AMD Radeon 550X vs NVIDIA GeForce GTX 660 Comparison

The NVIDIA GeForce GTX 660 and AMD Radeon 550X occupy the same percentile ranking among all GPUs, yet their benchmark results tell a story of clear separation. The GTX 660, a Kepler-generation card from 2012, consistently outperforms the newer Polaris-based Radeon 550X in compute workloads, despite the latter arriving seven years later with a more efficient manufacturing process. The data shows a straightforward hierarchy: the older NVIDIA card leads in every head-to-head comparison, though the Radeon 550X counters with dramatically lower power demands and a smaller physical footprint.

Head-to-Head Benchmarks

The GeForce GTX 660 dominates both available benchmark comparisons, and the margins are substantial. In Geekbench OpenCL, the GTX 660 scores 11,347 against the Radeon 550X's 8,866, producing a 28% advantage. The Vulkan result is nearly identical in percentage terms: 11,415 versus 8,970, a 27.3% lead for the NVIDIA card. These are not marginal wins; they represent a full tier of performance separation in compute throughput.

Looking at the aggregate picture, the GTX 660's average benchmark score of 9,022 places it 1.2% ahead of the Radeon 550X's 8,918. Interestingly, the GTX 660 also edges out the NVIDIA TITAN V CEO Edition by 0.2% in average score, though that comparison sits outside the head-to-head scope. The Radeon 550X, for its part, beats the AMD Radeon Pro WX 5100 by 0.6% and the AMD Radeon R9 M265X by 0.8% in average score. The GTX 660's nearest rival, the NVIDIA TITAN V CEO Edition, sits just 0.2% above it, while the GeForce GTX 560 trails by 0.4%.

The Vulkan test highlights the GTX 660's stronger compute architecture. A 27.3% delta in Vulkan suggests that the Kepler design's raw shader throughput translates well into modern API workloads, even with its older feature set. The OpenCL gap of 28% reinforces that this is not a single-API anomaly but a consistent compute advantage. For any workload that leverages OpenCL or Vulkan, the GTX 660 is the clear pick based on these numbers.

Architecture Differences

The GTX 660 uses the GK106 chip built on TSMC's 28 nm process, packing 2,540 million transistors into a 221 mm² die. That yields a transistor density of 11.5M per mm². The Radeon 550X counters with the Lexa chip on GlobalFoundries' 14 nm node, containing 2,200 million transistors on a much smaller 103 mm² die, achieving 21.4M per mm² density. The newer process node gives AMD a density advantage, but the NVIDIA chip still fields more raw compute resources.

The GTX 660's compute configuration is substantially larger: 960 shading units, 80 texture mapping units, and 24 ROPs. The Radeon 550X halves those numbers in most categories, with 512 shading units, 32 TMUs, and 16 ROPs. This explains the benchmark gap — the GTX 660 has nearly double the shading units and more than double the texture units. Clock speeds tell a different story: the Radeon 550X runs at 1,082 MHz base and 1,218 MHz boost, while the GTX 660 sits lower at 980 MHz base and 1,032 MHz boost. The AMD card's higher clocks cannot compensate for its smaller execution footprint.

Memory configurations differ in bus width and bandwidth. The GTX 660 uses a 192-bit bus with 144.2 GB/s bandwidth, while the Radeon 550X operates on a 128-bit bus with 112.0 GB/s. Both have 2 GB of GDDR5, but the NVIDIA card's wider bus gives it 28.7% more memory bandwidth. The Radeon 550X has a higher effective memory speed at 7 Gbps versus 6 Gbps, but the narrower bus limits the practical benefit.

Power consumption is where the Radeon 550X takes a decisive lead. The GTX 660 draws 140 W TDP with a 1x 6-pin power connector and a 300 W suggested PSU. The Radeon 550X sips just 50 W TDP, requires no power connectors, and only suggests a 250 W PSU. The AMD card also runs cooler in terms of power envelope, though both are dual-slot designs. Physically, the GTX 660 measures 241 mm (9.5 inches) in length, while the Radeon 550X is a compact 145 mm (5.7 inches).

API support also differs. The GTX 660 supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The Radeon 550X supports DirectX 12 (12_0), OpenGL 4.6, and Vulkan 1.3. The newer AMD card has a more modern feature set, including full DirectX 12_0 support and a newer Vulkan version, which could matter for specific compatibility scenarios.

Where Each One Wins

The GTX 660 wins in pure compute performance. Its 1.981 TFLOPS FP32 throughput dwarfs the Radeon 550X's 1,247.2 GFLOPS. The pixel rate is slightly higher at 20.64 GPixel/s versus 19.49 GPixel/s, and the texture rate is more than double at 82.56 GTexel/s versus 38.98 GTexel/s. For any application that scales with shader count, texture fill, or memory bandwidth, the GTX 660 is the superior choice. The benchmark data confirms this across both OpenCL and Vulkan.

