Vulkan, and the NVIDIA P104-100 wins both decisively. In Geekbench OpenCL, the P104-100 scores 52368 against the Radeon Pro 570's 27702, a 47.1% advantage. This is a substantial margin that underscores the NVIDIA card's compute-oriented design. The Pascal architecture's higher clock speeds, additional shaders, and doubled ROP count contribute directly to this result, as does the superior 320.3 GB/s memory bandwidth.

In Geekbench Vulkan, the gap narrows but remains significant. The P104-100 scores 45165 versus the Radeon Pro 570's 31974, a 29.2% difference. While still a clear win for NVIDIA, the smaller margin suggests that the GCN 4.0 architecture holds up relatively better in graphics-oriented workloads compared to pure compute tasks. Notably, the AMD card supports Vulkan 1.3 while the NVIDIA card supports Vulkan 1.4, which may partially account for the closer result.

Despite these two decisive losses, the Radeon Pro 570 still manages to edge out the P104-100 in average benchmark score (33207 vs 32982). This is possible because the AMD card has additional benchmark results in Geekbench Metal (39945) that are not available for the NVIDIA card, which lacks display outputs and Apple platform compatibility. The average score calculation includes these Metal results, allowing the Radeon Pro 570 to offset its OpenCL and Vulkan deficits. The head-to-head record stands at 2 wins for NVIDIA and 0 for AMD, but the overall average tells a more nuanced story where the AMD card's broader workload coverage gives it a slight edge.

Specification Differences

| Specification | AMD Radeon Pro 570 | NVIDIA P104-100 |

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

| Architecture | GCN 4.0 | Pascal |

| Process Node | 14 nm | 16 nm |

| Foundry | GlobalFoundries | TSMC |

| Transistors | 5,700 million | 7,200 million |

| Die Size | 232 mm² | 314 mm² |

| Transistor Density | 24.6M / mm² | 22.9M / mm² |

| Base Clock | 1000 MHz | 1607 MHz |

| Boost Clock | 1105 MHz | 1733 MHz |

| Memory Type | GDDR5 | GDDR5X |

| Memory Clock | 1695 MHz (6.8 Gbps) | 1251 MHz (10 Gbps) |

| Bandwidth | 217.0 GB/s | 320.3 GB/s |

| Shading Units | 1792 | 1920 |

| TMUs | 112 | 120 |

| ROPs | 32 | 64 |

| Pixel Rate | 35.36 GPixel/s | 110.9 GPixel/s |

| Texture Rate | 123.8 GTexel/s | 208.0 GTexel/s |

| FP32 | 3.960 TFLOPS | 6.655 TFLOPS |

| FP16 | 3.960 TFLOPS (1:1) | 104.0 GFLOPS (1:64) |

| TDP | 150 W | Not specified |

| Slot Width | IGP | Dual-slot |

| Power Connectors | None | 1x 8-pin |

| Suggested PSU | Not specified | 200 W |

| Bus Interface | PCIe 3.0 x16 | PCIe 1.0 x4 |

| Display Outputs | Portable Device Dependent | No outputs |

| DirectX Support | 12 (12_0) | 12 (12_1) |

| Vulkan Support | 1.3 | 1.4 |

| Length | Not specified | 267 mm (10.5 inches) |

| Release Date | 2017-06-04 | 2017-12-11 |

The Verdict

The data presents a clear split between raw compute capability and practical usability. For users prioritizing maximum performance in OpenCL or Vulkan workloads, the NVIDIA P104-100 is the superior choice. It delivers 47.1% higher OpenCL scores and 29.2% higher Vulkan scores, backed by 6.655 TFLOPS of FP32 performance, 320.3 GB/s of memory bandwidth, and double the ROP count. The Pascal architecture's higher clock speeds and additional shading units make it a formidable compute engine, and its 110.9 GPixel/s pixel rate is more than three times that of the Radeon Pro 570.

However, the P104-100's mining-oriented design imposes significant limitations. It has no display outputs, meaning it cannot drive a monitor without a secondary GPU. Its PCIe 1.0 x4 bus interface is severely bandwidth-constrained compared to the Radeon Pro 570's PCIe 3.0 x16, which could impact data transfer in some workloads. The card also requires a dual-slot cooler and an 8-pin power connector, with a suggested 200 W PSU, making it a less flexible option for general-purpose systems. Its FP16 performance is negligible at 104.0 GFLOPS, rendering it unsuitable for workloads that rely on half-precision arithmetic.

The AMD Radeon Pro 570, by contrast, offers a more balanced profile. Despite losing both shared benchmarks, its average score of 33207 is 0.7% higher than the NVIDIA card's 32982, thanks to its strong Geekbench Metal result of 39945. The card operates at 150 W with no external power connectors and an IGP slot width, making it far easier to integrate into portable or space-constrained systems. Its FP16 performance of 3.960 TFLOPS matches its FP32 output, a feature completely absent on the P104-100. The 14 nm process and higher transistor density (24.6M / mm² vs 22.9M / mm²) suggest a more efficient design per unit area.

The verdict depends entirely on use case. The P104-100 is the pick for compute-focused tasks where raw throughput matters and display output is irrelevant, provided the system can accommodate its dual-slot size, 8-pin power requirement, and PCIe 1.0 x4 interface. The Radeon Pro 570 is the better all-round solution for systems requiring integrated graphics, display capability, and FP16 support, with a slight overall performance edge in average benchmarks. Neither card is currently in production, both being end-of-life products, so availability and platform compatibility will ultimately dictate the practical choice.