AMD Radeon R9 M275X vs NVIDIA GeForce GTX 660 Comparison

Head-to-Head Benchmarks

The recorded data shows a clear, if narrow, overall victory for the NVIDIA GeForce GTX 660 in the two benchmark tests where both GPUs appear. The AMD Radeon R9 M275X does not manage a single win in these head-to-head comparisons. In the Geekbench OpenCL test, the GTX 660 scores 11347 against the R9 M275X's 11200, a difference of 1.3%. This is a close margin, indicating that in raw compute-oriented OpenCL workloads, the two parts are nearly interchangeable. The AMD card trails by only 147 points, a gap that could easily be influenced by driver optimization or thermal state during the run.

The larger separation appears in the Geekbench Vulkan test. Here, the GTX 660 posts 11415, while the R9 M275X manages 9964. The delta expands to 12.7% in favor of NVIDIA. This is a substantial gap and suggests that the Kepler architecture's Vulkan implementation, or the GTX 660's higher raw throughput, provides a meaningful advantage in modern API workloads. The R9 M275X, despite its GCN 1.0 architecture which is generally regarded as Vulkan-capable, falls behind by 1451 points in this test. The data implies that for users prioritizing Vulkan-based applications, the GTX 660 is the stronger choice by a wide margin.

Looking at the broader context within the database, the average benchmark score for the R9 M275X is 10582, while the GTX 660's average is 9022. This is an interesting inversion. Despite losing both head-to-head tests, the AMD part has a higher average score across all recorded benchmarks. The explanation lies in the composition of their respective benchmark suites. The R9 M275X has only two recorded scores (11200 and 9964), both of which are relatively high. The GTX 660 has three recorded scores, including a Geekbench Metal score of 4305, which is significantly lower than its OpenCL and Vulkan results. This Metal score drags down the GTX 660's average, illustrating how a single weak point in a specific API can skew the overall picture. The percentileVsAllGpus metric reflects this divergence: the R9 M275X sits at the 49th percentile, while the GTX 660 is at the 45th percentile. So, while the head-to-head data favors NVIDIA, the average-position data slightly favors AMD, making the overall narrative more nuanced than a simple win-loss tally.

Architecture Differences

The two GPUs represent different architectural philosophies from their respective manufacturers. The AMD Radeon R9 M275X is built on the Venus chip, utilizing the GCN 1.0 architecture. It belongs to the "Gem System (R9 M200)" generation and was released on 2014-01-27. The NVIDIA GeForce GTX 660 uses the GK106 chip with the Kepler architecture, part of the GeForce 600 generation, released earlier on 2012-09-05. Both are fabricated on a 28 nm process at TSMC, but the similarities end there.

The transistor counts differ significantly. The NVIDIA chip packs 2,540 million transistors on a 221 mm² die, while the AMD chip has 1,500 million transistors on a smaller 123 mm² die. This results in a higher transistor density for the AMD chip at 12.2M / mm², compared to NVIDIA's 11.5M / mm². Despite the smaller die, the AMD part achieves a denser packing of transistors, though the absolute count is much lower. The GTX 660's larger die and higher transistor budget provide the physical foundation for its superior raw throughput in many tests.

Shader configuration is another point of divergence. The GTX 660 has 960 shading units, 80 texture mapping units (TMUs), and 24 raster operators (ROPs). The R9 M275X has 640 shading units, 40 TMUs, and 16 ROPs. The NVIDIA part has 50% more shading units, double the TMUs, and 50% more ROPs. These differences directly translate into the recorded fill rates. The GTX 660's pixel rate is 20.64 GPixel/s and its texture rate is 82.56 GTexel/s. The R9 M275X trails with 14.80 GPixel/s and 37.00 GTexel/s. In terms of raw compute throughput, the GTX 660 delivers 1.981 TFLOPS of FP32 performance, while the R9 M275X offers 1,184.0 GFLOPS (or 1.184 TFLOPS). This is a substantial advantage for NVIDIA, yet the OpenCL benchmark scores are nearly identical. This suggests that the benchmark workload is not purely compute-bound, or that the AMD architecture is more efficient per FLOP in that specific test.

Clock speeds and memory subsystems also differ. The GTX 660 has a base clock of 980 MHz and a boost clock of 1032 MHz, while the R9 M275X operates at 900 MHz base and 925 MHz boost. The NVIDIA card also has a higher effective memory clock of 6 Gbps, compared to the AMD's 4.5 Gbps. Memory bus width is a major factor: the GTX 660 uses a 192 bit bus, while the R9 M275X is limited to 128 bit. This yields a bandwidth of 144.2 GB/s for NVIDIA versus 72.00 GB/s for AMD, exactly double. Both cards ship with 2 GB of GDDR5 memory, but the GTX 660's wider bus and faster memory clock give it a decisive edge in bandwidth-bound scenarios, which likely contributes to its larger Vulkan victory.

The Verdict

Based strictly on the recorded data, the NVIDIA GeForce GTX 660 is the superior performer in direct comparisons. It wins both head-to-head benchmarks, with a particular dominance in Vulkan where it leads by 12.7%. The GTX 660 also offers higher raw specifications across the board: more shading units, TMUs, ROPs, higher clock speeds, and double the memory bandwidth. For any user who knows they will be running Vulkan-based applications or workloads that stress memory bandwidth, the GTX 660 is the clear choice from the data.

