NVIDIA GeForce GTX 660 vs NVIDIA GeForce GTX 950M Comparison

FAQ

Q: Which GPU has the higher average benchmark score?

A: The NVIDIA GeForce GTX 660 has an average benchmark score of 9022, placing it in the 45th percentile of all GPUs. The NVIDIA GeForce GTX 950M scores 8135 on average, which puts it in the 42nd percentile.

Q: How much faster is the GTX 660 in OpenCL workloads?

A: In the Geekbench OpenCL test, the GTX 660 scores 11347 compared to the GTX 950M's 9745. That is a 16.4% advantage for the desktop card.

Q: What is the largest performance gap between these two GPUs?

A: The largest gap appears in the Geekbench Vulkan test. The GTX 660 scores 11415, while the GTX 950M scores 6525, a difference of 74.9% in favor of the GTX 660.

Q: Do both GPUs support DirectX 12?

A: Yes, both support DirectX 12 (11_0). They also both support OpenGL 4.6. The GTX 660 supports Vulkan 1.2.175, while the GTX 950M supports Vulkan 1.4.

Q: What are the memory configurations of each card?

A: The GTX 660 comes with 2 GB of GDDR5 memory on a 192-bit bus, delivering 144.2 GB/s of bandwidth. The GTX 950M has 4 GB of DDR3 memory on a 128-bit bus, providing 28.80 GB/s of bandwidth.

Q: Which card has a higher transistor density?

A: The GTX 950M has a higher transistor density at 12.6M per mm², despite having fewer total transistors (1,870 million vs. 2,540 million) and a smaller die size (148 mm² vs. 221 mm²).

Where Each One Wins

The data is unambiguous in favor of the NVIDIA GeForce GTX 660 across both recorded benchmark tests. The GTX 660 wins in Geekbench OpenCL with a score of 11347 against the GTX 950M's 9745, a 16.4% margin. It also wins in Geekbench Vulkan by an enormous 74.9%, scoring 11415 versus 6525. The head-to-head record stands at 2 wins for the GTX 660 and 0 for the GTX 950M.

The GTX 660's wins are not narrow. In OpenCL, the 16.4% edge represents a meaningful generational and architectural gap. In Vulkan, the 74.9% margin is decisive, indicating that the older Kepler architecture is dramatically stronger in this particular low-level API workload. The GTX 950M, by contrast, does not have a single benchmark where it overtakes the GTX 660. The GTX 950M's only relative strength is its average score being closer to its nearest rivals; it sits only 0.1% above the AMD Radeon R9 M360 and 0.4% above the NVIDIA GeForce 945M, whereas the GTX 660 sits between the NVIDIA TITAN V CEO Edition and the AMD Radeon 550X in its own comparison group.

For users prioritizing compute workloads or Vulkan-based applications, the GTX 660 is the clear choice. For scenarios where the GTX 950M might be the only option, such as in a laptop, its 4 GB memory capacity and lower power draw are practical advantages, but the benchmark data shows no performance win.

Architecture Differences

The GTX 660 is built on the Kepler architecture using the GK106 chip, while the GTX 950M uses the Maxwell architecture with the GM107 chip. Both are manufactured on a 28 nm process at TSMC, so the node is identical. The fundamental architectural shift from Kepler to Maxwell explains several key differences in how these GPUs allocate resources.

Kepler, as implemented in GK106, is a larger chip: 2,540 million transistors on a 221 mm² die. Maxwell's GM107 is smaller, with 1,870 million transistors on a 148 mm² die. The transistor density actually increases with Maxwell, going from 11.5M per mm² on the GTX 660 to 12.6M per mm² on the GTX 950M. This density improvement reflects Maxwell's more efficient use of transistors per unit area.

The GTX 660 has 960 shading units, 80 texture mapping units, and 24 render output units. The GTX 950M has fewer of each: 640 shading units, 40 TMUs, and 16 ROPs. This reduction in execution resources is a direct consequence of Maxwell's design philosophy, which favored per-core efficiency over raw core counts.

Both GPUs lack dedicated ray tracing cores and tensor cores, as these are older architectures predating those features. The GTX 950M supports Vulkan 1.4, a newer version than the GTX 660's Vulkan 1.2.175, which is notable given the GTX 950M's lower raw performance. The Maxwell architecture also enables the GTX 950M to be an integrated graphics package (IGP) with no power connectors, whereas the GTX 660 is a dual-slot card requiring a 6-pin power connector.

