Home / PC H/W Reviews / Core PC Components / Desktop GPU Reviews / Asus GTX 660 Direct CU II TOP 2GB GDDR5 GPU Review
Asus GTX660 Boxshot 1

Asus GTX 660 Direct CU II TOP 2GB GDDR5 GPU Review

  1. Introduction
  2. Packaging and Specification
  3. Closer Look
  4. Test bench and testing methodology
  5. Futuremark Benchmarks
  6. Cinebench OpenGL Benchmarks
  7. Catzilla BETA ALLBenchmark
  8. Unigine Heavens DX11 Benchmark
  9. Battlefield 3 (Episode Operation: Swordfish) Benchmark
  10. Dirt 3 DX 11 Benchmark
  11. Passion Leads Army (Unreal 3 Engine) Benchmark
  12. Resident Evil 5 DX10 Benchmark
  13. Sniper Elite V2 DX11 Benchmark
  14. Sleeping Dogs DX 11 Benchmark
  15. Overclocked Performance Comparison
  16. Conclusion
  17. View All

Nvidia’s GTX 660 uses Nvidia GK106 Kepler SMX architecture made with 28nm fabrication process manufactured by TSMC. It was first introduced on GTX 680.


In GTX 660, there are 3 processing clusters with 5 SMC units and an L2 Cache shared between them all. There are 192 CUDA cores per unit, 80 texture units and 24 ROPs. The reference version of the card core clock of 980 MHz and a memory clock of 1502 MHz with a maximum capacity of 2GB GDDR5. It takes power from a single 6-pin power connector and has TDP of 140w.

The card which I have is from Asus. A non-reference version.


Direct CUII TOP versions usually have a similar illustration on the box. They could work on putting 1-2 add-ons though.


The sample came with (I think) pretty much everything compared to a retail pack- except the card was gift wrapped using bubble wraps. I wouldn’t be surprised if you get the anti static bag by default. You get a DVI-to-VGA converter so it should be used if anyone requires them for any reason.


It would be nice if non-reference versions were bundled with a game just like in the old days. Also, unlike GTX 650 Ti Direct CU II that I’ve reviewed before, the packaging of this GTX 660 variant doesn’t use Polyethylene Foam protection, rather cardboard box is used to the card in place. But you are getting the driver utility with GPU Tweak software.


The non-reference GTX 660 comes with the GPU core clocked at 1072 MHz and 1527 MHz respectively.

GeForce GTX 660 (Reference) ASUS GTX 660-DC2T-2GD5
Shader Units 960 960
ROPs 24 24
Graphics Processor GK106 GK106
Transistors 2540M 2540M
Memory Size 2048 MB 2048 MB
Memory Bus Width 192 bit 192 bit
Core Clock 980 MHz 1072 MHz
Memory Clock 1502 MHz 1527 MHz

The key difference in specs is the core clock and memory clock. Rest of it on paper is the same.

The PCB of the gtx 660 GPU is 4.5” x (a little under) 10” and looking at the dimensions including the cooler is 5.5” x 10.7”. This card takes up 2 slots.



As a speciality of Direct CU II series GPUs, the card comes with dual fans and 2x heatpipe with direct heatpipe contact on the GPU core. As Asus claims, it helps to keep the card 20% cooler and vastly quieter.


From left: 1x DVI-I /Compatible connector to DVI-to-VGA converter with a DVI-D port below it, followed by HDMI and Display Port with a small vent below it.


The cooler is pretty easy to remove once you remove the 4x spring screws.


A generous amount of thermal paste is on the direct heatpipe as well as a small strip of the thermal pad for 2 of the memory chips.


Just like the reference version, this card requires a 6 pin PCIe connector to power up the card. The PCB has Asus’ Digi+ VRM, with the array of Asus’ Super Alloy chokes and capacitors. There are a total of 8x Samsung K46G20325FD-FC03 chips (4x on each side) on both sides of the PCB.


