AMD Radeon Pro 555X vs NVIDIA GeForce GTX 670 Comparison

The AMD Radeon Pro 555X and NVIDIA GeForce GTX 670 represent two very different eras of GPU design, with the former built for mobile professional work on a 14 nm process and the latter a desktop gaming card from the 28 nm generation. Benchmark results show a clear split: the AMD card dominates in Metal workloads, while the NVIDIA card takes decisive leads in OpenCL and Vulkan. The average scores are close, with the Radeon Pro 555X posting 13,321 against the GTX 670's 12,773, a difference of roughly 4.3%, but the individual test results reveal distinct performance profiles rather than a single overall winner.

Head-to-Head Benchmarks

The most dramatic margin in this comparison belongs to the AMD Radeon Pro 555X in the Geekbench Metal test. It scores 14,479 against the GTX 670's 7,424, a 95% advantage. This is not a marginal win; it is nearly double the performance, and it reflects the AMD card's modern GCN 4.0 architecture being far better suited to Apple's Metal API. The GTX 670, a Kepler part from 2012, was never designed with Metal in mind, and its score of 7,424 puts it at roughly half the capability of the 555X in this specific workload. For any application that relies on Metal, the choice is unambiguous.

The tables turn completely in the OpenCL test. Here, the NVIDIA GeForce GTX 670 scores 15,341, while the Radeon Pro 555X trails at 12,628, giving NVIDIA a 17.7% lead. The GTX 670's raw hardware resources—1,344 shading units, 112 TMUs, and a 256-bit memory bus—come into play in this more generic compute workload, overwhelming the 555X's 768 shading units and 128-bit bus. The 17.7% delta is substantial, and it shows that the older card still has significant compute power when the software stack plays to its strengths.

Vulkan results follow a similar pattern. The GTX 670 wins again, scoring 15,553 to the 555X's 12,855, a 17.3% margin. This is nearly identical to the OpenCL gap, suggesting that the NVIDIA card's advantage in these cross-platform APIs is consistent and architectural rather than test-specific. The Radeon Pro 555X does support Vulkan 1.3, a newer specification version than the GTX 670's 1.2.175, but the actual performance numbers favor the NVIDIA card regardless of API version. In the two tests that matter for general-purpose compute and modern game engines, the GTX 670 is the clear winner.

Overall, the head-to-head record stands at 1 win for the AMD Radeon Pro 555X and 2 wins for the NVIDIA GeForce GTX 670. However, the magnitude of the AMD win in Metal is far larger than either of NVIDIA's wins. The 95% Metal delta dwarfs the 17.7% and 17.3% deltas in OpenCL and Vulkan, meaning that the total performance picture is more nuanced than a simple win count. The average benchmark scores reflect this: the 555X's 13,321 average is 4.3% higher than the GTX 670's 12,773, driven entirely by that massive Metal score.

Where Each One Wins

The AMD Radeon Pro 555X is the clear choice for any workload that runs on Apple's Metal API. Its 14,479 Metal score is nearly double the GTX 670's, and it positions the card at the 54th percentile of all GPUs, slightly above the GTX 670's 52nd percentile. This makes the 555X a better fit for macOS ecosystems, where Metal is the primary graphics and compute interface. It also consumes just 75 W of power with no power connectors required, which is a massive efficiency advantage over the GTX 670's 170 W TDP and dual 6-pin connectors, though this is a qualitative observation rather than a benchmark-driven one.

The NVIDIA GeForce GTX 670 wins in OpenCL and Vulkan, making it the better option for cross-platform compute tasks and game engines that use these APIs. Its OpenCL score of 15,341 and Vulkan score of 15,553 are both consistently higher than the 555X's corresponding scores of 12,628 and 12,855. This consistency suggests that the GTX 670's Kepler architecture, despite its age, has robust compute throughput that scales well across different abstraction layers. The card also has a wider 256-bit memory bus and 192.3 GB/s of bandwidth, compared to the 555X's 128-bit bus and 94.08 GB/s, which contributes to its performance in bandwidth-sensitive workloads.

For users who need a mix of both, the data presents a trade-off. The 555X offers a massive single-API advantage but loses the other two tests by a combined margin of 35%. The GTX 670 offers balanced wins in two APIs but suffers a catastrophic loss in Metal. Neither card is a general-purpose all-rounder; each has a clear domain where it excels, and the choice depends entirely on which API the user's software targets. The 555X is also an integrated GPU (IGP) in the slot width sense, meaning it is built into a portable device, while the GTX 670 is a dual-slot desktop card with a 241 mm length, so form factor is another consideration that the benchmarks do not capture.

The Verdict

The data supports a clear recommendation for specific use cases rather than a blanket winner. Choose the AMD Radeon Pro 555X if your primary environment is macOS and your applications leverage Metal. The 95% lead in the Metal benchmark is overwhelming, and the card's 14,479 score is a strong result for a mobile integrated GPU. This is particularly relevant for creative professionals running Metal-accelerated software on Mac laptops, where the 555X's 14 nm process and 3,000 million transistors provide modern features like 1.3 TFLOPS of FP16 performance, which the GTX 670 lacks entirely.

