NVIDIA GeForce GTX 660 Ti vs NVIDIA GeForce GTX 680 Comparison

Head-to-Head Benchmarks

The data shows a clear pattern of dominance for the NVIDIA GeForce GTX 680 across both benchmark tests where direct comparisons are available. In Geekbench OpenCL, the GTX 680 scores 16,906 against the GTX 660 Ti's 15,113, representing a 10.6% advantage for the GTX 680. This is a substantial margin that reflects the GTX 680's higher raw compute capabilities. The Vulkan test shows an even wider gap: the GTX 680 reaches 17,646 while the GTX 660 Ti manages 15,012, a 14.9% difference in favor of the GTX 680. Neither benchmark goes to the GTX 660 Ti; the head-to-head tally stands at zero wins for the GTX 660 Ti and two wins for the GTX 680.

Looking at average benchmark scores, the picture becomes more nuanced. The GTX 660 Ti averages 15,063 across its two benchmark entries, while the GTX 680 averages 14,150 across three entries — despite winning both head-to-head tests. This discrepancy stems from the GTX 680's inclusion of a Geekbench Metal score of 7,897, which drags its average down. The GTX 680's OpenCL and Vulkan scores are both notably higher than the GTX 660 Ti's corresponding results, yet the Metal result substantially lowers the overall average. This illustrates why single-metric comparisons can mislead; the per-test breakdown provides the more reliable picture. In every directly comparable test, the GTX 680 outperforms the GTX 660 Ti.

The percentile rankings place both cards in similar territory overall: the GTX 660 Ti sits at the 57th percentile of all GPUs, while the GTX 680 sits at the 55th percentile. The GTX 660 Ti's higher percentile reflects its more consistent benchmark performance without the low Metal outlier. When examining nearest rivals, the GTX 660 Ti's average score of 15,063 places it 0.1% behind the AMD Radeon Pro 560X (15,082), 0.7% behind the AMD Radeon RX 7600 (15,171), 0.9% behind the NVIDIA GeForce RTX 3050 OEM (15,199), and 1.4% behind the AMD Radeon 680M (15,270). These are tight margins, all within 1.5%, indicating the GTX 660 Ti is competitive with a cluster of much newer hardware. The GTX 680's average of 14,150 sits 0.9% ahead of the NVIDIA Tesla K10 (14,029), 0.9% behind the NVIDIA GeForce GTX 1070 Ti (14,277), 1.2% behind the Intel Iris Xe MAX Graphics (14,315), and 1.4% behind the AMD Radeon Vega 11 (14,352).

The deltas between the two cards themselves are more pronounced than the deltas against their nearest rivals. A 10.6% OpenCL gap and a 14.9% Vulkan gap represent meaningful performance stratification within the same generation. The GTX 680 consistently leverages its superior specifications to deliver higher frame rates in compute workloads, while the GTX 660 Ti stays closer to its competitor cluster, trading blows within a narrow band.

The Verdict

From the benchmark data alone, the NVIDIA GeForce GTX 680 is the stronger card. It wins both head-to-head tests, posting double-digit percentage advantages in OpenCL and Vulkan workloads. Anyone prioritizing raw compute performance should choose the GTX 680 without hesitation. The 14.9% Vulkan lead is particularly telling, as Vulkan represents modern graphics API workloads that stress the full pipeline. The GTX 660 Ti's average benchmark score is higher due to the GTX 680's Metal outlier, but that is a statistical artifact rather than a performance advantage.

However, the GTX 660 Ti holds its own in a different sense. Its percentile ranking (57th) exceeds the GTX 680's (55th), and its average score of 15,063 versus 14,150 reflects greater consistency across its tested workloads. The GTX 660 Ti also launches at a lower MSRP of 299 USD compared to the GTX 680's 499 USD — a fact worth noting, though pricing analysis is beyond the scope of this data-driven comparison. For users who value the higher raw performance ceiling and can tolerate the lower average score driven by one benchmark entry, the GTX 680 is the clear pick. For users who want a card that performs consistently across a broader set of tests without the performance trough, the GTX 660 Ti offers the steadier profile.