4x Samsung chips surround the Solid Cap that is soldered on the side of the PCB, just behind the core.

It gives me great pleasure to say that I get hardware support from the manufacturers to review not only their own products but others on behalf of the readers. Its some companies such as Asus, Gigabyte, Kingston, Western Digital and Coolermaster who give me hardware support by updating my test rig no strings attached and that helps me to help you! Thumbs up to these guys!

I would like to thank:

  • Gigabyte India for providing Gigabyte 890GPA UD3H Rev 1.0 motherboard
  • Asus India for providing Asus 990FX Sabertooth motherboard
  • Kingston Taiwan for providing hardware support with memory kits and SSD drive.
  • WD India for providing WD 300GB HLFS Velociraptor Hard Drive.
  • Coolermaster India for providing Coolermaster GX450 RS-450-ACAA-D3 Power Supply
Test Setup for: Asus GTX 660 Direct CU II TOP 2GB GDDR5
Motherboard+ Processor Gigabyte 890GPAUD3H Rev 1.0+ AMD 965BE
Memory Kingston KHX1600C9D3P1K28G HyperX Genesis 8GB 1600MHz DDRIII
Primary OS drive WD 3000HLFS Velociraptor 300GB
Power Supply Corsair TX750
Chassis Open Test Bench
LCD Panel Dell E228WFP 22 Inch 16:10 (1680x 1050 Max. Resolution)
Operating System Windows 7 64 bit Ultimate SP1
GPU Driver Used Nvidia 310.70 Drivers

Every Benchmark (Except Battlefield 3) has been done by running the tests 3 times. Scores that are posted are of 3rd run. The drive is formatted and OS + GPU Driver via Asus’ website installed. PhysX is disabled throughout the test.

3DMark 11 Performance Preset: Tested with standard setting preset by FutureMark with 1280x 720 Resolution+ Moderate Load

3DMark Vantage GPU benchmark:

Performance Preset High Preset
Screen Resolution: 1280×1024 1680×1050
Multisample Count: 1 2
Multisample Quality: 0 1
Texture Filtering: Trilinear Anisotropic
Max Anisotropy: N/A 8
Texture Quality: Performance High
Shadow Shader Quality: Performance High
Shadow Resolution Quality: Performance High
POM Shader Quality: Performance High
Volumetric Rendering Quality: Performance High
Post-Processing Scale: 1:2 1:2

Cinebench Benchmarks

Cinebench 10- OpenGL Score

Cinebench 11.5- OpenGL Score

Unigine Heaven 3.0 DX11 Benchmark

Preset 1: Tessellation: Disabled; Shaders: Low; Anisotropy: Off AA: off

Preset 2: Tessellation: Extreme; Shaders: High; Anisotropy: 16x AA: 8X

Preset 3: Tessellation: Normal; Shaders: High Anisotropy: 4x AA: 8x

Battlefield 3 (Chapter: Operation Swordfish)

Preset 1: 1600 x 900 Resolution/ Preset: Ultra

Preset 2: 1680 x 1050 Resolution/Preset: Ultra

Preset 3: 1680 x 1050 Resolution/Preset: High

Preset 4: 1680 x 1050 Resolution/Preset: Medium

Benchmarked using Fraps

Sleeping Dogs DX 11 Benchmark

Preset 1: Quality AA: Normal; HD Textures: Off; Shadow Resolution: Off; Shadow Filtering: Normal; SSAO: Normal; Motion Blur: Off; World Density: Extreme

Preset 2: Quality AA: High; HD Textures: Off; Shadow Resolution: Normal; Shadow Filtering: Normal; SSAO: Normal; Motional Blur: Off; World Density: Extreme

Preset 3: Quality AA: High; HD Textures: On; Shadow Resolution: High; Shadow Filtering: High; SSAO: High; Motion Blur: Off; World Density: Extreme

Preset 4: Quality AA: Extreme; HD Textures: On; Shadow Resolution: High; Shadow Filtering: High; SSAO: High; Motion Blur: Off; World Density: Extreme

