NVIDIA GeForce GTX 660 Ti

The NVIDIA GeForce GTX 660 Ti, built on the Kepler architecture with the GK104 chip, delivers benchmark scores that place it in the 57th percentile of all GPUs tested, with an average benchmark score of 15,303. Its performance profile is closely matched by several rivals, with deltas of less than 2% in either direction, making it a balanced mid-generation option that now sits at end-of-life status.

Power and Cooling

The GTX 660 Ti carries a TDP of 150 W, which is modest for its era and aligns with its dual-slot cooler design. The card requires two 6-pin power connectors, and NVIDIA recommends a system power supply of 450 W. This combination of power draw and connector requirements indicates that the card is suitable for typical desktop builds of its generation without demanding oversized PSUs. The 28 nm manufacturing process from TSMC, housing 3,540 million transistors on a 294 mm² die, contributes to a transistor density of 12.0M per mm², which helps explain the relatively contained power envelope. The dual-slot footprint and 241 mm (9.5 inches) length mean the card fits standard mid-tower cases, but users should verify clearance given the two-slot cooler. There is no evidence in the data suggesting that this card supports lower-power idle states beyond what the 150 W TDP implies, so cooling should be planned around sustained load behavior.

Ray Tracing and Feature Set

The GTX 660 Ti does not include dedicated RT cores or tensor cores, as the data lists these fields as null. Consequently, hardware-accelerated ray tracing is not part of this card’s feature set. Instead, the card relies on traditional rasterization through its 1,344 shading units, 112 texture mapping units, and 24 ROPs. For API support, the card offers DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175. The DirectX 12 (11_0) designation indicates feature level 11_0, which means the card can run DirectX 12 titles but without the higher-tier features like bindless resources or conservative rasterization. Vulkan 1.2.175 support provides access to modern cross-platform graphics APIs, though the absence of RT cores limits any real-time ray tracing workloads. Display outputs include two DVI ports, one HDMI 1.4a, and one DisplayPort 1.2, enabling multi-monitor setups but without the bandwidth for high refresh rates at 4K over a single connection. The pixel rate of 27.44 GPixel/s and texture rate of 109.8 GTexel/s define the card’s fill-rate capabilities, which are more aligned with 1080p gaming than with ray-traced effects.

Benchmark Performance

Benchmark results show the GTX 660 Ti achieving a Geekbench OpenCL score of 15,095 and a Geekbench Vulkan score of 15,511. These two scores produce an average benchmark score of 15,303, which places the card in the 57th percentile of all GPUs in the database. The Vulkan score is notably higher than the OpenCL score by about 2.8%, suggesting that the card handles modern API workloads slightly better than generic compute tasks, though the difference is small. Compared to its nearest rivals, the GTX 660 Ti sits in a tightly contested performance band. It is 0.7% ahead of the NVIDIA GeForce GTX 580, which scores 15,201, meaning the two cards are effectively performance equals in synthetic benchmarks. Similarly, the GTX 660 Ti is 0.7% faster than the NVIDIA GeForce RTX 3050 OEM, which scores 15,199. Against the NVIDIA GeForce GTX 965M, the GTX 660 Ti leads by 1.3%, with the mobile chip scoring 15,109. However, the AMD Radeon R9 M380 edges out the GTX 660 Ti by 1.3%, scoring 15,504. These deltas are all within a 2.6% spread, indicating that the GTX 660 Ti’s performance is indistinguishable from its direct competitors in aggregate benchmarks. The FP32 compute throughput of 2.634 TFLOPS provides context for these scores, though the benchmarking data itself is the primary metric for comparison.

How It Compares

NVIDIA GeForce GTX 580: The GTX 660 Ti is 0.7% faster than the GTX 580, making the two cards near-identical in synthetic performance. This is noteworthy because the GTX 580 is from the previous generation, yet the Kepler architecture only manages a marginal improvement in these tests. The 150 W TDP of the GTX 660 Ti versus the older card’s higher power draw is not directly stated, but the performance parity suggests the 660 Ti offers a generational efficiency gain without a speed leap.

