AMD Radeon Pro 560X vs NVIDIA GeForce GTX 670 Comparison

Head-to-Head Benchmarks

The recorded data shows a clear split between the AMD Radeon Pro 560X and the NVIDIA GeForce GTX 670, with each card claiming victories in different API workloads. The AMD Radeon Pro 560X wins 2 of the 3 head-to-head benchmarks, but the NVIDIA GeForce GTX 670 secures the single largest margin of victory.

The most dramatic result appears in the Geekbench Metal test. The AMD Radeon Pro 560X scores 18,763, while the NVIDIA GeForce GTX 670 manages only 7,424. That is a 152.7% advantage for the AMD part, a massive gap that reflects the NVIDIA card's lack of optimized Metal support rather than raw compute capability. The GeForce GTX 670 predates Apple's Metal API by several years, and the database shows its score is less than half of the AMD card's output.

In Geekbench Vulkan, the AMD Radeon Pro 560X again takes the lead, scoring 16,819 against the NVIDIA GeForce GTX 670's 15,553. The delta here is 8.1%, a more modest but still decisive margin. This result indicates that in modern cross-platform graphics workloads, the newer AMD architecture holds a consistent edge.

The NVIDIA GeForce GTX 670 fights back in Geekbench OpenCL, where it scores 15,341 versus the AMD Radeon Pro 560X's 9,663. That represents a 37% deficit for the AMD card, flipping the script entirely. OpenCL has historically favored NVIDIA's compute-oriented Kepler design, and the data confirms that the GTX 670 remains substantially faster in this legacy compute API.

Looking at aggregate performance, the AMD Radeon Pro 560X posts an average benchmark score of 15,082 across all tests, while the NVIDIA GeForce GTX 670 averages 12,773. This places the AMD card in the 57th percentile of all GPUs, compared to the 52nd percentile for the NVIDIA card. The overall average favors the AMD part by roughly 18%, driven almost entirely by the Metal result.

The head-to-head deltas tell a nuanced story. The AMD card's OpenCL loss is severe at 37%, but its Metal victory is overwhelming at 152.7%. The Vulkan win of 8.1% is the closest contest. For users prioritizing Metal, the choice is obvious. For OpenCL compute tasks, the older NVIDIA card retains a significant advantage.

Where Each One Wins

The AMD Radeon Pro 560X is the clear winner in Metal-based workloads. Its Geekbench Metal score of 18,763 not only beats the NVIDIA GeForce GTX 670 by 152.7%, but also positions it well above its own average benchmark score of 15,082. This suggests Metal is a strength of the Polaris 21 chip, likely due to its GCN 4.0 architecture and native support for modern Apple frameworks.

The AMD card also wins in Vulkan, scoring 16,819. This is 8.1% ahead of the NVIDIA GeForce GTX 670 and also exceeds its own average score. Vulkan 1.3 support on the AMD card, versus 1.2.175 on the NVIDIA card, aligns with the performance gap. For cross-platform gaming and compute using Vulkan, the AMD Radeon Pro 560X is the safer pick.

The NVIDIA GeForce GTX 670 dominates OpenCL. Its score of 15,341 is 37% higher than the AMD card's 9,663. This is the only test where the NVIDIA card wins, but it wins big. The GTX 670's Kepler architecture, with 1,344 shading units and 112 texture mapping units, clearly excels at OpenCL compute tasks. The AMD card's 1,024 shading units and 64 TMUs cannot keep pace in this API.

In terms of use cases, the AMD Radeon Pro 560X is better suited for macOS environments, creative applications that leverage Metal, and modern Vulkan-based software. The NVIDIA GeForce GTX 670 is better suited for OpenCL compute workloads, particularly those that can utilize its higher texture rate of 109.8 GTexel/s and pixel rate of 27.44 GPixel/s.

The data also shows the AMD card has a higher overall percentile ranking (57th vs 52nd), meaning it outperforms a larger share of the GPU population in the database. The NVIDIA card's nearest rivals include the AMD Radeon Pro 455 with a score of 12,831, while the AMD card's rivals cluster around 15,000 to 15,270, including the NVIDIA GeForce RTX 3050 OEM at 15,199.

The Verdict

Based strictly on the benchmark data, the AMD Radeon Pro 560X is the superior overall card. It wins 2 of 3 head-to-head tests and has a higher average benchmark score of 15,082 versus 12,773. Its percentile ranking of 57th versus 52nd confirms that it sits higher in the overall GPU hierarchy.

The AMD card should be chosen by anyone working in Metal-based environments. The 152.7% lead in Geekbench Metal is decisive and cannot be overcome by any other metric. For macOS users, creative professionals using Metal-accelerated applications, or developers targeting Apple platforms, the Radeon Pro 560X is unequivocally the right pick.

The NVIDIA GeForce GTX 670 should be chosen for OpenCL-heavy compute tasks. Its 37% lead in OpenCL is substantial, and its texture rate of 109.8 GTexel/s and pixel rate of 27.44 GPixel/s indicate strong raw throughput. If the workload is exclusively OpenCL and does not touch Metal or Vulkan, the GTX 670 delivers more compute performance.

However, the overall picture favors AMD. The Vulkan win of 8.1% may appear small, but it represents modern API efficiency that the NVIDIA card lacks. The GTX 670's Vulkan 1.2.175 support is older, and its Metal score of 7,424 suggests poor compatibility with contemporary graphics stacks. The AMD card's 4 GB of GDDR5 memory also doubles the GTX 670's 2 GB, providing more headroom for modern textures and datasets.