Dirt 3 DX11 Benchmark

Benchmarks are done with Low, Medium, High and Ultra in-game presets using 1680 x 1050

Passion Leads Army (Unreal 3 Engine) Benchmark

DX9 Low-Performance Preset

DX9 Medium Performance Preset

DX11 High-Performance Preset

Metro 2033 Benchmark

Preset 1: DX11; Resolution: 1680 x 1050; Quality: Medium; AA: MSAA4X; Texture Filtering: AF 4X; DOF: Off
Preset 2: DX11; Resolution: 1280 x 1024; Quality: Medium; AA: MSAA4X; Texture Filtering: AF 4X; DOF: Off
Preset 3: DX11; Resolution: 1680 x 1050; Quality: Very High; AA: MSAA4X; Texture Filtering: AF 16X; DOF: On

Residential Evil 5 Benchmark

Preset 1- Resolution: 1680x 1050; AA: 2x; Motion Blur: On; Shadow DETAIL: High ; Texture Detail: High ; Overall Quality: High

Preset 2– Resolution: 1680x 1050; AA: c16x; Motion Blur: On; Shadow DETAIL: High ; Texture Detail: High ; Overall Quality: High

Sniper Elite V2 DX11 Benchmark

Preset 1- Texture Detail: Low; Shadow Detail: Low; AA: Low; Draw Distance: Low Anisotropic Filtering= off ; Motion Blur: On; Shader: On ; Advance Shadows= off; Super Sampling= 0

Preset 2- Texture Detail: high; Shadow Detail: high; AA: med; Draw Distance: high Anisotropic Filtering= off ; Motion Blur: On; Shader: On ; Advance Shadows= high; Super Sampling= 2.25x

Preset 3- Texture Detail: ultra; Shadow Detail: ultra; AA: med; Draw Distance: ultra Anisotropic Filtering= 16 ; Motion Blur: On; Shader: On ; Advance Shadows= high; Super Sampling= 4x

Overclock Gaming Performance Comparison Test

AllBenchmark Beta Benchmark

Preset “Cat” with 720p

Preset “Tiger” with 1080p

Preset “Catzilla” with 1440

Just like Anvil Benchmark for storage drives, I won’t be using Catzilla for comparing with other cards as it’s in BETA. Still, it’s an interesting test that I’ve grown fond of and I would like to run this on a regular basis. Catzilla ALLBenchmark is a test to check your computer’s graphics capabilities to see if it’s suited for latest games in best quality with the resolution preset and custom settings. What I really like about this test is that even due to lack of resources I can run tests without being limited to my screen’s resolution. Unlike 3DMark Benchmarks, I can not only run 1080p tests but also on 2560 x 1440 resolution preset.

3D Mark Vantage: Performance and High Preset

3D Mark 11: Performance Preset

Sleeping Dogs

Metro 2033

3DMark Vantage is the industry standard performance benchmark for DirectX 10 gaming PCs. It includes two graphics tests, two CPU tests and six feature tests. Trusted by gamers worldwide to give accurate and unbiased results, 3DMark Vantage is the best way to consistently and reliably test DirectX 10 under game-like loads.

With GPU Clock of 1072 MHz with a Memory clock of 1527 MHz, the non-reference Asus GTX 660 Direct CU II 2GB GDDR5 stays above GTX 650 Ti and HD7850, both Direct CU II models with either both performance and high preset.

3DMark 11 is the latest version of the world’s most popular benchmark. Designed to measure your PC’s gaming performance 3DMark 11 makes extensive use of all the new features in DirectX 11 including tessellation, compute shaders and multi-threading. Trusted by gamers worldwide to give accurate and unbiased results, 3DMark 11 is the best way to consistently and reliably test DirectX 11 under game-like loads.

Same as Vantage.