NVIDIA GeForce RTX 3050 OEM: The GTX 660 Ti outperforms the RTX 3050 OEM by 0.7%, which is surprising given the latter’s newer architecture and RT capabilities. However, the RTX 3050 OEM’s score of 15,199 is nearly identical, so the delta is within measurement noise. This comparison highlights that raw benchmark scores do not account for feature differences like ray tracing, which the GTX 660 Ti lacks entirely.

NVIDIA GeForce GTX 965M: The GTX 660 Ti leads the GTX 965M by 1.3%, with the mobile GPU scoring 15,109. The gap is small, but the desktop card’s higher sustained performance potential may be more relevant in long gaming sessions. The GTX 965M’s mobile nature likely means lower power limits, yet its benchmark score is close enough to make the 660 Ti’s advantage modest.

AMD Radeon R9 M380: The R9 M380 is 1.3% faster than the GTX 660 Ti, scoring 15,504. This is the only rival that beats the GTX 660 Ti in this comparison group, but the margin is slim. The R9 M380 is also a mobile part, which makes its slight lead over a desktop card notable. In practice, the difference would be imperceptible in most games.

Who Should Consider It

The GTX 660 Ti’s benchmark scores indicate it is suited for 1080p gaming at medium to high settings, but the data does not provide specific frame rate figures. Given its 57th percentile standing and the tight competition with the GTX 580 and RTX 3050 OEM, the card can handle older DirectX 11 titles comfortably. For modern games, the DirectX 12 (11_0) support means the card will run them but may lack optimizations for newer features. The pixel rate of 27.44 GPixel/s and texture rate of 109.8 GTexel/s suggest that 1080p resolution is the sweet spot; at 1440p, the 24 ROPs and 192-bit memory bus would likely become bottlenecks, though no specific resolution-based scores are provided. Users targeting high refresh rates at 1080p should look elsewhere, as the card’s raw compute of 2.634 TFLOPS is limited. The 2 GB VRAM is sufficient for 1080p textures in most titles, but for large open-world games or modded content, capacity may be tight. The card’s end-of-life status means it is only relevant for budget builds or legacy systems, not for future-proofing.

FAQ

Q: Does the GTX 660 Ti support hardware ray tracing?

A: No. The data lists RT cores as null, and the card relies on standard shading units without dedicated ray tracing hardware.

Q: What is the average benchmark score of the GTX 660 Ti?

A: The average benchmark score is 15,303, derived from a Geekbench OpenCL score of 15,095 and a Geekbench Vulkan score of 15,511.

Q: How does the GTX 660 Ti compare to the GTX 580 in performance?

A: The GTX 660 Ti is 0.7% faster than the GTX 580, with the latter scoring 15,201, making them performance equivalents.

Q: What power supply is recommended for the GTX 660 Ti?

A: The suggested PSU is 450 W, and the card uses two 6-pin power connectors with a TDP of 150 W.

Q: What is the VRAM capacity and memory bandwidth?

A: The card has 2 GB of GDDR5 memory on a 192-bit bus, providing a bandwidth of 144.2 GB/s.

Q: Which API versions does the GTX 660 Ti support?

A: It supports DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175.

Memory Subsystem

The GTX 660 Ti is equipped with 2 GB of GDDR5 memory, which is adequate for 1080p gaming but may be limiting at higher resolutions or with high-resolution texture packs. The memory bus is 192 bits wide, and the memory clock runs at 1502 MHz, yielding an effective data rate of 6 Gbps. This configuration produces a memory bandwidth of 144.2 GB/s. In the context of its benchmark scores, this bandwidth is sufficient for the card’s fill-rate capabilities, but it is a potential constraint when compared to wider-bus competitors. The 24 ROPs combined with the 192-bit bus mean that at resolutions above 1080p, pixel throughput and memory bandwidth could become the limiting factors, especially in games that demand large frame buffers. The 2 GB VRAM size is also a consideration for modern titles that often exceed this capacity at high settings. However, for the card’s intended era and performance tier, the memory subsystem is balanced relative to the GPU’s compute power, as evidenced by the close rival deltas of 0.7% to 1.3%. The end-of-life production status further suggests that users should not expect driver optimizations to improve memory handling beyond current levels.