Choose the NVIDIA GeForce GTX 670 if your work involves OpenCL or Vulkan compute. Its 17.7% lead in OpenCL and 17.3% lead in Vulkan are significant, and its 2.634 TFLOPS of FP32 performance is nearly double the 555X's 1,393.2 GFLOPS. The GTX 670 also offers 4 GB less memory than the 555X (2 GB vs 4 GB), but its 256-bit bus and 192.3 GB/s bandwidth partially compensate for the smaller capacity in bandwidth-limited workloads. For users running Linux or Windows with Vulkan-based game engines, the GTX 670 is the better performer, despite its 28 nm process node and higher power draw.

For a general-purpose user who cannot guarantee which API their software will use, the average benchmark scores offer a tiebreaker. The 555X's 13,321 average is 4.3% higher than the GTX 670's 12,773, so on a pure average basis, the AMD card is slightly ahead. However, this average is heavily skewed by the Metal result, and the GTX 670's more consistent performance across two of three tests may be more valuable in heterogeneous workloads. Ultimately, the GTX 670 is the safer choice for compute-heavy tasks, while the 555X is the only choice for Metal-centric ecosystems.

FAQ

Q: Which GPU has the higher average benchmark score?

A: The AMD Radeon Pro 555X has an average benchmark score of 13,321, which is 4.3% higher than the NVIDIA GeForce GTX 670's average of 12,773.

Q: How large is the AMD Radeon Pro 555X's lead in the Metal benchmark?

A: The 555X scores 14,479 in the Geekbench Metal test, while the GTX 670 scores 7,424, giving the AMD card a 95% advantage.

Q: In which benchmarks does the NVIDIA GeForce GTX 670 win?

A: The GTX 670 wins in Geekbench OpenCL with a score of 15,341 (17.7% ahead) and in Geekbench Vulkan with a score of 15,553 (17.3% ahead).

Q: What are the memory specifications of each card?

A: The AMD Radeon Pro 555X has 4 GB of GDDR5 memory on a 128-bit bus with 94.08 GB/s bandwidth, while the NVIDIA GeForce GTX 670 has 2 GB of GDDR5 memory on a 256-bit bus with 192.3 GB/s bandwidth.

Q: Which card has a higher transistor count and larger die size?

A: The NVIDIA GeForce GTX 670 has 3,540 million transistors on a 294 mm² die, compared to the AMD Radeon Pro 555X's 3,000 million transistors on a 123 mm² die.

Q: What is the FP32 performance difference between the two?

A: The GTX 670 delivers 2.634 TFLOPS of FP32 compute, while the 555X delivers 1,393.2 GFLOPS, making the NVIDIA card roughly 89% higher in raw single-precision throughput.

Architecture Differences

The AMD Radeon Pro 555X is built on the Polaris 21 chip using the GCN 4.0 architecture, manufactured on a 14 nm process at GlobalFoundries. This is a modern, dense design with 3,000 million transistors packed into a 123 mm² die, yielding a transistor density of 24.4 million transistors per mm². The card features 768 shading units, 48 texture mapping units, and 16 ROPs, with a pixel rate of 14.51 GPixel/s and a texture rate of 43.54 GTexel/s. It supports FP16 compute at a 1:1 ratio with FP32, offering 1,393.2 GFLOPS for both precision levels, which is a hallmark of GCN 4.0's flexible compute capabilities. The memory runs at 1,470 MHz (5.9 Gbps effective) on a 128-bit bus, providing 94.08 GB/s of bandwidth.

The NVIDIA GeForce GTX 670 uses the GK104 chip with the Kepler architecture, manufactured on a 28 nm process at TSMC. This is an older, less dense design with 3,540 million transistors on a 294 mm² die, giving a transistor density of just 12.0 million transistors per mm²—less than half of the 555X's density. The GTX 670 has significantly more execution resources: 1,344 shading units, 112 TMUs, and 32 ROPs, with a pixel rate of 27.44 GPixel/s and a texture rate of 109.8 GTexel/s. Its FP32 throughput is 2.634 TFLOPS, but it has no FP16 support, which limits its usefulness in modern machine learning workloads. The memory operates at 1,502 MHz (6 Gbps effective) on a 256-bit bus, providing 192.3 GB/s of bandwidth, which is more than double the 555X's bandwidth.

Key architectural differences extend beyond raw specs. The 555X supports DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.3, while the GTX 670 supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The newer Vulkan version on the AMD card is notable, though the benchmarks show the older Vulkan implementation on NVIDIA still performs better. The 555X uses a PCIe 3.0 x8 interface and has no power connectors, consuming just 75 W, while the GTX 670 uses PCIe 3.0 x16 and requires two 6-pin power connectors with a 170 W TDP and a 450 W recommended PSU. The GTX 670 also has a full suite of display outputs (2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2), while the 555X's outputs are portable device dependent, reflecting its intended integration into laptops. The GTX 670's release date of May 2012 and launch MSRP of 399 USD place it firmly in the desktop gaming era, while the 555X's July 2018 release targets professional Mac users, and both are now end-of-life products.