NVIDIA GeForce GTX 660 Ti vs NVIDIA GeForce RTX 3060 Ti Comparison

The NVIDIA GeForce RTX 3060 Ti and the NVIDIA GeForce GTX 660 Ti represent two distinct eras of NVIDIA graphics architecture, separated by nearly a decade of silicon evolution. The data shows a stark generational divide: the RTX 3060 Ti wins both head-to-head benchmark comparisons decisively, yet the GTX 660 Ti retains relevance in the database's percentile rankings, sitting only two percentiles below its modern counterpart. This analysis examines where each card wins, what architectural changes drove the performance gap, and which user profiles align with each product based strictly on the benchmark and specification data provided.

Where Each One Wins

The benchmark results paint a clear picture: the NVIDIA GeForce RTX 3060 Ti wins every direct comparison, but the nature of those wins matters for workload selection. In the two shared benchmarks, the RTX 3060 Ti dominates compute-oriented tasks. Its Geekbench OpenCL score of 78,927 is 422.2% higher than the GTX 660 Ti's 15,113, indicating a massive advantage in general-purpose GPU compute workloads such as rendering, physics simulation, and data processing. The Vulkan score tells a similar story: 47,784 versus 15,012, a 218.3% delta, showing the RTX 3060 Ti excels in modern graphics APIs that leverage asynchronous compute and explicit multi-threading.

The GTX 660 Ti's wins are not performance-based but rather contextual. It holds a 57th percentile ranking among all GPUs, while the RTX 3060 Ti sits at the 59th percentile. This narrow 2-percentile gap is notable given the raw score difference—the GTX 660 Ti's average benchmark score of 15,063 is only 6.6% below the RTX 3060 Ti's 16,129 average. This suggests that in legacy DirectX 10 and DirectX 9 workloads, which the GTX 660 Ti supports with native scores of 132 and 234 respectively (in PassMark tests), the older card may remain competitive in older game titles or applications that do not scale with modern hardware. The GTX 660 Ti also carries lower power requirements (150 W TDP) and a simpler power connector setup (2x 6-pin), making it a more accessible option in systems with older power supplies.

Architecture Differences

The architectural chasm between these two cards is fundamental. The RTX 3060 Ti uses the GA104 chip built on Ampere architecture, fabricated on Samsung's 8 nm process node. It packs 17,400 million transistors into a 392 mm² die, achieving a transistor density of 44.4 million per square millimeter. In contrast, the GTX 660 Ti uses the GK104 chip on Kepler architecture, built on TSMC's 28 nm node, with 3,540 million transistors on a 294 mm² die—a density of just 12.0 million per square millimeter. This represents a 3.7x increase in transistor density, enabling the RTX 3060 Ti to house 4,864 shading units versus 1,344 on the GTX 660 Ti, and 152 texture mapping units versus 112.

Memory architecture diverges equally dramatically. The RTX 3060 Ti features 8 GB of GDDR6 memory on a 256-bit bus, delivering 448.0 GB/s of bandwidth. The GTX 660 Ti is limited to 2 GB of GDDR5 on a 192-bit bus, providing 144.2 GB/s. The RTX 3060 Ti also introduces hardware features absent from Kepler: 38 ray tracing cores and 152 tensor cores, enabling DirectX 12 Ultimate (12_2) API support. The GTX 660 Ti maxes out at DirectX 12 (11_0) and lacks Vulkan 1.4 support, capping at Vulkan 1.2.175. Clock speeds follow the generational trend: the RTX 3060 Ti boosts to 1,665 MHz versus 980 MHz on the GTX 660 Ti, while memory runs at 14 Gbps effective versus 6 Gbps.

Head-to-Head Benchmarks

The two direct benchmark comparisons in the data reveal the scale of the performance gap. In Geekbench OpenCL, the RTX 3060 Ti scores 78,927 against the GTX 660 Ti's 15,113. This 422.2% delta is staggering, reflecting not just raw compute throughput but also the efficiency of Ampere's design. The RTX 3060 Ti's FP32 performance is 16.20 TFLOPS, six times the GTX 660 Ti's 2.634 TFLOPS. Pixel fill rate tells a similar story: 133.2 GPixel/s versus 27.44 GPixel/s, a 4.85x advantage. Texture fill rate reaches 253.1 GTexel/s versus 109.8 GTexel/s, a 2.3x improvement.

The Geekbench Vulkan result shows a narrower but still decisive win: 47,784 versus 15,012, a 218.3% delta. This smaller gap compared to OpenCL suggests that Vulkan's lower overhead helps the Kepler architecture close some distance, but the RTX 3060 Ti's superior memory bandwidth (448.0 GB/s versus 144.2 GB/s) and larger frame buffer (8 GB versus 2 GB) provide headroom that Kepler cannot match. The RTX 3060 Ti's PassMark G3D score of 20,349 further reinforces its dominance, though no comparable score exists for the GTX 660 Ti in the provided data. The average benchmark score difference—16,129 versus 15,063—represents a 7.1% aggregate advantage for the RTX 3060 Ti, though this figure is heavily influenced by the GTX 660 Ti's limited benchmark suite.

FAQ

Q: Which card has better raw compute performance?

