AMD Radeon Pro 555 vs NVIDIA GeForce GTX 680 Comparison

# NVIDIA GeForce GTX 680 vs AMD Radeon Pro 555

The data presents a fascinating generational clash: a 2012 desktop flagship from NVIDIA's Kepler architecture against a 2017 mobile-oriented AMD Polaris part, both landing within 5.5% of each other in average benchmark score (14,150 vs 13,407). The GTX 680 claims two of three benchmark victories, yet the Radeon Pro 555's single win is the most dramatic margin in the entire comparison. The percentile rankings tell a tight story too—55th vs 54th percentile among all GPUs—making this a genuine near-tie decided by workload, not raw dominance.

Where Each One Wins

The AMD Radeon Pro 555 owns the Metal API outright. Its Geekbench Metal score of 16,236 dwarfs the GTX 680's 7,897, a 51.4% deficit for NVIDIA. This is not a marginal edge; it is a landslide. For any application that leans on Apple's Metal graphics framework, the Radeon Pro 555 is the clear choice. The data suggests this is the only scenario where the AMD part leads, but it leads so decisively that it redefines the matchup for Metal-centric users.

NVIDIA's GeForce GTX 680, meanwhile, sweeps the other two compute APIs. In Geekbench OpenCL, it scores 16,906 against 11,682 for the Radeon Pro 555—a 44.7% advantage. The Vulkan gap is nearly identical in percentage terms: 17,646 vs 12,303, a 43.4% lead. These are not close contests either. The GTX 680's wins are slightly smaller in relative margin than AMD's Metal victory, but they cover two APIs instead of one. For OpenCL and Vulkan workflows, the 2012 NVIDIA card is decisively faster than the 2017 AMD part.

The takeaway is clear: pick the Radeon Pro 555 for Metal-heavy applications, and the GTX 680 for everything else. The benchmark deltas are so lopsided that there is no middle ground—each card wins its respective territory by 43–51%.

Architecture Differences

The GTX 680 uses the GK104 chip built on TSMC's 28 nm process, packing 3,540 million transistors into a 294 mm² die with a transistor density of 12.0M per mm². The Radeon Pro 555 counters with the Polaris 21 chip on GlobalFoundries' 14 nm process, containing 3,000 million transistors in a much smaller 123 mm² die, achieving 24.4M transistors per mm². The density difference is stark—AMD fit nearly twice the transistors per square millimeter, a direct consequence of the newer process node.

Kepler architecture brings 1,536 shading units, 128 texture mapping units, and 32 ROPs. GCN 4.0 in the Radeon Pro 555 delivers 768 shading units, 48 TMUs, and only 16 ROPs. NVIDIA's card has exactly double the shading units, nearly 2.7 times the TMUs, and double the ROPs. That hardware advantage explains why the GTX 680 wins in raw compute throughput tests—more parallel units working on OpenCL and Vulkan workloads.

Memory configurations differ substantially too. Both cards use 2 GB of GDDR5, but the GTX 680 runs a 256-bit bus with 192.3 GB/s bandwidth, while the Radeon Pro 555 is limited to a 128-bit bus and 81.60 GB/s. The memory clock also favors NVIDIA: 1502 MHz (6 Gbps effective) vs 1275 MHz (5.1 Gbps effective). The GTX 680 delivers more than double the memory bandwidth, which likely contributes to its compute API wins.

Power and physical design tell a story of different intended use cases. The GTX 680 draws 195 W TDP, requires dual-slot cooling, two 6-pin power connectors, and a 450 W suggested PSU. The Radeon Pro 555 is an integrated-class part at 75 W TDP, uses no power connectors, and is described as "IGP" slot width—it is designed to live inside a laptop, not a desktop tower. The GTX 680 is explicitly a desktop card with 256 mm length and 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2 outputs. The Radeon Pro 555's display outputs are "Portable Device Dependent," confirming its mobile-first design.

API support also diverges. The GTX 680 supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The Radeon Pro 555 supports DirectX 12 (12_0), OpenGL 4.6, and the newer Vulkan 1.3. AMD's part has a more complete DirectX 12 feature set and a higher Vulkan version, yet still loses decisively in Vulkan benchmarks—evidence that API version alone does not determine performance.

FAQ

Q: Which card has the higher average benchmark score?

A: The NVIDIA GeForce GTX 680 averages 14,150 across all benchmarks, while the AMD Radeon Pro 555 averages 13,407. That is a 5.5% gap in NVIDIA's favor, placing the cards in adjacent percentile ranks (55th vs 54th).

Q: Why does the Radeon Pro 555 win Geekbench Metal by such a large margin?

A: The Radeon Pro 555 scores 16,236 in Metal versus 7,897 for the GTX 680, a 51.4% delta. The data does not specify the cause, but the magnitude suggests architectural optimization for Metal in GCN 4.0 rather than any minor tweak.

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

A: The GTX 680 leads with 16,906 points against 11,682 for the Radeon Pro 555, a 44.7% advantage. This is one of two major wins for NVIDIA in the head-to-head benchmarks.

Q: What is the memory bandwidth difference?

A: The GTX 680 delivers 192.3 GB/s over a 256-bit bus, while the Radeon Pro 555 provides 81.60 GB/s over a 128-bit bus. NVIDIA's bandwidth is more than double AMD's figure.

Q: Which card has more shading units?

A: The GTX 680 has 1,536 shading units, exactly double the 768 in the Radeon Pro 555. The TMU count is 128 vs 48, and ROPs are 32 vs 16—all favoring NVIDIA.