Specification Differences

The specifications diverge significantly in several areas. Clock speeds: the GTX 660 has a base clock of 980 MHz and a boost clock of 1032 MHz. The GTX 950M runs slightly higher at 993 MHz base and 1124 MHz boost. Memory clocks differ substantially: the GTX 660's memory runs at 1502 MHz with 6 Gbps effective speed, while the GTX 950M's memory runs at 900 MHz with 1800 Mbps effective speed.

Memory configuration is a major differentiator. The GTX 660 uses 2 GB of GDDR5 on a 192-bit bus, achieving 144.2 GB/s bandwidth. The GTX 950M uses 4 GB of DDR3 on a 128-bit bus, achieving only 28.80 GB/s. The GTX 660's bandwidth advantage is massive, roughly five times that of the GTX 950M, which directly impacts performance in memory-intensive workloads.

Compute throughput also differs. The GTX 660 delivers 1.981 TFLOPS of FP32 performance, while the GTX 950M delivers 1,438.7 GFLOPS. Pixel rates are 20.64 GPixel/s for the GTX 660 and 17.98 GPixel/s for the GTX 950M. Texture rates are 82.56 GTexel/s versus 44.96 GTexel/s, respectively.

Power and physical characteristics: the GTX 660 has a TDP of 140 W, a dual-slot width, and a 6-pin power connector, with a suggested PSU of 300 W. The GTX 950M has a 75 W TDP, is an IGP, and requires no power connectors. The GTX 660 is 241 mm long (9.5 inches), while the GTX 950M has no recorded dimensions. The GTX 660 uses PCIe 3.0 x16, while the GTX 950M uses PCIe 3.0 x8. Display outputs: the GTX 660 offers 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2, whereas the GTX 950M's outputs are portable device dependent.

Release timing: the GTX 660 launched on September 5, 2012, with a launch MSRP of 229 USD. The GTX 950M launched on March 12, 2015. Both are end-of-life products. The GTX 660's predecessor is the GeForce 500 series and its successor is the GeForce 700 series, while the GTX 950M's predecessor is the GeForce 800M and its successor is the GeForce 10 Mobile.

Head-to-Head Benchmarks

The recorded head-to-head data covers two benchmark tests, and the GTX 660 wins both. The first is Geekbench OpenCL, where the GTX 660 scores 11347 against the GTX 950M's 9745. The 16.4% delta is substantial but not overwhelming, reflecting the GTX 660's larger shader count and far higher memory bandwidth. The 960 shading units and 144.2 GB/s bandwidth give the GTX 660 a clear edge in compute tasks that scale with parallel throughput.

The second test is Geekbench Vulkan, and here the gap widens dramatically. The GTX 660 scores 11415, while the GTX 950M scores 6525. The 74.9% delta is enormous and suggests that the GTX 660's Kepler architecture handles Vulkan's driver overhead and command processing far better than Maxwell in this specific workload. The GTX 950M's lower Vulkan version support (1.4 vs. 1.2.175) does not explain the performance deficit; rather, the GTX 660's higher raw compute and memory bandwidth appear to be the deciding factors.

Looking at the wider context from the database, the GTX 660's average score of 9022 places it 0.2% below the NVIDIA TITAN V CEO Edition's 9037 and 0.4% below the NVIDIA GeForce GTX 560's 9058. It sits 1.2% above the AMD Radeon 550X's 8918 and 1.8% above the AMD Radeon Pro WX 5100's 8863. The GTX 950M's average of 8135 puts it 0.1% above the AMD Radeon R9 M360's 8129, 0.4% below the NVIDIA GeForce GTX 980's 8167, 0.4% above the NVIDIA GeForce 945M's 8099, and 0.7% above the NVIDIA GRID K2's 8080.

These nearest-rival comparisons show that the GTX 660 is clustered with much stronger desktop parts, while the GTX 950M sits among lower-end mobile and workstation chips. The 74.9% Vulkan win is the single largest margin in the head-to-head, and it underscores the GTX 660's dominance in this metric. In every recorded benchmark, the GTX 660 emerges ahead, making it the superior performer in the database's measurements.