The Radeon 550X wins on efficiency and physical integration. Its 50 W TDP means it can run in systems with modest power supplies and no auxiliary power connectors. The 145 mm length makes it suitable for small form factor cases where the GTX 660's 241 mm board would not fit. The newer 14 nm process node also means less heat generation, which could translate to quieter operation in constrained chassis.

The Radeon 550X also offers a more modern display output suite with HDMI 2.0b and DisplayPort 1.4a, versus the GTX 660's HDMI 1.4a and DisplayPort 1.2. For users connecting to high-refresh or high-resolution displays, the AMD card's newer output standards are a practical advantage. The Radeon 550X also supports Vulkan 1.3, which is a newer API version than the GTX 660's 1.2.175.

The Verdict

The data is unambiguous: the GTX 660 outperforms the Radeon 550X in every measured benchmark. If compute performance is the priority, the GTX 660 is the only logical choice. Its 28% OpenCL lead and 27.3% Vulkan lead are decisive margins that no configuration change can overcome. The GTX 660's larger shader count, wider memory bus, and higher texture rate all contribute to this dominance.

However, the Radeon 550X is not without merit. Its 50 W TDP is a fraction of the GTX 660's 140 W, and its 145 mm length opens up installation options that the 241 mm GTX 660 cannot match. For a compact build with a limited power supply, the Radeon 550X is the practical pick. The newer display outputs and Vulkan 1.3 support also give it a modern connectivity edge.

Who should pick which? Users with an existing power supply of 300 W or more and room for a 241 mm card should choose the GTX 660 for its superior compute performance. Users building a small, low-power system with a 250 W PSU and limited chassis space should choose the Radeon 550X, accepting the performance trade-off for physical and power compatibility. The GTX 660 is the performance winner; the Radeon 550X is the efficiency and form-factor winner.

FAQ

Q: Which GPU has a higher average benchmark score?

A: The NVIDIA GeForce GTX 660 has an average benchmark score of 9,022, which is 1.2% higher than the AMD Radeon 550X's 8,918.

Q: How large is the performance gap in OpenCL compute?

A: The GTX 660 scores 11,347 in Geekbench OpenCL versus the Radeon 550X's 8,866, a 28% advantage for the NVIDIA card.

Q: What is the TDP difference between the two cards?

A: The GTX 660 has a 140 W TDP and requires a 1x 6-pin power connector with a 300 W suggested PSU. The Radeon 550X has a 50 W TDP, needs no power connectors, and only suggests a 250 W PSU.

Q: Do both cards have the same memory capacity?

A: Yes, both have 2 GB of GDDR5 memory, but the GTX 660 uses a 192-bit bus with 144.2 GB/s bandwidth, while the Radeon 550X uses a 128-bit bus with 112.0 GB/s bandwidth.

Q: Which card supports a newer version of Vulkan?

A: The AMD Radeon 550X supports Vulkan 1.3, while the NVIDIA GeForce GTX 660 supports Vulkan 1.2.175.

Q: What are the physical length differences?

A: The GTX 660 is 241 mm (9.5 inches) long, while the Radeon 550X is 145 mm (5.7 inches) long, making the AMD card significantly more compact.

Specification Differences

| Specification | NVIDIA GeForce GTX 660 | AMD Radeon 550X |

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

| Chip | GK106 | Lexa |

| Architecture | Kepler | GCN 4.0 |

| Process Node | 28 nm | 14 nm |

| Foundry | TSMC | GlobalFoundries |

| Transistors | 2,540 million | 2,200 million |

| Die Size | 221 mm² | 103 mm² |

| Transistor Density | 11.5M / mm² | 21.4M / mm² |

| Base Clock | 980 MHz | 1082 MHz |

| Boost Clock | 1032 MHz | 1218 MHz |

| Memory Clock | 1502 MHz (6 Gbps effective) | 1750 MHz (7 Gbps effective) |

| Memory Bus Width | 192 bit | 128 bit |

| Memory Bandwidth | 144.2 GB/s | 112.0 GB/s |

| Shading Units | 960 | 512 |

| TMUs | 80 | 32 |

| ROPs | 24 | 16 |

| Pixel Rate | 20.64 GPixel/s | 19.49 GPixel/s |

| Texture Rate | 82.56 GTexel/s | 38.98 GTexel/s |

| FP32 Performance | 1.981 TFLOPS | 1,247.2 GFLOPS |

| FP16 Performance | N/A | 1,247.2 GFLOPS (1:1) |

| TDP | 140 W | 50 W |

| Power Connectors | 1x 6-pin | None |

| Suggested PSU | 300 W | 250 W |

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

| Display Outputs | 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 | 1x DVI, 1x HDMI 2.0b, 1x DisplayPort 1.4a |

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

| Vulkan Support | 1.2.175 | 1.3 |

| Length | 241 mm (9.5 inches) | 145 mm (5.7 inches) |

| Release Date | 2012-09-05 | 2019-03-26 |

| Launch MSRP | 229 USD | N/A |