NVIDIA GeForce GTX 660 vs NVIDIA GeForce GTX 960 Comparison

The NVIDIA GeForce GTX 960 and the NVIDIA GeForce GTX 660 are two end-of-life graphics cards from different generations, and the benchmark data shows a clear generational shift in performance, though not without a surprise. The GTX 960 wins two of the three head-to-head benchmark comparisons, but the GTX 660 takes a significant victory in the Vulkan API test. The most dramatic difference is in the Geekbench Metal test, where the GTX 960 scores 8,773 against the GTX 660’s 4,305, a massive 103.8% advantage. This is a doubling of performance, indicating that the Maxwell 2.0 architecture in the GTX 960 is far more efficient in this specific workload. The OpenCL results show a similar, though less extreme, trend, with the GTX 960 scoring 18,925 compared to the GTX 660’s 11,347, a 66.8% lead. However, the Vulkan test flips the script entirely; the GTX 660 scores 11,415, which is 19.1% higher than the GTX 960’s 9,231. This single result prevents the GTX 960 from being a clean sweep and highlights that the older Kepler architecture still has strengths in certain modern API scenarios.

Head-to-Head Benchmarks

The GTX 960’s dominance in compute-heavy tests is the defining story of this comparison. The 103.8% delta in the Geekbench Metal benchmark is the largest gap between the two cards, with the GTX 960 scoring 8,773 versus the GTX 660’s 4,305. This suggests that for applications leveraging Apple’s Metal API, the GTX 960 is in a completely different performance class. The OpenCL test reinforces this narrative, with the GTX 960 achieving 18,925 points against the GTX 660’s 11,347. The 66.8% improvement is substantial, indicating that the newer card’s shading units and memory subsystem are better utilized in general-purpose GPU computing.

The Vulkan result is the outlier and the GTX 660’s only win. With a score of 11,415, it beats the GTX 960’s 9,231 by 19.1%. This is a notable upset, as Vulkan is a modern, low-level API that typically favors newer architectures. The data suggests that the GTX 660’s Kepler design, despite its age, has a specific advantage in this test. It is importantly the GTX 960’s average benchmark score across all tests is 9,273, while the GTX 660’s average is 9,022. While the GTX 960 leads overall, the delta between their average scores is only 2.7%, which is far narrower than the individual test deltas suggest. This indicates that the GTX 960’s wins are concentrated in specific areas, while the GTX 660 remains competitive in others.

FAQ

Q: Which card has a higher average benchmark score?

A: The NVIDIA GeForce GTX 960 has a higher average benchmark score of 9,273, compared to the GTX 660’s 9,022. This places the GTX 960 slightly ahead in overall synthetic performance.

Q: Is the GTX 960 faster in every benchmark?

A: No. The GTX 960 wins in Geekbench Metal and OpenCL, but the GTX 660 wins in Geekbench Vulkan with a score of 11,415 versus 9,231.

Q: What is the biggest performance gap between the two cards?

A: The biggest gap is in the Geekbench Metal test, where the GTX 960 leads by 103.8%. This is a doubling of performance and the most decisive result in the comparison.

Q: How do these cards compare to other GPUs in the database?

A: Both cards sit at the 45th percentile of all GPUs. The GTX 960’s nearest rival is the NVIDIA GeForce GTX 465, which has an average score of 9,294 and is 0.2% faster. The GTX 660’s nearest rival is the NVIDIA TITAN V CEO Edition, which scores 9,037 and is 0.2% faster.

Q: Which card has a higher boost clock speed?

A: The GTX 960 has a higher boost clock of 1178 MHz, while the GTX 660’s boost clock is 1032 MHz. The base clocks are 1127 MHz and 980 MHz, respectively.

Q: Do both cards have the same amount of memory?

A: Yes, both cards feature 2 GB of GDDR5 memory. However, the GTX 660 has a wider 192-bit memory bus, giving it higher bandwidth of 144.2 GB/s, while the GTX 960 has a 128-bit bus and 112.2 GB/s bandwidth.

Architecture Differences

The GTX 960 is built on the Maxwell 2.0 architecture with the GM206 chip, while the GTX 660 uses the older Kepler architecture with the GK106 chip. Both are fabricated on a 28 nm process at TSMC, but the GTX 960 packs more transistors: 2,940 million versus 2,540 million. The die sizes are similar, with the GTX 960 at 228 mm² and the GTX 660 at 221 mm², resulting in a higher transistor density of 12.9M / mm² for the GTX 960 compared to 11.5M / mm² for the GTX 660. This denser design is a key reason for the GTX 960’s efficiency gains.