Cinebench 10 OpenGL Benchmark

The MAXON CINEBENCH R10 uses a wide range of algorithms and instructions which ensure that the tests performed to deliver a very good overview of a system’s overall performance capabilities, also with regard to other CPU and graphics card intensive tasks. MAXON CINEBENCH R10 supports up to 16 CPUs or CPU cores and runs under Windows (2000, XP, Vista in 32-bit or 64-bit) or Mac OS X (with PowerPC and Intel processors).

It’s a real-world test suite that assesses your computer’s performance capabilities. MAXON CINEBENCH is based on MAXON’s award-winning animation software CINEMA 4D, which is used extensively by studios and production houses worldwide for 3D content creation. MAXON software has been used in blockbuster movies such and Spiderman, Star Wars, The Chronicles of Narnia and many more.

MAXON CINEBENCH runs several tests on your computer to measure the performance of the main processor and the graphics card under real world circumstances. The benchmark application makes use of up to 16 CPUs or CPU cores and is available for Windows (32- and 64-Bit) and Macintosh (PPC- and Intel-based).

The resulting values among different Operating Systems are 100% comparable and therefore very useful with regard to purchasing decision-making. It can also be used as a marketing tool for hardware vendors or simply to compare hardware among colleagues or friends.


Cinebench 11.5 OpenGL Benchmark

This procedure uses a complex 3D scene depicting a car chase which measures the performance of your graphics card in OpenGL mode. The performance depends on various factors, such as the GPU processor on your hardware, but also on the drivers used. The graphics card has to display a huge amount of geometry (nearly 1 million polygons) and textures, as well as a variety of effects, such as environments, bump maps, transparency, lighting and more to evaluate the performance across different disciplines and give a good average overview of the capabilities of your graphics hardware. The result given is measured in frames per second (fps). The higher the number, the faster your graphics card.


When it comes to OpenGL rendering, AMD Radeon’s 7850 has an advantage over these cards.

It’s a computer test focused to check your computer graphics capabilities. The test can analyze if your computer is suitable for latest games or watching HD movies in best quality. It can also suggest what hardware modules of your computer should be replaced in order to get the best performance.

The benchmark will also check software configuration of your system and will suggest improvements that may cause your computer to work faster and more efficiently.

There are four tests available:

KITTY – 576p (1024 x 576) – for testing netbooks and ultrabooks, requires 256 MB of GPU memory.
CAT – 720p (1280 x 720) – for testing the low end desktop computers, requires 512 MB of GPU memory
TIGER – 1080p (1920 x 1080) – for current desktop computers, requires 1024 MB of GPU memory
CATZILLA – 1440p (2560 x 1440) – for most demanding high-end computers, requires more than 1024 MB of GPU memory
CUSTOM based on your own options and features settings



  • Heavy GPU load, extreme hardware stability test.
  • Support of DirectX 9, DirectX 10, DirectX 11 and OpenGL 4.0.
  • Comprehensive use of hardware tessellation technology (3 presets).
  • Advanced ambient occlusion.
  • Dynamic global illumination.
  • Volumetric cumulonimbus clouds of high physical fidelity.
  • Simulation of day-night shift and changing light conditions.
  • Dynamic sky with light scattering.
  • Interactive experience with fly/walk-through modes.
  • Support of NVIDIA SLI and ATI CrossFire technologies.
  • Stereo 3D modes:
    • Anaglyph
    • Separate images
    • 3D Vision
    • 3D Surround
    • iZ3D
  • Support of multi-monitor configurations.
  • Based on the latest, most advanced version version of powerful UNIGINE engine.


The card performs pretty well against Direct CUII version of 7850 with the first preset where the settings are low, but will normal settings on Tessellation and shaders, with 4x Anisotropy and Anti-aliasing of 8x, it performs on same par with the non-reference 7850.