However, the AMD Radeon R9 M275X is not without its merits. Its average benchmark score is higher (10582 vs 9022), and it sits at a higher percentile (49th vs 45th). This is driven by its consistent performance in both of its recorded tests, neither of which is as low as the GTX 660's Metal score. The R9 M275X also has a smaller die, which historically implies lower manufacturing cost, though pricing data is not available for this comparison. If a user's workload is primarily OpenCL-based, the difference is negligible (1.3%). The R9 M275X also offers support for DirectX 12 (11_1) and Vulkan 1.2.170, while the GTX 660 supports DirectX 12 (11_0) and Vulkan 1.2.175. Both support OpenGL 4.6. The API versions are close, but the AMD part has a slightly higher DirectX feature level.

The verdict depends on the specific use case. For Vulkan-centric gaming or compute, the GTX 660 wins decisively. For general compute where OpenCL is the primary API, the two are nearly tied, and the R9 M275X's better average standing makes it a defensible pick. The data does not support a universal winner; it supports a workload-specific choice.

FAQ

Q: Which GPU wins in OpenCL performance?

A: The NVIDIA GeForce GTX 660 wins by a narrow margin. It scores 11347 in Geekbench OpenCL, while the AMD Radeon R9 M275X scores 11200, a difference of 1.3%.

Q: How large is the Vulkan performance gap between the two?

A: The gap is substantial. The GTX 660 scores 11415 in Geekbench Vulkan, while the R9 M275X scores 9964, giving NVIDIA a 12.7% advantage.

Q: Which GPU has higher memory bandwidth?

A: The GTX 660 has exactly double the memory bandwidth of the R9 M275X. The NVIDIA card delivers 144.2 GB/s, while the AMD card delivers 72.00 GB/s, due to its wider 192 bit bus and faster 6 Gbps effective memory clock.

Q: What is the FP32 compute difference?

A: The GTX 660 provides 1.981 TFLOPS of FP32 performance, while the R9 M275X provides 1,184.0 GFLOPS. The NVIDIA part is roughly 67% higher in this raw compute metric.

Q: Why does the R9 M275X have a higher average benchmark score if it loses both head-to-head tests?

A: The R9 M275X has only two recorded benchmark scores, both relatively high (11200 and 9964). The GTX 660 has three scores, including a Geekbench Metal score of 4305, which is much lower and pulls its average down to 9022.

Q: Which GPU has a higher transistor count?

A: The GTX 660 has 2,540 million transistors on a 221 mm² die. The R9 M275X has 1,500 million transistors on a 123 mm² die.

Where Each One Wins

The NVIDIA GeForce GTX 660 wins in scenarios that demand high memory bandwidth and raw shader throughput. Its 144.2 GB/s bandwidth and 960 shading units make it the clear pick for Vulkan-based games and applications, as evidenced by its 12.7% lead in the Geekbench Vulkan test. The higher pixel rate (20.64 GPixel/s) and texture rate (82.56 GTexel/s) also suggest an edge in rasterization-heavy tasks, even though no specific gaming benchmarks are recorded in the data. For users with a GeForce 600 series legacy or those targeting the OpenGL 4.6 API, the GTX 660's specifications are consistently higher.

The AMD Radeon R9 M275X wins in the context of average benchmark standing. Its 49th percentile ranking versus the GTX 660's 45th, and its higher average score of 10582, indicate that in a mixed workload that includes the API types not tested head-to-head, the AMD part holds up better. Its OpenCL score is within 1.3% of the GTX 660, making it a viable choice for OpenCL compute tasks where the performance difference is negligible. The smaller die size (123 mm²) and lower transistor count (1,500 million) may imply lower power draw, though no TDP data is available for the AMD card to confirm this. For a mobile or space-constrained system, the R9 M275X's compact design could be a deciding factor. The data clearly shows that for Vulkan, pick NVIDIA, but for a balanced average across multiple APIs, the AMD card is competitive.

Specification Differences

The following table highlights the key specification differences between the AMD Radeon R9 M275X and the NVIDIA GeForce GTX 660, based solely on the recorded data.

| Specification | AMD Radeon R9 M275X | NVIDIA GeForce GTX 660 |

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

| Chip | Venus | GK106 |

| Architecture | GCN 1.0 | Kepler |

| Generation | Gem System (R9 M200) | GeForce 600 |

| Release Date | 2014-01-27 | 2012-09-05 |

| Transistors | 1,500 million | 2,540 million |

| Die Size | 123 mm² | 221 mm² |

| Transistor Density | 12.2M / mm² | 11.5M / mm² |

| Base Clock | 900 MHz | 980 MHz |

| Boost Clock | 925 MHz | 1032 MHz |

| Memory Clock | 4.5 Gbps effective | 6 Gbps effective |

| Memory Bus Width | 128 bit | 192 bit |

| Memory Bandwidth | 72.00 GB/s | 144.2 GB/s |

| Shading Units | 640 | 960 |

| TMUs | 40 | 80 |

| ROPs | 16 | 24 |

| Pixel Rate | 14.80 GPixel/s | 20.64 GPixel/s |

| Texture Rate | 37.00 GTexel/s | 82.56 GTexel/s |

| FP32 Performance | 1,184.0 GFLOPS | 1.981 TFLOPS |

| TDP | Not specified | 140 W |

| Slot Width | Not specified | Dual-slot |

| Power Connectors | Not specified | 1x 6-pin |

| Suggested PSU | Not specified | 300 W |

| Display Outputs | Not specified | 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 |

| DirectX Support | 12 (11_1) | 12 (11_0) |

| Vulkan Support | 1.2.170 | 1.2.175 |

| Length | Not specified | 241 mm (9.5 inches) |

| Launch MSRP | Not specified | 229 USD |