The shading units differ, with the GTX 960 featuring 1,024 shading units and 64 TMUs, while the GTX 660 has 960 shading units and 80 TMUs. The ROP count is higher on the GTX 960 at 32, versus 24 on the GTX 660. These differences explain the GTX 960’s superior pixel rate of 37.70 GPixel/s, which is nearly double the GTX 660’s 20.64 GPixel/s. The texture rate, however, favors the GTX 660 at 82.56 GTexel/s, compared to the GTX 960’s 75.39 GTexel/s, due to its higher TMU count.

The memory clocks also differ, with the GTX 960 running at 7 Gbps effective and the GTX 660 at 6 Gbps. Despite this, the GTX 660’s wider 192-bit bus gives it a bandwidth advantage. The GTX 960’s FP32 compute is 2.413 TFLOPS, which is higher than the GTX 660’s 1.981 TFLOPS. In terms of API support, the GTX 960 supports DirectX 12 (12_1) and Vulkan 1.4, while the GTX 660 only supports DirectX 12 (11_0) and Vulkan 1.2.175. Both support OpenGL 4.6, but the GTX 960’s newer feature set is a clear architectural advantage.

Specification Differences

The most significant specification differences are in clock speeds and compute rates. The GTX 960 has a base clock of 1127 MHz and a boost clock of 1178 MHz, while the GTX 660 operates at 980 MHz base and 1032 MHz boost. This clock advantage contributes to the GTX 960’s higher FP32 performance of 2.413 TFLOPS versus 1.981 TFLOPS. The pixel rate is another major differentiator: 37.70 GPixel/s for the GTX 960 compared to 20.64 GPixel/s for the GTX 660. The texture rate goes the other way, with the GTX 660 at 82.56 GTexel/s and the GTX 960 at 75.39 GTexel/s.

Memory bandwidth is a clear win for the GTX 660. Its 192-bit bus provides 144.2 GB/s, while the GTX 960’s 128-bit bus offers 112.2 GB/s. However, the GTX 960 has a faster memory clock at 7 Gbps effective versus 6 Gbps. The GTX 960 also has a lower TDP of 120 W, compared to the GTX 660’s 140 W. Both cards are dual-slot and use a single 6-pin power connector, with a suggested PSU of 300 W. The display outputs differ: the GTX 960 offers 1x DVI, 1x HDMI 2.0, and 3x DisplayPort 1.2, while the GTX 660 has 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2. Both have the same physical length of 241 mm (9.5 inches).

Where Each One Wins

The GTX 960 is the clear winner for compute-heavy workloads and modern API performance, except for Vulkan. Its 103.8% lead in Geekbench Metal and 66.8% lead in OpenCL make it the superior choice for general-purpose GPU computing and applications that leverage these APIs. The higher pixel rate and FP32 throughput also suggest it is better suited for tasks that involve heavy rasterization and shader math. The lower TDP of 120 W versus 140 W makes it a more power-efficient option, which can be beneficial for smaller form-factor builds or systems with modest power supplies.

The GTX 660’s sole win in Geekbench Vulkan is significant for users who prioritize Vulkan-based games or applications. Its 19.1% advantage in this test is not trivial, and it indicates that the Kepler architecture has a specific optimization for this low-level API. Additionally, the GTX 660’s higher texture rate of 82.56 GTexel/s and greater memory bandwidth of 144.2 GB/s could provide advantages in scenarios that are heavily texture-bound or require high data throughput, such as certain gaming titles or content creation tasks. For users stuck on a platform that relies heavily on Vulkan, the GTX 660 might be the more practical pick despite its older architecture.

The Verdict

The data points to the GTX 960 as the overall better card. It wins two of three benchmarks, has a higher average score (9,273 vs 9,022), and delivers a dominating performance in Metal and OpenCL. The 103.8% and 66.8% deltas in those tests are decisive, and the card’s higher clock speeds, more shading units, and lower power draw make it a more capable and efficient choice for most users. The GTX 960 is the right pick for anyone prioritizing compute performance, modern API support, or power efficiency.

However, the GTX 660 is not without merit. Its Vulkan victory is a concrete, measurable advantage that cannot be ignored. For a user whose primary workload is Vulkan-based, the GTX 660’s 19.1% lead in that specific test could translate to real-world performance gains. The GTX 660 also has higher memory bandwidth and texture rate, which may be beneficial in specific scenarios. The choice comes down to the user’s workload: the GTX 960 for general excellence and compute, the GTX 660 for Vulkan-specific tasks. Both cards sit at the 45th percentile of all GPUs, but the GTX 960’s benchmark profile is stronger overall.