AMD Radeon HD 7970M vs NVIDIA GeForce GTX 660 Ti Comparison

# AMD Radeon HD 7970M vs NVIDIA GeForce GTX 660 Ti

The mobile graphics segment often forces a choice between raw compute throughput and feature set. This comparison places the AMD Radeon HD 7970M against the NVIDIA GeForce GTX 660 Ti, two designs from the same generation that approach performance from different architectural directions. The database records show a close contest in raw compute, but a decisive split in application-specific tests.

FAQ

Q: How does the AMD Radeon HD 7970M compare to the NVIDIA GeForce GTX 660 Ti in raw compute benchmarks?

A: In the Geekbench OpenCL test, the AMD Radeon HD 7970M scores 17,019 points, which is 12.6% ahead of the NVIDIA GeForce GTX 660 Ti's score of 15,113. The AMD card wins this head-to-head matchup.

Q: What is the performance difference in the Vulkan API test?

A: The AMD Radeon HD 7970M records a Geekbench Vulkan score of 17,019, while the NVIDIA GeForce GTX 660 Ti scores 15,012. The AMD part leads by 13.4% in this API.

Q: How do these GPUs compare to their nearest rivals in the database?

A: The AMD Radeon HD 7970M sits at the 60th percentile of all GPUs, with an average score of 17,019. Its nearest rival, the NVIDIA GeForce GTX 690, scores 17,037 (a 0.1% difference). The NVIDIA GeForce GTX 660 Ti sits at the 57th percentile, averaging 15,063 points. Its nearest rival, the AMD Radeon Pro 560X, scores 15,082 (0.1% slower).

Q3: Which GPU has the higher memory bandwidth?

A: The AMD Radeon HD 7970M provides 153.6 GB/s of bandwidth over a 256-bit bus with GDDR5 memory running at 1200 MHz (4.8 Gbps effective). The NVIDIA GeForce GTX 660 Ti offers 144.2 GB/s over a 192-bit bus with GDDR5 at 1502 MHz (6 Gbps effective).

Q: Are there differences in the feature set?

A: Yes. The AMD chip is built on the GCN 1.0 architecture and integrates 1,280 shading units, 80 texture units, and 32 ROPs. The NVIDIA part uses the Kepler architecture with 1,344 shading units, 112 texture units, and 24 ROPs.

The Verdict

Strictly from the recorded data, the AMD Radeon HD 7970M is the clear winner for compute workloads. In the OpenCL test, it beats the GTX 660 Ti by 12.6% (17,019 vs 15,113). The same pattern holds in Vulkan, where the AMD part leads by 13.4%. The delta is consistent across these APIs, which suggests an architectural advantage in raw shading throughput rather than a driver quirk. The AMD chip has a 33% higher texture rate (68.00 GTexel/s vs 109.8 GTexel/s) and a pixel rate of 27.20 GPixel/s vs 27.44 GPixel/s, so it should handle geometry-heavy scenes better. However, the NVIDIA card is no slouch: it has a higher transistor density (12.0M/mm² vs 13.2M/mm²) and, despite a lower overall score, it is surprisingly close in the OpenCL test. The data does not support choosing the NVIDIA for raw performance, but it should be considered for its broader feature set. The AMD part is the one to choose from the numbers.

Head-to-Head Benchmarks

The biggest confirmed win for AMD comes from the Geekbench OpenCL test. The Radeon HD 7970M posts 17,019 points, which is 1,906 points more than the GeForce GTX 660 Ti (15,113 points). That is a 12.6% difference. This is the single largest margin in this comparison.

The NVIDIA GeForce GTX 660 Ti has no wins in the recorded head-to-head tests. The AMD card wins the only matchup in the database. The closest results are in the OpenCL test, where the NVIDIA card trails by 12.6%, and the gap is even bigger in the Vulkan test, where the AMD lead extends to 13.4%.

The data shows a consistent pattern: the AMD design delivers higher compute scores in every recorded API. The 12.6% lead in OpenCL is mirrored by a 13.4% lead in Vulkan, indicating a stable architectural advantage. The NVIDIA card, however, is not far behind: its average score of 15,063 is 91.3% of the AMD's average score of 17,019.

Specification Differences

| Specification | AMD Radeon HD 7970M | NVIDIA GeForce GTX 660 Ti |

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

| Chip | Wimbledon | GK104 |

| Architecture | GCN 1.0 | Kepler |

| Process | 28 nm | 28 nm |

| Transistors | 2,800 million | 3,540 million |

| Die size | 212 mm² | 294 mm² |

| Memory size | 2 GB | 2 GB |

| Memory type | GDDR5 | GDDR5 |

| Memory bus width | 256 bit | 192 bit |

| Memory bandwidth | 153.6 GB/s | 144.2 GB/s |

| Shading units | 1,280 | 1,344 |

| TMUs | 80 | 112 |

| ROPs | 32 | 24 |

Note: The AMD chip packs 2,800 million transistors on a 28 nm process, resulting in a density of 13.2 million per mm². The NVIDIA chip integrates 3,540 million transistors on a 28 nm node and reaches a density of 12.0 million per mm².

Architecture Differences

The architectural split between these two GPUs is clearly visible in the recorded data. AMD's GCN 1.0 design is the more compute-heavy: it integrates 1,280 shaders, 80 texture units and 32 ROPs. NVIDIA's Kepler architecture chooses a different balance: 1,344 shaders, 112 TMUs and 24 ROPs. This explains the AMD's lead in the OpenCL and Vulkan workloads, as the larger ROP count and wider 256-bit memory interface help fill geometry and shader throughput more efficiently. The NVIDIA chip uses a 192-bit interface, which is narrower, but it compensates with a higher memory clock (1502 MHz vs 1200 MHz), which brings its effective data rate to 6 Gbps.

Both cards are mobile MXM modules (AMD uses the MXM-B 3.0 bus, NVIDIA uses a dual-slot, 6-pin power connector). The AMD chip has no external power connector, while the NVIDIA requires two 6-pin connectors. The NVIDIA is the more power-hungry card, with a 150 W TDP versus 100 W for the AMD. The AMD is end-of-life, while the NVIDIA is also end-of-life. The AMD was released on April 23, 2012, replacing the Vancouver chip. The NVIDIA was released on August 15, 2012, and its predecessor was the GeForce 500 series.