NVIDIA GeForce GTX 660

NVIDIA GeForce GTX 660
NVIDIA GeForce GTX 660 graphics card review

NVIDIA GeForce GTX 660: Tests of Different Versions, FPS in Games, and Whether It's Worth Buying Today

The NVIDIA GeForce GTX 660 was released in 2012 as a mid-range graphics card for gaming in Full HD. The retail version featured the Kepler architecture, 960 CUDA cores, and 2 GB of GDDR5 memory. Today, the GTX 660 is mainly interesting as a cheap card for older games: the performance of the GK106 is still sufficient for many projects from the DirectX 9-11 era, but the 2 GB of video memory and discontinued driver support severely limit its performance in newer games.

Results of Specific GeForce GTX 660 Models

The GTX 660 was produced by ASUS, MSI, EVGA, Gigabyte, Point of View, and other manufacturers. Most models differed in factory overclocking and cooling systems, but the performance difference between them was usually minimal.

Below are the results of four non-reference GTX 660 models tested in the same system.

Graphics Card GPU Base / Boost 3DMark 11
ASUS GTX660-DC2T-2GD5 DirectCU II TOP 1072 / 1137 MHz 7082
EVGA GeForce GTX 660 SC 2GB 1046 / 1111 MHz 6807
Point of View GTX 660 Ultra Charged 1084 / 1163 MHz 6645
MSI N660 TF 2GD5/OC Twin Frozr 1033 / 1098 MHz 6632

The ASUS DirectCU II TOP turned out to be faster than the other test participants. Meanwhile, the Point of View Ultra Charged had a higher claimed Boost frequency, but it did not provide any advantage over the ASUS card.

The gap between the four cards is less than 7%. On the secondary market, the condition of the cooling system and the board itself is now significantly more important than a minor factory overclock.

GTX 660 and GTX 660 OEM - Not the Same Card

Under the GeForce GTX 660 name, NVIDIA released two significantly different versions.

Parameter GTX 660 GTX 660 OEM
CUDA Cores 960 1152
Base Frequency 980 MHz 823 MHz
Boost 1033 MHz 888 MHz
Video Memory 2 GB GDDR5 1.5 or 3 GB GDDR5

The OEM modification used a different GPU configuration and was mainly found in pre-built computers. Therefore, the results of the GTX 660 OEM should not be mixed in the same sample with the standard retail GTX 660 solely based on the adapter name.

When adding old results to the database, it is advisable to check the number of CUDA cores, clock speeds, and memory size of the specific card.

GTX 660 Performance in Games

Game tests clearly show the initial positioning of the GTX 660. In projects of its time, the card was primarily aimed at 1920×1080 resolutions with high or maximum settings.

Game Settings GTX 660
Battlefield 3 1080p, Ultra ~42-45 FPS
The Witcher 2 1080p, Ultra ~48-51 FPS
Sleeping Dogs 1080p, Ultra ~53-55 FPS

For games from the early 2010s, this is still an acceptable level. Many projects from that period can be run in Full HD without dropping to minimum settings.

With modern games, the situation is much worse. The main limitation is the 2 GB VRAM: heavy textures quickly fill up the video memory, causing drops in performance and data loading issues, while some new games either run poorly or aren't designed for such an old architecture at all.

Power and System Requirements

The reference GTX 660 has a rated power consumption of about 140 W and uses one six-pin power connector. NVIDIA recommended a system power supply of at least 450 W.

For an old office PC, however, the more important factor is that the GTX 660 requires a separate 6-pin power supply, which means it cannot be installed into just any prebuilt system.

Driver Support Has Ended

The GTX 660 belongs to the Kepler generation of desktop graphics cards. NVIDIA stopped releasing full Game Ready Drivers for such models in 2021, and the period for subsequent security updates has also concluded.

For older games, this is not critical: existing drivers are already well-optimized for projects from that time. However, there will be no fixes for new releases, optimization of modern game engines, or support for new features for the GTX 660.

Buying the GTX 660 on the Secondary Market

The age of a specific unit is now more important than whether it was released by ASUS, MSI, or EVGA. Most remaining GTX 660 cards are already over ten years old.

Before buying, it's worth checking:

  • Stable performance under load without artifacts;
  • GPU temperature;
  • Condition and noise of the fans;
  • Functionality of video outputs;
  • Condition of the six-pin power connector.

Dried thermal paste and worn-out fans are common for cards of this age. Replacing the thermal paste may help with temperatures, but it does not resolve issues related to memory degradation, power delivery, or the GPU itself.

Overpaying for a supposed GTX 660 TOP or SuperClocked just for factory overclocking is nearly pointless. It’s better to choose a unit that operates stably and has functional cooling.

