AMD Radeon HD 8690M vs NVIDIA GeForce GTX 670M Comparison

The NVIDIA GeForce GTX 670M and AMD Radeon HD 8690M are both end-of-life mobile graphics solutions from different architectural generations, yet they land surprisingly close in overall compute performance. Based on the single available OpenCL benchmark, the GTX 670M edges out the HD 8690M by a 6.1% margin, with scores of 6513 and 6137 respectively. This places the NVIDIA part in the 38th percentile of all GPUs, while the AMD part sits at the 36th percentile, indicating both are near the lower-midrange of the performance spectrum. The data reveals a classic trade-off: one card leverages a larger, older chip with more memory bandwidth, while the other uses a smaller, newer process node with higher clock speeds. This analysis breaks down where each GPU wins, how their architectures diverge, and what the benchmark numbers mean for real-world usage.

Where Each One Wins

The GTX 670M takes the clear win in raw compute performance, as shown by the sole head-to-head benchmark result. It scores 6513 in Geekbench OpenCL, which is 6.1% ahead of the HD 8690M’s 6137. That 6.1% delta is the only benchmark where both cards are directly compared, and it favors NVIDIA across the board. Looking at the GTX 670M’s nearest rivals, it sits just 0.3% above the GT 555M and 0.5% above the Quadro M5000M, while trailing the Intel UHD Graphics P750 by 0.6%. This tight clustering suggests the GTX 670M is competitive with a broad range of integrated and discrete solutions, but it does not dramatically outclass any of them.

The HD 8690M, despite losing the head-to-head, still has its own competitive positioning. Its nearest rivals include the Intel Iris Pro Graphics 6200, which it beats by 0.3%, and the NVIDIA RTX A400 and GeForce MX230, which it leads by 1% each. The data shows the HD 8690M is only 1.2% behind the AMD Radeon HD 6950, a desktop card from an older generation. In practical terms, the HD 8690M wins against a handful of integrated and entry-level discrete GPUs, but it does not have a single benchmark victory over the GTX 670M. The GTX 670M wins on overall compute, while the HD 8690M wins on architectural efficiency and newer feature support, though not on raw score.

Architecture Differences

The two GPUs come from fundamentally different design philosophies. The GTX 670M is built on NVIDIA’s Fermi 2.0 architecture, using the GF114 chip manufactured on a 40 nm process at TSMC. This is a large die at 332 mm², packing 1,950 million transistors, which yields a transistor density of 5.9 million per square millimeter. The HD 8690M, by contrast, uses AMD’s GCN 1.0 architecture with the Sun chip on a 28 nm process, also from TSMC. Its die is drastically smaller at 56 mm², with 690 million transistors, giving it a much higher density of 12.3 million per square millimeter. The 28 nm node is a full process generation ahead of the 40 nm node, which explains how AMD fits a modern architecture into a fraction of the silicon area.

The memory subsystems are also very different. The GTX 670M uses 1536 MB of GDDR5 on a 192-bit bus, delivering a memory bandwidth of 72.00 GB/s with a memory clock of 750 MHz (3 Gbps effective). The HD 8690M has 2 GB of GDDR5, but on a much narrower 64-bit bus, which limits its bandwidth to 32.00 GB/s despite a faster memory clock of 1000 MHz (4 Gbps effective). This means the GTX 670M has more than double the memory bandwidth, which is critical for texture-heavy workloads. In terms of compute units, the GTX 670M has 336 shading units, 56 texture mapping units, and 24 ROPs, while the HD 8690M has 320 shading units, 20 TMUs, and only 8 ROPs. The NVIDIA card also leads in fill rates, with 8.372 GPixel/s and 33.49 GTexel/s versus the AMD card’s 7.800 GPixel/s and 19.50 GTexel/s.

Clock speeds tell the opposite story. The HD 8690M has a base clock of 925 MHz and a boost clock of 975 MHz, while the GTX 670M does not list base or boost clocks in the data, only its memory clock. This higher clock speed helps the AMD card partially compensate for its lower core counts. The GTX 670M’s FP32 throughput is 803.7 GFLOPS, while the HD 8690M is lower at 624.0 GFLOPS. In terms of API support, the GTX 670M supports DirectX 12 (11_0) and OpenGL 4.6, but has no Vulkan support listed. The HD 8690M supports DirectX 12 (11_1), OpenGL 4.6, and Vulkan 1.2.170, making it slightly more future-proof on the API front.

Head-to-Head Benchmarks

The only direct benchmark comparison is the Geekbench OpenCL test, where the GTX 670M scores 6513 against the HD 8690M’s 6137. This 6.1% difference is modest but consistent with the hardware specifications. The GTX 670M’s advantage comes primarily from its wider memory bus and higher bandwidth. With 72.00 GB/s versus 32.00 GB/s, the NVIDIA card can feed its 336 shading units much faster than the AMD card can feed its 320 units. The GTX 670M also has more texture units (56 versus 20) and more ROPs (24 versus 8), which directly improves texture and pixel throughput.

