NVIDIA GeForce GTX 1660 vs NVIDIA GeForce GTX 670 Comparison

The NVIDIA GeForce GTX 670 and the NVIDIA GeForce GTX 1660 are separated by nearly seven years of GPU architecture evolution, yet both cards occupy a similar performance tier in the current benchmark database. The GTX 670, a Kepler-generation part from 2012, and the GTX 1660, a Turing-based card from 2019, show a clear generational gap in raw compute results, but their average benchmark scores place them within a narrow band of each other. The data shows the GTX 1660 delivers decisively higher scores in modern API benchmarks, while the GTX 670 holds a slight edge in the aggregate average across all tests, a contradiction that stems from the different benchmark suites available for each card.

Where Each One Wins

The benchmark data splits cleanly along the lines of compute API and application age. The GTX 670 wins no head-to-head comparisons in the supplied dataset, with a win count of zero against the GTX 1660. Its strength lies not in any single modern test, but in its overall average benchmark score of 12773 points, which is higher than the GTX 1660's 11680 average. This aggregate advantage is driven by the GTX 670's scores in Geekbench compute tests, where it posts 15341 in OpenCL and 15553 in Vulkan. These numbers are competitive with its nearest rivals, including the NVIDIA GeForce GTX 590 and AMD Radeon Pro 455, both of which score within 0.4% of the GTX 670's average.

The GTX 1660, by contrast, wins both head-to-head benchmarks decisively. Its Geekbench OpenCL score of 47850 is nearly three times the GTX 670's result, and its Geekbench Vulkan score of 50137 is even more dominant. The GTX 1660 also has a broader benchmark profile, including PassMark tests across DirectX 9 through 12, with a strong G3D score of 11646 and a GPU compute score of 4963. However, the GTX 1660's average benchmark score of 11680 places it in the 51st percentile of all GPUs, just below the GTX 670's 52nd percentile ranking. For applications that rely on older DirectX 9 or 10 paths, the GTX 1660 shows respectable results, with a DirectX 9 score of 177 and a DirectX 10 score of 61, suggesting it retains compatibility with legacy workloads.

Architecture Differences

The architectural gap between these two cards is substantial. The GTX 670 uses the GK104 chip built on a 28 nm process at TSMC, featuring 3,540 million transistors on a 294 mm² die. This yields a transistor density of 12.0 million transistors per square millimeter. The GTX 1660 uses the TU116 chip on a 12 nm process, also from TSMC, with 6,600 million transistors packed into a 284 mm² die, achieving a density of 23.2 million per square millimeter. The newer process node allows the GTX 1660 to nearly double the transistor count while maintaining a similar die size, which is a clear indicator of generational advancement.

The GTX 670 is based on the Kepler architecture and belongs to the GeForce 600 generation, with a predecessor in the GeForce 500 series and a successor in the GeForce 700 series. The GTX 1660 is Turing-based, part of the GeForce 16 generation, and lists the GeForce 10 series as its predecessor and GeForce 20 as its successor. Neither card features dedicated ray tracing or tensor cores, so the architectural differences manifest primarily in compute throughput and memory efficiency. The Turing architecture in the GTX 1660 supports DirectX 12 (12_1) and Vulkan 1.4, while the Kepler card only reaches DirectX 12 (11_0) and Vulkan 1.2.175, limiting the GTX 670's compatibility with modern graphics APIs.

Head-to-Head Benchmarks

The two available head-to-head comparisons show a dominant performance swing in favor of the GTX 1660. In Geekbench OpenCL, the GTX 670 scores 15341 while the GTX 1660 reaches 47850, a delta of -67.9% from the perspective of the older card. This means the GTX 1660 is roughly 212% faster in this compute test, a massive gap that reflects not only higher clock speeds but also improved shader efficiency. The GTX 1660's base clock of 1530 MHz and boost of 1785 MHz dwarf the GTX 670's 915 MHz base and 980 MHz boost, which alone would account for a significant portion of the performance difference.