The data also shows both cards are end-of-life products from the GeForce 600 generation, sharing the same GK104 chip, 28 nm process node, and TSMC foundry. The GTX 680 launched earlier on 2012-03-21, while the GTX 660 Ti followed on 2012-08-15. Both have the same predecessor (GeForce 500) and successor (GeForce 700). The decision ultimately hinges on whether the 10.6% to 14.9% performance advantage matters more than the consistency and percentile placement that favor the GTX 660 Ti.

Architecture Differences

Both cards share the fundamental architecture: the GK104 chip built on the Kepler architecture, manufactured by TSMC on a 28 nm process. Both pack 3,540 million transistors on a 294 mm² die, yielding a transistor density of 12.0 million transistors per square millimeter. The architectural lineage is identical, with both belonging to the GeForce 600 generation and carrying the same predecessor and successor family designations.

The divergence begins with the execution resources. The GTX 680 fields 1,536 shading units, 128 texture mapping units, and 32 raster output units. The GTX 660 Ti scales down to 1,344 shading units, 112 TMUs, and 24 ROPs. That reduction in every major compute block explains the performance differential: fewer shaders mean lower FP32 throughput, fewer TMUs mean lower texture fill, and fewer ROPs mean lower pixel throughput. The GTX 680 runs its shader cores at a 1006 MHz base clock with a 1058 MHz boost, while the GTX 660 Ti operates at 915 MHz base and 980 MHz boost. The combination of more units and higher clocks gives the GTX 680 a substantial theoretical advantage.

Memory configuration differs sharply. The GTX 680 uses a 256-bit memory bus with 192.3 GB/s of bandwidth, while the GTX 660 Ti uses a 192-bit bus with 144.2 GB/s. Both cards carry 2 GB of GDDR5 memory running at 1502 MHz with 6 Gbps effective data rate. The wider bus on the GTX 680 provides 33% more memory bandwidth, which matters in high-resolution textures and compute workloads that saturate memory throughput. Pixel rate and texture rate follow the hardware counts: the GTX 680 delivers 33.86 GPixel/s and 135.4 GTexel/s, versus the GTX 660 Ti's 27.44 GPixel/s and 109.8 GTexel/s. FP32 compute peaks at 3.250 TFLOPS for the GTX 680 and 2.634 TFLOPS for the GTX 660 Ti, a 23% gap.

Power and physical characteristics differ as well. The GTX 680 draws 195 W TDP versus 150 W for the GTX 660 Ti, though both use dual-slot coolers and require two 6-pin power connectors with a suggested 450 W power supply. The GTX 680 is longer at 256 mm (10.1 inches) compared to 241 mm (9.5 inches) for the GTX 660 Ti, and the GTX 680 also has recorded height and width dimensions of 111 mm and 38 mm, respectively, while the GTX 660 Ti lacks those measurements. Display outputs match exactly: 2x DVI, 1x HDMI 1.4a, and 1x DisplayPort 1.2 on both cards. API support is identical — DirectX 12 (11_0), OpenGL 4.6, and Vulkan 1.2.175 — meaning no software feature gap exists between the two.

FAQ

Q: Which card has the higher raw performance in head-to-head tests?

A: The NVIDIA GeForce GTX 680 wins both head-to-head benchmarks. It scores 16,906 in Geekbench OpenCL versus 15,113 for the GTX 660 Ti, and 17,646 in Geekbench Vulkan versus 15,012, representing advantages of 10.6% and 14.9%, respectively.

Q: Why does the GTX 660 Ti have a higher average benchmark score than the GTX 680?

A: The GTX 660 Ti averages 15,063 across two tests, while the GTX 680 averages 14,150 across three tests. The GTX 680's average is dragged down by its Geekbench Metal score of 7,897, which is far lower than its OpenCL and Vulkan results, while the GTX 660 Ti has no Metal entry to dilute its average.

Q: Are these cards from the same architecture generation?

A: Yes. Both use the GK104 chip on the Kepler architecture, manufactured by TSMC on a 28 nm process. Both belong to the GeForce 600 generation, and both pack 3,540 million transistors on a 294 mm² die with a transistor density of 12.0 million transistors per square millimeter.

Q: How do the memory configurations compare?

A: Both cards have 2 GB of GDDR5 memory running at 1502 MHz with 6 Gbps effective data rate. The GTX 680 uses a 256-bit bus with 192.3 GB/s bandwidth, while the GTX 660 Ti uses a 192-bit bus with 144.2 GB/s bandwidth.