Q: What are the power requirements?

A: The GTX 680 has a 195 W TDP, needs dual-slot cooling and two 6-pin connectors, with a 450 W suggested PSU. The Radeon Pro 555 has a 75 W TDP, requires no power connectors, and has no suggested PSU listed.

Specification Differences

| Specification | NVIDIA GeForce GTX 680 | AMD Radeon Pro 555 |

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

| Process Node | 28 nm (TSMC) | 14 nm (GlobalFoundries) |

| Transistors | 3,540 million | 3,000 million |

| Die Size | 294 mm² | 123 mm² |

| Transistor Density | 12.0M / mm² | 24.4M / mm² |

| Base Clock | 1006 MHz | Not listed |

| Boost Clock | 1058 MHz | Not listed |

| Memory Clock | 1502 MHz (6 Gbps effective) | 1275 MHz (5.1 Gbps effective) |

| Memory Bus Width | 256 bit | 128 bit |

| Memory Bandwidth | 192.3 GB/s | 81.60 GB/s |

| Shading Units | 1536 | 768 |

| TMUs | 128 | 48 |

| ROPs | 32 | 16 |

| Pixel Rate | 33.86 GPixel/s | 13.60 GPixel/s |

| Texture Rate | 135.4 GTexel/s | 40.80 GTexel/s |

| FP32 | 3.250 TFLOPS | 1,305.6 GFLOPS |

| FP16 | Not listed | 1,305.6 GFLOPS (1:1) |

| TDP | 195 W | 75 W |

| Slot Width | Dual-slot | IGP |

| Power Connectors | 2x 6-pin | None |

| Suggested PSU | 450 W | Not listed |

| Bus Interface | PCIe 3.0 x16 | PCIe 3.0 x8 |

| Display Outputs | 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 | Portable Device Dependent |

| DirectX | 12 (11_0) | 12 (12_0) |

| Vulkan | 1.2.175 | 1.3 |

| Release Date | 2012-03-21 | 2017-06-04 |

Head-to-Head Benchmarks

The Metal benchmark is the outlier that demands attention. The Radeon Pro 555 scores 16,236 against 7,897 for the GTX 680, a 51.4% delta. This is the single largest percentage gap in any test between these two cards. The GTX 680 does not just lose; it loses by more than half its own score. For context, the GTX 680's nearest rival in average score is the NVIDIA Tesla K10 at 14,029 (0.9% delta), and the Radeon Pro 555's closest competitor is the AMD Radeon HD 8950M at 13,376 (0.2% delta)—neither of those comparisons approaches the 51.4% swing seen in Metal.

The OpenCL test flips the script entirely. The GTX 680 posts 16,906, beating the Radeon Pro 555's 11,682 by 44.7%. This is a mirror image of the Metal result, just with slightly smaller magnitude. The GTX 680's nearest rival, the GeForce GTX 1070 Ti, sits at 14,277 (0.9% away), meaning the GTX 680's OpenCL score is far above its own average—this is a workload where Kepler shines.

Vulkan shows similar NVIDIA dominance: 17,646 vs 12,303, a 43.4% delta. The GTX 680's Vulkan score is its best of the three tests, even exceeding its OpenCL result. The Radeon Pro 555's Vulkan score (12,303) is actually its second-best, ahead of OpenCL (11,682) but far behind Metal (16,236). This pattern suggests the Radeon Pro 555 is heavily optimized for Metal specifically, while the GTX 680 maintains consistent high performance across OpenCL and Vulkan.

The win count is 2–1 in NVIDIA's favor, but the aggregate averages tell a closer story: 14,150 vs 13,407. That 5.5% overall gap hides the fact that each card utterly dominates in its preferred API. The GTX 680's wins are consistent across two APIs, while the Radeon Pro 555's single win is the most extreme outlier.

The Verdict

The data points to a clear split decision. For users running Metal-based applications—primarily macOS environments, given the Radeon Pro 555's "Radeon Pro Mac" generation label—the AMD card is the only rational choice. Its 16,236 Metal score is not just better; it is 51.4% ahead, a margin that no driver tweak or workload adjustment is likely to overcome.

For OpenCL or Vulkan workloads, the GTX 680 is the definitive winner. Its 44.7% OpenCL lead and 43.4% Vulkan lead are nearly as decisive as AMD's Metal advantage, and they cover more API territory. The GTX 680 also wins on raw hardware specifications: double the shading units, more than double the memory bandwidth, higher pixel and texture rates, and 2.5 times the FP32 throughput (3.250 TFLOPS vs 1,305.6 GFLOPS).

The architecture gap explains the outcome. The GTX 680's 28 nm Kepler design uses more transistors (3,540M vs 3,000M), a larger die (294 mm² vs 123 mm²), and a wider memory bus (256-bit vs 128-bit) to deliver superior raw compute. The Radeon Pro 555's 14 nm Polaris chip achieves higher transistor density (24.4M/mm² vs 12.0M/mm²) and much lower power draw (75 W vs 195 W), but that efficiency does not translate into compute API wins outside of Metal.

Choose the GTX 680 if your work involves OpenCL or Vulkan acceleration and you have the power budget (195 W TDP, dual-slot, 450 W PSU). Choose the Radeon Pro 555 if Metal is your primary API and you need a low-power, connector-free solution at 75 W. The benchmarks do not support a single "best" card—they support two different cards for two different worlds.