NVIDIA GeForce GTX 660 vs NVIDIA TITAN V CEO Edition Comparison

# NVIDIA TITAN V CEO Edition vs NVIDIA GeForce GTX 660

The NVIDIA TITAN V CEO Edition and the NVIDIA GeForce GTX 660 represent two vastly different eras of GPU design, yet their average benchmark scores place them almost exactly level. The TITAN V CEO Edition posts an average benchmark score of 9037, while the GTX 660 scores 9022 — a difference of just 0.2% in favor of the TITAN V. Both cards sit at the 45th percentile among all GPUs, meaning they occupy the same middle tier of the performance distribution. This near-equality in aggregate scoring is striking given the six years separating their releases, and the data invites a closer look at what each card actually does well.

Head-to-Head Benchmarks

The head-to-head benchmark data is empty, so the comparison must rely on each card's individual benchmark results and their relative positions. The TITAN V CEO Edition delivers a Passmark G3D score of 16988, which is a strong showing for a compute-oriented card. The GTX 660, by contrast, has no Passmark G3D entry in the data, but it does have Geekbench scores across three APIs: Metal at 4305, OpenCL at 11347, and Vulkan at 11415. These Geekbench results show the GTX 660 achieving roughly 2.6 times higher OpenCL scores than Metal scores, and Vulkan scores slightly above OpenCL.

The TITAN V CEO Edition's Passmark G2D score of 1086 suggests modest 2D performance relative to its 3D capability. The GTX 660's nearest rival data shows it trailing the TITAN V by 0.2% on average score, while the TITAN V's nearest rivals include the GTX 660 at a 0.2% delta, the GTX 560 at -0.2%, and the AMD Radeon 890M at -1.9%. The fact that a 2018 flagship-class card and a 2012 mid-range card land within 0.2% of each other on average benchmarks suggests that the aggregate metric masks enormous differences in raw compute capability.

The TITAN V CEO Edition's FP32 throughput is 14.90 TFLOPS, while the GTX 660 manages 1.981 TFLOPS — a 7.5-fold gap. Similarly, texture rate differs by a factor of 5.6 (465.6 GTexel/s versus 82.56 GTexel/s), and pixel rate by a factor of 9.0 (186.2 GPixel/s versus 20.64 GPixel/s). Yet the average benchmark scores barely move. This implies the aggregate benchmark methodology heavily weights factors other than pure compute throughput, or that the GTX 660's driver optimizations and lighter feature set allow it to punch above its raw specifications.

Architecture Differences

The architecture gap between these two cards is generational. The TITAN V CEO Edition uses the GV100 chip on the Volta architecture, built on a 12 nm process at TSMC. The GTX 660 uses the GK106 chip on Kepler, also at TSMC but on a 28 nm process. Transistor counts tell the story: the TITAN V packs 21,100 million transistors on an 815 mm² die, yielding a transistor density of 25.9 million per square millimeter. The GTX 660 has 2,540 million transistors on a 221 mm² die, with a density of 11.5 million per square millimeter. The TITAN V thus has roughly 8.3 times more transistors and a density advantage of 2.25 times.

Memory architecture differs fundamentally. The TITAN V uses 32 GB of HBM2 on a 4096-bit bus, delivering 868.4 GB/s bandwidth. The GTX 660 uses 2 GB of GDDR5 on a 192-bit bus, achieving 144.2 GB/s. The TITAN V's bandwidth advantage is 6.0 times, and its memory capacity is 16 times larger. The TITAN V also features 640 tensor cores, which the GTX 660 lacks entirely — a capability designed for deep learning workloads that simply did not exist in the Kepler generation.

Shading resources diverge sharply: the TITAN V has 5120 shading units, 320 TMUs, and 128 ROPs, versus the GTX 660's 960 shading units, 80 TMUs, and 24 ROPs. The TITAN V supports FP16 at 29.80 TFLOPS (2:1 ratio), while the GTX 660 has no FP16 capability listed. Both cards support DirectX 12, but the TITAN V achieves 12_1 while the GTX 660 is limited to 11_0. Vulkan support also differs: the TITAN V runs Vulkan 1.4, while the GTX 660 runs 1.2.175.

FAQ

Q: Which card has the higher average benchmark score?

A: The NVIDIA TITAN V CEO Edition scores 9037 on average, while the NVIDIA GeForce GTX 660 scores 9022. The difference is 0.2% in favor of the TITAN V.

Q: How do the memory bandwidths compare?

A: The TITAN V CEO Edition delivers 868.4 GB/s from 32 GB of HBM2 on a 4096-bit bus. The GTX 660 provides 144.2 GB/s from 2 GB of GDDR5 on a 192-bit bus. The TITAN V has 6.0 times the bandwidth and 16 times the capacity.

Q: What is the transistor density difference?

A: The TITAN V CEO Edition has a density of 25.9 million transistors per square millimeter on a 12 nm process, while the GTX 660 has 11.5 million per square millimeter on a 28 nm process. That is a 2.25 times density advantage for the newer card.

Q: Does the GTX 660 support tensor cores?

A: No. The GTX 660 has no tensor cores listed, while the TITAN V CEO Edition includes 640 tensor cores. This reflects the Volta architecture's focus on AI and compute workloads.

Q: What are the pixel rate figures for each card?