The HD 8690M fights back with higher clocks. Its 975 MHz boost clock is significantly higher than the GTX 670M’s unlisted core clock, but the AMD card’s smaller core configuration limits its overall output. The FP32 numbers confirm this: 803.7 GFLOPS for the GTX 670M versus 624.0 GFLOPS for the HD 8690M, a 28.8% gap in raw compute. Despite this, the HD 8690M’s 6.1% deficit in the actual benchmark is smaller than the compute gap would suggest, indicating that GCN 1.0 may be more efficient per FLOP in this workload. The HD 8690M’s 2 GB memory capacity is larger than the GTX 670M’s 1536 MB, which could help in memory-constrained scenarios, but the bandwidth bottleneck of the 64-bit bus likely negates that advantage in OpenCL compute tasks.

The nearest rival data further contextualizes these scores. The GTX 670M is 0.3% above the GT 555M, which suggests it is only marginally better than its immediate predecessor. It is 0.5% above the Quadro M5000M, a professional card, and 0.6% above the AMD Vega 10 Mobile. The HD 8690M is 0.3% above the Iris Pro 6200, 1% above both the RTX A400 and MX230, and 1.2% below the HD 6950. These deltas show that both cards are clustered within a narrow performance band, with the GTX 670M holding a slight edge over most rivals while the HD 8690M often trails by less than 1%.

FAQ

Q: Which GPU has the higher benchmark score?

A: The NVIDIA GeForce GTX 670M scores 6513 in Geekbench OpenCL, which is 6.1% higher than the AMD Radeon HD 8690M’s score of 6137.

Q: How does the memory bandwidth compare between the two?

A: The GTX 670M has a 192-bit bus and 72.00 GB/s bandwidth, while the HD 8690M has a 64-bit bus and 32.00 GB/s bandwidth. The NVIDIA card has more than double the bandwidth.

Q: What are the process node differences?

A: The GTX 670M is built on a 40 nm process, while the HD 8690M uses a 28 nm process. Both are manufactured by TSMC, but the 28 nm node allows the AMD card to have a much smaller die size of 56 mm² versus 332 mm².

Q: Does the HD 8690M support any API that the GTX 670M does not?

A: Yes, the HD 8690M supports Vulkan 1.2.170, while the GTX 670M has no Vulkan support listed. Both support OpenGL 4.6, but the GTX 670M is limited to DirectX 12 (11_0) compared to the HD 8690M’s DirectX 12 (11_1).

Q: Which card has higher clock speeds?

A: The HD 8690M has a base clock of 925 MHz and a boost clock of 975 MHz. The GTX 670M does not list base or boost clocks in the data, only a memory clock of 750 MHz (3 Gbps effective).

Q: How do the two cards compare in terms of shading units and texture units?

A: The GTX 670M has 336 shading units and 56 texture units, while the HD 8690M has 320 shading units and 20 texture units. The NVIDIA card leads in both counts.

The Verdict

The data points to the NVIDIA GeForce GTX 670M as the stronger performer for compute-heavy tasks. It wins the only available benchmark by 6.1%, and its hardware specifications support that result. The GTX 670M offers higher FP32 throughput (803.7 GFLOPS versus 624.0 GFLOPS), more texture units (56 versus 20), more ROPs (24 versus 8), and significantly more memory bandwidth (72.00 GB/s versus 32.00 GB/s). For any workload that depends on texture filtering, pixel fill, or memory throughput, the GTX 670M is the clear choice. Its larger die and older architecture are compensated by raw resource counts.

The AMD Radeon HD 8690M is the more efficient design on paper, with a 28 nm process, smaller die, and higher clock speeds, but that efficiency does not translate into a benchmark win. Its 2 GB memory capacity is higher than the GTX 670M’s 1536 MB, and it supports Vulkan 1.2.170, which the NVIDIA card lacks. However, the 64-bit memory bus is a severe limitation, cutting bandwidth to less than half of the GTX 670M’s. Users who prioritize API compatibility, particularly Vulkan support, or who need more than 1.5 GB of VRAM might prefer the HD 8690M. Those who prioritize raw compute performance, texture throughput, or memory bandwidth should choose the GTX 670M. The benchmark data gives the NVIDIA card the overall win, with the AMD card offering a more modern feature set but lower performance.

Specification Differences

| Specification | NVIDIA GeForce GTX 670M | AMD Radeon HD 8690M |

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

| Chip | GF114 | Sun |

| Architecture | Fermi 2.0 | GCN 1.0 |

| Process Node | 40 nm | 28 nm |

| Transistors | 1,950 million | 690 million |

| Die Size | 332 mm² | 56 mm² |

| Transistor Density | 5.9M / mm² | 12.3M / mm² |

| Memory Size | 1536 MB | 2 GB |

| Memory Type | GDDR5 | GDDR5 |

| Memory Bus Width | 192 bit | 64 bit |

| Memory Bandwidth | 72.00 GB/s | 32.00 GB/s |

| Shading Units | 336 | 320 |

| TMUs | 56 | 20 |

| ROPs | 24 | 8 |

| Pixel Rate | 8.372 GPixel/s | 7.800 GPixel/s |

| Texture Rate | 33.49 GTexel/s | 19.50 GTexel/s |

| FP32 | 803.7 GFLOPS | 624.0 GFLOPS |

| TDP | 75 W | Not listed |

| Bus Interface | MXM-B (3.0) | PCIe 3.0 x8 |

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

| Vulkan Support | Not listed | 1.2.170 |

| Release Date | 2012-03-21 | 2013-02-28 |