The launch MSRP for the NVIDIA GeForce GTX 670 was 399 USD, but the database shows both cards are now end-of-life products. For current use, the AMD Radeon Pro 560X offers better all-around performance, higher memory capacity, lower power draw (75 W versus 170 W), and superior API support. The only scenario where the GTX 670 makes sense is a legacy OpenCL compute pipeline where its specific strengths are required.

FAQ

Q: Which card wins in Metal benchmarks?

A: The AMD Radeon Pro 560X wins decisively, scoring 18,763 against the NVIDIA GeForce GTX 670's 7,424, a 152.7% advantage.

Q: Does the NVIDIA GeForce GTX 670 win any benchmark?

A: Yes, the GTX 670 wins the Geekbench OpenCL test, scoring 15,341 versus the AMD card's 9,663, a 37% margin.

Q: What are the average benchmark scores for each card?

A: The AMD Radeon Pro 560X has an average benchmark score of 15,082, while the NVIDIA GeForce GTX 670 averages 12,773.

Q: How do the cards compare in Vulkan performance?

A: The AMD Radeon Pro 560X wins in Vulkan, scoring 16,819 against the GTX 670's 15,553, a 8.1% difference.

Q: Which card has more memory?

A: The AMD Radeon Pro 560X has 4 GB of GDDR5 memory, double the 2 GB on the NVIDIA GeForce GTX 670.

Q: What is the power consumption difference?

A: The AMD Radeon Pro 560X has a TDP of 75 W, while the NVIDIA GeForce GTX 670 has a TDP of 170 W.

Architecture Differences

The AMD Radeon Pro 560X is built on the Polaris 21 chip using GCN 4.0 architecture, manufactured on a 14 nm process at GlobalFoundries. The die size is 123 mm², containing 3,000 million transistors, which yields a transistor density of 24.4 million per mm². This modern process node allows for high efficiency despite the modest 75 W TDP.

The NVIDIA GeForce GTX 670 uses the GK104 chip with Kepler architecture, manufactured on a 28 nm process at TSMC. The die size is 294 mm², containing 3,540 million transistors, but the transistor density is only 12.0 million per mm². The older, larger process node explains the higher 170 W TDP.

The compute resources differ significantly. The AMD card has 1,024 shading units, 64 TMUs, and 16 ROPs. The NVIDIA card has 1,344 shading units, 112 TMUs, and 32 ROPs. This gives the GTX 670 higher pixel rate (27.44 GPixel/s versus 16.06 GPixel/s) and texture rate (109.8 GTexel/s versus 64.26 GTexel/s). The NVIDIA card also has higher FP32 performance at 2.634 TFLOPS versus 2.056 TFLOPS.

The AMD card supports FP16 at a 1:1 ratio with FP32, both at 2.056 TFLOPS, while the NVIDIA card has no listed FP16 capability. This matters for workloads that can utilize half-precision math. The AMD card also supports newer API versions, including DirectX 12 (12_0) versus the NVIDIA card's DirectX 12 (11_0), and Vulkan 1.3 versus 1.2.175.

The memory subsystems differ in bus width and bandwidth. The AMD card uses a 128-bit bus with 94.08 GB/s bandwidth, while the NVIDIA card uses a 256-bit bus with 192.3 GB/s bandwidth. The GTX 670's wider bus doubles the bandwidth, but the AMD card has twice the memory capacity at 4 GB versus 2 GB.

Specification Differences

The AMD Radeon Pro 560X and NVIDIA GeForce GTX 670 differ in nearly every major specification field. The process node is 14 nm for AMD versus 28 nm for NVIDIA. The AMD card is an IGP slot width, while the NVIDIA card is dual-slot. Power connectors are absent on the AMD card, but the GTX 670 requires 2x 6-pin connectors, with a suggested PSU of 450 W.

The bus interface differs: PCIe 3.0 x8 on the AMD card versus PCIe 3.0 x16 on the NVIDIA card. Display outputs are portable-device dependent for the AMD card, while the GTX 670 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. The NVIDIA card has physical dimensions of 241 mm length, 111 mm height, and 38 mm width, while the AMD card has no listed dimensions.

Clock speeds show the NVIDIA card has a base clock of 915 MHz and boost clock of 980 MHz, while the AMD card has no listed base or boost clocks. Memory clocks are similar in effective terms: 5.9 Gbps effective for AMD and 6 Gbps effective for NVIDIA. The memory type is GDDR5 for both, but capacity differs at 4 GB versus 2 GB.

The release dates are far apart: the AMD card launched on July 15, 2018, while the GTX 670 launched on May 9, 2012. The NVIDIA card has a predecessor (GeForce 500) and successor (GeForce 700), while the AMD card has neither listed. The generation names also differ: Radeon Pro Mac (500X Series) for AMD versus GeForce 600 for NVIDIA.

The NVIDIA card has a launch MSRP of 399 USD, while the AMD card has no MSRP listed. Both are end-of-life products. The foundries differ (GlobalFoundries for AMD, TSMC for NVIDIA), and the architectures are fundamentally different: GCN 4.0 versus Kepler, with different transistor counts (3,000 million versus 3,540 million) and die sizes (123 mm² versus 294 mm²).