A: The RTX 3060 Ti delivers 16.20 TFLOPS of FP32 compute versus 2.634 TFLOPS on the GTX 660 Ti, a six-fold advantage that manifests in the 422.2% OpenCL benchmark delta.

Q: Does the GTX 660 Ti support modern APIs?

A: It supports DirectX 12 (11_0) and Vulkan 1.2.175, but lacks ray tracing cores, tensor cores, and DirectX 12 Ultimate features found on the RTX 3060 Ti.

Q: Which card has more memory bandwidth?

A: The RTX 3060 Ti provides 448.0 GB/s over a 256-bit GDDR6 interface, compared to 144.2 GB/s on the GTX 660 Ti's 192-bit GDDR5 bus.

Q: How do their percentile rankings compare?

A: The RTX 3060 Ti sits at the 59th percentile of all GPUs, while the GTX 660 Ti ranks at the 57th percentile—a surprisingly small gap given the performance delta.

Q: What is the transistor density difference?

A: The RTX 3060 Ti achieves 44.4 million transistors per square millimeter on an 8 nm process, versus 12.0 million per square millimeter on the GTX 660 Ti's 28 nm process.

Q: Which card has more shading units?

A: The RTX 3060 Ti contains 4,864 shading units, compared to 1,344 on the GTX 660 Ti, a 3.6x increase that directly impacts pixel and texture throughput.

Specification Differences

The following specifications differ between the two cards, with all values taken directly from the data:

• Chip: GA104 (RTX 3060 Ti) versus GK104 (GTX 660 Ti)
• Architecture: Ampere versus Kepler
• Process Node: 8 nm Samsung versus 28 nm TSMC
• Transistors: 17,400 million versus 3,540 million
• Die Size: 392 mm² versus 294 mm²
• Transistor Density: 44.4M / mm² versus 12.0M / mm²
• Base Clock: 1,410 MHz versus 915 MHz
• Boost Clock: 1,665 MHz versus 980 MHz
• Memory Clock: 1,750 MHz (14 Gbps effective) versus 1,502 MHz (6 Gbps effective)
• Memory Size: 8 GB GDDR6 versus 2 GB GDDR5
• Memory Bus: 256-bit versus 192-bit
• Memory Bandwidth: 448.0 GB/s versus 144.2 GB/s
• Shading Units: 4,864 versus 1,344
• TMUs: 152 versus 112
• ROPs: 80 versus 24
• RT Cores: 38 versus none
• Tensor Cores: 152 versus none
• Pixel Rate: 133.2 GPixel/s versus 27.44 GPixel/s
• Texture Rate: 253.1 GTexel/s versus 109.8 GTexel/s
• FP32: 16.20 TFLOPS versus 2.634 TFLOPS
• TDP: 200 W versus 150 W
• Power Connectors: 1x 12-pin versus 2x 6-pin
• Suggested PSU: 550 W versus 450 W
• Bus Interface: PCIe 4.0 x16 versus PCIe 3.0 x16
• Display Outputs: 1x HDMI 2.1, 3x DisplayPort 1.4a versus 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2
• DirectX Support: 12 Ultimate (12_2) versus 12 (11_0)
• Vulkan Support: 1.4 versus 1.2.175
• Release Date: 2020-11-30 versus 2012-08-15
• Launch MSRP: 399 USD versus 299 USD

The Verdict

The benchmark data leaves no ambiguity for modern workloads: the RTX 3060 Ti is the superior card in every measurable dimension, with a 422.2% advantage in OpenCL compute and a 218.3% lead in Vulkan performance over the GTX 660 Ti. Its architectural features—38 ray tracing cores, 152 tensor cores, and DirectX 12 Ultimate support—make it the only viable option for contemporary gaming or compute applications that utilize these capabilities. The 8 GB frame buffer versus 2 GB is decisive for texture-heavy modern titles, and the 448.0 GB/s bandwidth prevents bottlenecks in high-resolution scenarios.

However, the GTX 660 Ti's 57th percentile ranking and 15,063 average score show it remains a functional card for legacy applications. Users running pre-2013 games that rely on DirectX 9 or DirectX 10 may find the GTX 660 Ti's PassMark scores (234 for DX9, 132 for DX10) adequate, and its lower 150 W TDP and 450 W suggested PSU make it easier to integrate into older systems. The 2x 6-pin power connectors are more universally compatible than the RTX 3060 Ti's 1x 12-pin, which may require an adapter for existing power supplies.

The decision ultimately comes down to workload recency. For any task that uses Vulkan, OpenCL, or modern DirectX 12 features, the RTX 3060 Ti's 16.20 TFLOPS of FP32 compute and 4,864 shading units provide a transformative performance uplift. For users with strict power budgets or systems that cannot support PCIe 4.0, the GTX 660 Ti's 150 W TDP and PCIe 3.0 interface offer a pragmatic fallback—but the data shows its 27.44 GPixel/s pixel rate and 144.2 GB/s bandwidth will struggle with any modern workload. The RTX 3060 Ti's 59th percentile ranking, while only 2 points higher, is backed by a benchmark suite that includes DirectX 12, compute, and modern API tests, making its average score of 16,129 far more representative of current software demands. Choose the RTX 3060 Ti for future-proofing; choose the GTX 660 Ti only if constrained by legacy system compatibility.