A: The TITAN V CEO Edition achieves 186.2 GPixel/s, while the GTX 660 achieves 20.64 GPixel/s. The TITAN V is 9.0 times faster in pixel throughput.

Q: How do their DirectX versions differ?

A: The TITAN V CEO Edition supports DirectX 12 (12_1), while the GTX 660 supports DirectX 12 (11_0). Both are DirectX 12 compatible, but the TITAN V exposes a higher feature level.

Specification Differences

| Specification | NVIDIA TITAN V CEO Edition | NVIDIA GeForce GTX 660 |

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

| Chip | GV100 | GK106 |

| Architecture | Volta | Kepler |

| Generation | GeForce 10 | GeForce 600 |

| Process Node | 12 nm | 28 nm |

| Transistors | 21,100 million | 2,540 million |

| Die Size | 815 mm² | 221 mm² |

| Transistor Density | 25.9M / mm² | 11.5M / mm² |

| Base Clock | 1200 MHz | 980 MHz |

| Boost Clock | 1455 MHz | 1032 MHz |

| Memory Clock | 848 MHz (1696 Mbps effective) | 1502 MHz (6 Gbps effective) |

| Memory Size | 32 GB | 2 GB |

| Memory Type | HBM2 | GDDR5 |

| Memory Bus Width | 4096 bit | 192 bit |

| Memory Bandwidth | 868.4 GB/s | 144.2 GB/s |

| Shading Units | 5120 | 960 |

| TMUs | 320 | 80 |

| ROPs | 128 | 24 |

| Tensor Cores | 640 | None |

| Pixel Rate | 186.2 GPixel/s | 20.64 GPixel/s |

| Texture Rate | 465.6 GTexel/s | 82.56 GTexel/s |

| FP32 | 14.90 TFLOPS | 1.981 TFLOPS |

| FP16 | 29.80 TFLOPS (2:1) | None |

| TDP | 250 W | 140 W |

| Power Connectors | 1x 6-pin + 1x 8-pin | 1x 6-pin |

| Suggested PSU | 600 W | 300 W |

| Display Outputs | 1x HDMI 2.0, 3x DisplayPort 1.4a | 2x DVI, 1x HDMI 1.4a, 1x DisplayPort 1.2 |

| DirectX | 12 (12_1) | 12 (11_0) |

| Vulkan | 1.4 | 1.2.175 |

| Length | 267 mm (10.5 inches) | 241 mm (9.5 inches) |

| Release Date | 2018-06-20 | 2012-09-05 |

| Launch MSRP | None listed | 229 USD |

Where Each One Wins

The TITAN V CEO Edition wins decisively in every raw compute metric in the data. Its FP32 throughput of 14.90 TFLOPS dwarfs the GTX 660's 1.981 TFLOPS, making it 7.5 times faster for general shader work. Texture-heavy workloads see a 5.6 times advantage (465.6 GTexel/s versus 82.56 GTexel/s), and pixel-fill tasks run 9.0 times faster. The 640 tensor cores give the TITAN V a capability the GTX 660 cannot match at all, positioning it for machine learning inference and training workloads. Memory bandwidth of 868.4 GB/s versus 144.2 GB/s means the TITAN V can feed its compute units far more effectively, particularly for large datasets that exceed the GTX 660's 2 GB capacity.

The GTX 660 wins in power efficiency per frame based on the data. It draws 140 W versus the TITAN V's 250 W, and its suggested PSU requirement is 300 W versus 600 W. For a system with a modest power supply or constrained thermal envelope, the GTX 660 is the more practical choice. Its smaller physical footprint — 241 mm versus 267 mm length — also makes it easier to fit in compact cases. The GTX 660's memory clock runs at 1502 MHz (6 Gbps effective) versus the TITAN V's 848 MHz (1696 Mbps effective), though the vastly wider bus of the TITAN V more than compensates. The GTX 660 also offers dual DVI outputs, which some legacy display setups may still require.

The Verdict

The data presents a fascinating contradiction: two cards with nearly identical average benchmark scores (9037 versus 9022) and identical percentile rankings (45th) that could not be more different in architecture and intent. The TITAN V CEO Edition is a compute monster from 2018, built on a 12 nm process with 5120 shading units, 640 tensor cores, and 32 GB of HBM2. The GTX 660 is a 2012 mid-range card on 28 nm with 960 shading units and 2 GB of GDDR5. Yet the aggregate benchmark numbers suggest they perform similarly in the specific workloads those benchmarks measure.

For users who need raw compute power, FP32 throughput, tensor core acceleration, or massive memory bandwidth, the TITAN V CEO Edition is the obvious pick. Its 14.90 TFLOPS FP32 and 868.4 GB/s bandwidth are in a different category entirely from the GTX 660's 1.981 TFLOPS and 144.2 GB/s. The GTX 660, however, offers a lower power draw at 140 W, a smaller physical footprint, and a 229 USD launch MSRP that reflects its original mid-range positioning. The GTX 660 also supports DirectX 12 (11_0) and Vulkan 1.2.175, so it remains functional for modern APIs.

The verdict depends entirely on workload. Benchmark aggregate scores suggest these cards are peers in some averaged sense, but the underlying specifications tell a different story. The TITAN V CEO Edition is for compute-heavy tasks and high-resolution texturing; the GTX 660 is for lightweight, power-conscious builds where its 2 GB memory and 192-bit bus are sufficient. The data does not support a single "better" card — it supports two different tools for two different jobs.