In Geekbench Vulkan, the results are even more lopsided. The GTX 670 scores 15553, while the GTX 1660 posts 50137, a delta of -69%. This is a 222% improvement for the newer card. The Vulkan results are particularly telling because both cards support the API, but the GTX 1660's implementation is far more efficient. The GTX 1660 also shows a strong PassMark G3D score of 11646, which is within 0.5% of the AMD Radeon RX 7800 XT's average score of 11627, according to its nearest rival data. This places the GTX 1660 in a competitive position against much more modern cards, despite its end-of-life production status.

Specification Differences

The specification sheet reveals where the GTX 1660 gains its advantage. The GTX 670 has 1344 shading units, 112 texture mapping units, and 32 ROPs, while the GTX 1660 has 1408 shading units, 88 TMUs, and 48 ROPs. The GTX 1660's higher ROP count and slightly higher shader count contribute to its pixel rate of 85.68 GPixel/s, which is over three times the GTX 670's 27.44 GPixel/s. Texture rate also favors the newer card at 157.1 GTexel/s versus 109.8 GTexel/s, despite the GTX 670 having more TMUs. The compute throughput gap is stark: the GTX 1660 delivers 5.027 TFLOPS of FP32 performance, compared to the GTX 670's 2.634 TFLOPS, a near doubling of raw compute power.

Memory configurations differ in capacity and bus width but converge on bandwidth. The GTX 670 has 2 GB of GDDR5 on a 256-bit bus, delivering 192.3 GB/s. The GTX 1660 has 6 GB of GDDR5 on a 192-bit bus, delivering 192.1 GB/s. The bandwidth is essentially identical, but the GTX 1660's larger frame buffer is a significant advantage for modern games and applications that require more than 2 GB of video memory. The memory clock is also higher on the GTX 1660 at 2001 MHz (8 Gbps effective) versus the GTX 670's 1502 MHz (6 Gbps effective). Power consumption is notably lower on the newer card at 120 W TDP versus 170 W, with a suggested power supply of 300 W compared to 450 W. The GTX 1660 also uses a single 8-pin power connector instead of two 6-pin connectors.

The physical dimensions are similar, with the GTX 670 measuring 241 mm in length and the GTX 1660 at 229 mm, both dual-slot cards. Display outputs differ, with the GTX 670 offering two DVI ports, one HDMI 1.4a, and one DisplayPort 1.2, while the GTX 1660 provides one DVI, one HDMI 2.0, and one DisplayPort 1.4a. The GTX 670 was released in May 2012 with a launch MSRP of 399 USD, while the GTX 1660 launched in March 2019 with a launch MSRP of 219 USD.

The Verdict

The data points to a clear choice for modern workloads: the GTX 1660 is the superior card for any application that leverages OpenCL or Vulkan. Its benchmark scores are roughly three times higher than the GTX 670 in both head-to-head tests, and its support for DirectX 12 (12_1) and Vulkan 1.4 ensures compatibility with current software. The 6 GB memory capacity is also a practical necessity for modern games, as the GTX 670's 2 GB frame buffer is likely to bottleneck performance in memory-intensive scenarios. The GTX 1660's lower power draw and single 8-pin connector make it easier to integrate into existing systems, and its 51st percentile ranking places it just below the GTX 670's 52nd percentile, a negligible difference in overall standing.

The GTX 670's only advantage is its higher average benchmark score of 12773, driven by its Geekbench results, but this is a misleading metric because the GTX 1660's average is calculated across a different, more extensive set of tests including PassMark suite. The GTX 670's nearest rivals, such as the AMD Radeon Pro 455 and NVIDIA GeForce GTX 590, are all within 0.6% of its average score, indicating that it performs at a level consistent with its era. For users with legacy software that relies on older DirectX versions and does not scale well with newer architectures, the GTX 670 remains a functional option, but it offers no path forward for modern gaming or compute tasks. The verdict is straightforward: the GTX 1660 wins on every relevant performance metric, and the GTX 670 should only be considered for its historical value or extremely narrow legacy use cases.