Q: What is the percentile ranking for each card against all GPUs?

A: The GTX 660 Ti sits at the 57th percentile of all GPUs, while the GTX 680 sits at the 55th percentile. Despite the GTX 680's higher raw scores in head-to-head tests, the GTX 660 Ti ranks slightly higher overall.

Q: Which card has the higher power draw?

A: The GTX 680 has a TDP of 195 W, while the GTX 660 Ti has a TDP of 150 W. Both require two 6-pin power connectors and suggest a 450 W power supply.

Where Each One Wins

The GTX 680 wins decisively in compute-heavy workloads. Its OpenCL advantage of 10.6% and Vulkan advantage of 14.9% indicate that any application leveraging these APIs will see meaningful gains. The GTX 680's higher shader count (1,536 versus 1,344), higher clock speeds (1006 MHz base versus 915 MHz base), and wider memory bus (256-bit versus 192-bit) all contribute to its superior throughput. The 3.250 TFLOPS FP32 peak versus 2.634 TFLOPS gives it a 23% theoretical compute edge, and the 33.86 GPixel/s pixel rate versus 27.44 GPixel/s suggests better fill-rate-bound performance. The GTX 680 also offers 192.3 GB/s memory bandwidth versus 144.2 GB/s, which helps in bandwidth-sensitive scenarios.

The GTX 660 Ti wins in consistency and efficiency. Its average benchmark score of 15,063 exceeds the GTX 680's 14,150, and it holds a higher percentile ranking (57th versus 55th) despite losing every direct comparison. The GTX 660 Ti achieves this with a lower TDP of 150 W versus 195 W, making it the more power-efficient option. It also occupies a shorter physical footprint at 241 mm versus 256 mm, which matters for compact builds. The GTX 660 Ti's nearest-rival deltas are all within 1.4% of its average score, indicating tight competition with a range of modern GPUs, whereas the GTX 680's nearest-rival deltas span a similar range but from a lower baseline.

For users focused on raw performance in OpenCL and Vulkan workloads, the GTX 680 is the unambiguous choice. For users who prioritize consistent benchmark performance, higher percentile placement, lower power draw, and a more compact card, the GTX 660 Ti offers distinct advantages. Neither card supports ray tracing or tensor cores, as both predate those features. The GTX 680's earlier release date of 2012-03-21 versus the GTX 660 Ti's 2012-08-15 means the latter came to market roughly five months later with scaled-down specifications.

Specification Differences

| Specification | NVIDIA GeForce GTX 660 Ti | NVIDIA GeForce GTX 680 |

|---|---|---|

| Base Clock | 915 MHz | 1006 MHz |

| Boost Clock | 980 MHz | 1058 MHz |

| Memory Bus Width | 192 bit | 256 bit |

| Memory Bandwidth | 144.2 GB/s | 192.3 GB/s |

| Shading Units | 1344 | 1536 |

| TMUs | 112 | 128 |

| ROPs | 24 | 32 |

| Pixel Rate | 27.44 GPixel/s | 33.86 GPixel/s |

| Texture Rate | 109.8 GTexel/s | 135.4 GTexel/s |

| FP32 Performance | 2.634 TFLOPS | 3.250 TFLOPS |

| TDP | 150 W | 195 W |

| Length | 241 mm (9.5 inches) | 256 mm (10.1 inches) |

| Height | Not specified | 111 mm (4.4 inches) |

| Width | Not specified | 38 mm (1.5 inches) |

| Release Date | 2012-08-15 | 2012-03-21 |

| Launch MSRP | 299 USD | 499 USD |

Both cards share identical specifications for memory size (2 GB), memory type (GDDR5), memory clock (1502 MHz, 6 Gbps effective), chip (GK104), architecture (Kepler), process node (28 nm), foundry (TSMC), transistors (3,540 million), die size (294 mm²), transistor density (12.0M / mm²), slot width (Dual-slot), power connectors (2x 6-pin), suggested PSU (450 W), bus interface (PCIe 3.0 x16), display outputs (2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2), API support (DirectX 12 (11_0), OpenGL 4.6, Vulkan 1.2.175), production status (End-of-life), predecessor (GeForce 500), and successor (GeForce 700). The GTX 680 carries additional dimension measurements for height and width that the GTX 660 Ti does not report.