GTX 660 Today: Only for Old Games and Budget Builds

The GTX 660 should only be considered at a very low price. For older games and as a temporary PC solution, its performance is still sufficient, but the 2 GB of memory and lack of current drivers virtually exclude it from modern gaming builds.

When choosing between several GTX 660 units, it's better to get the most reliable one instead of overpaying for factory overclocking or a more expensive version from ASUS, MSI, or EVGA.

Basic

Label Name
NVIDIA
Platform
Desktop
Launch Date
September 2012
Model Name
GeForce GTX 660
Generation
GeForce 600
Base Clock
980MHz
Boost Clock
1032MHz
Bus Interface
PCIe 3.0 x16
Transistors
2,540 million
TMUs
?
Texture Mapping Units (TMUs) serve as components of the GPU, which are capable of rotating, scaling, and distorting binary images, and then placing them as textures onto any plane of a given 3D model. This process is called texture mapping.
80
Foundry
TSMC
Process Size
28 nm
Architecture
Kepler

Memory Specifications

Memory Size
2GB
Memory Type
GDDR5
Memory Bus
?
The memory bus width refers to the number of bits of data that the video memory can transfer within a single clock cycle. The larger the bus width, the greater the amount of data that can be transmitted instantaneously, making it one of the crucial parameters of video memory. The memory bandwidth is calculated as: Memory Bandwidth = Memory Frequency x Memory Bus Width / 8. Therefore, when the memory frequencies are similar, the memory bus width will determine the size of the memory bandwidth.
192bit
Memory Clock
1502MHz
Bandwidth
?
Memory bandwidth refers to the data transfer rate between the graphics chip and the video memory. It is measured in bytes per second, and the formula to calculate it is: memory bandwidth = working frequency × memory bus width / 8 bits.
144.2 GB/s

Display and Media

Outputs
2x DVI
1x HDMI 1.4a
1x DisplayPort 1.2

Theoretical Performance

Pixel Rate
?
Pixel fill rate refers to the number of pixels a graphics processing unit (GPU) can render per second, measured in MPixels/s (million pixels per second) or GPixels/s (billion pixels per second). It is the most commonly used metric to evaluate the pixel processing performance of a graphics card.
20.64 GPixel/s
Texture Rate
?
Texture fill rate refers to the number of texture map elements (texels) that a GPU can map to pixels in a single second.
82.56 GTexel/s
FP64 (double)
?
An important metric for measuring GPU performance is floating-point computing capability. Double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy, while single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable.
82.56 GFLOPS
FP32 (float)
?
An important metric for measuring GPU performance is floating-point computing capability. Single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks, while double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable.
2.021 TFLOPS

Miscellaneous

Shading Units
?
The most fundamental processing unit is the Streaming Processor (SP), where specific instructions and tasks are executed. GPUs perform parallel computing, which means multiple SPs work simultaneously to process tasks.
960
L1 Cache
16 KB (per SMX)
L2 Cache
384KB
TDP
140W
Vulkan Version
?
Vulkan is a cross-platform graphics and compute API by Khronos Group, offering high performance and low CPU overhead. It lets developers control the GPU directly, reduces rendering overhead, and supports multi-threading and multi-core processors.
1.1
OpenCL Version
3.0
OpenGL
4.6
CUDA
3.0
DirectX
12 (11_0)
Power Connectors
1x 6-pin
ROPs
?
The Raster Operations Pipeline (ROPs) is primarily responsible for handling lighting and reflection calculations in games, as well as managing effects like anti-aliasing (AA), high resolution, smoke, and fire. The more demanding the anti-aliasing and lighting effects in a game, the higher the performance requirements for the ROPs; otherwise, it may result in a sharp drop in frame rate.
24
Shader Model
5.1
Suggested PSU
300W

Benchmarks

FP32 (float)
Score
2.021 TFLOPS
3DMark Time Spy
Score
1285
Blender
Score
126
OctaneBench
Score
28
Vulkan
Score
11719
OpenCL
Score
11135
Hashcat
Score
25551 H/s

Compared to Other GPU

FP32 (float) / TFLOPS
2.157 +6.7%
2.099 +3.9%
3DMark Time Spy
4682 +264.4%
3619 +181.6%
2329 +81.2%
1526 +18.8%
Blender
1408.56 +1017.9%
802 +536.5%
391 +210.3%
191.62 +52.1%
OctaneBench
288 +928.6%
123 +339.3%
69 +146.4%
Vulkan
83205 +610%
55601 +374.5%
34493 +194.3%
16654 +42.1%
OpenCL
53439 +379.9%
34541 +210.2%
18130 +62.8%
884 -92.1%
Hashcat / H/s
33607 +31.5%
31509 +23.3%
24493 -4.1%
23908 -6.4%