Battlefield 3 is a first-person shooter video game developed by EA Digital Illusions CE and published by Electronic Arts. It is a direct sequel to 2005’s Battlefield 2, and the twelfth installment in the Battlefield franchise. It is the first game in the series that does not support versions of Windows prior to Windows Vista as the game only supports DirectX 10 and 11.Battlefield 3 debuts the new Frostbite 2 engine. This updated Frostbite engine can realistically portray the destruction of buildings and scenery to a greater extent than previous versions. Unlike previous iterations, the new version can also support dense urban areas.

Battlefield 3 uses a new type of character animation technology called ANT. ANT technology is used in EA Sports games, such as FIFA, but for Battlefield 3 is adapted to create a more realistic soldier, with the ability to transition into cover and turn the head before the body, as well as “drag fallen comrades into safety and mount weapons on almost any part of the terrain”.


GTX 650Ti has a small advantage on lower resolution compared to GTX 660, even with the in-game Ultra setting. But when it comes to at least 1680 x 1050 with any presets, the card has a significant lead over the 2 cards that were evaluated before.

Working with AMD, the developers of Dirt 3 have continued their support for DirectX® 11 with the inclusion of contact hardening shadows through Shader Model 5.0, improved high definition ambient occlusion, and stunning tessellation support with realistic water, snow, and cloth.


With a resolution of 1680 x 1050, just like in BF3 benchmark, the card gives better performance over the other graphics cards.

Giant Iron Horse’s shooter is one of the most graphically advanced Chinese titles to date, in part thanks to the use of Epic Games’ Unreal Engine 3. Featuring a range of DirectX 11 effects, such as tessellation and Bokeh depth of field, PLA also makes extensive use of NVIDIA’s GPU-accelerated PhysX technology to generate realistic, physics-led effects. To help promote PLA, and to get gamers ready for its launch later this year, Giant Iron Horse has released the first-ever Chinese-developed DirectX 11 benchmark featuring the aforementioned range of effects.


With Low and Medium Performance preset with DX 9, Asus’ non-reference HD7850 churns more FPS But with high preset with DX11 with 1680 x 1050 resolution, the tables have turned, giving GTX 660 has an advantage.

Resident Evil 5 is a survival horror video game developed and published by Capcom using MT Framework v1.4 Havok engine


Sniper Elite V2 is a 2012 tactical shooter video game developed by Rebellion Developments using Asura Engine.


With low preset, GTX 660 has a higher advantage, but with 2nd and 3rd presets, it’s not the same par as HD7850.


I didn’t run this benchmark with 650Ti and with HD7850, but this is where the performance stands.


With no changes with the voltage, I managed to achieve 45MHz more on the GPU core and +29MHz on the memory core. I can increase it a bit more, but in certain tests such as Catzilla Benchmark, the last benchmark part of Sleeping Dogs and METRO 2033 gets locked.

3DMark 11 Benchmark


Catzilla BETA ALLBenchmark


METRO: 2033


Sleeping Dogs


The performance difference between the overclocked and non overclocked non-reference GTX 660 from Asus is not very satisfied with higher preset/resolution.

About Nvidia’s GTX 660 GK106:

The core’s overclocked performance in comparison with the non-reference @ stock setting is not impressive. One would be curious about the performance difference between the reference and non-reference to see if there is any significant boost.

There’s also the issue of games getting locked if I go more than 1117MHz on GPU and/or 1542 MHz on the memory clock. The card’s core ability to give nice overclock performance results is restricted. The card uses 6-pin power on the non-reference card, and one may suspect if this got to do with Power limiter that Nvidia has on this by default. I’ve read GTX 660Ti gives better tweaks and performance difference even on reference editions are pretty good.

The performance is pretty decent only with 1680 x 1050 (and possibly 1080p resolution) with low/medium preset and high in certain games. But the beauty of the ability to overclock a PC hardware for gaming and therefore get better performance is stifled.

About Asus’ card:

Asus has a good card but with a mediocre GPU. Not that I am saying GTX 660 it’s a bad product, it’s just that the part where certain games/benchmarks get locked after raising GPU Core and memory clock to a level. If the performance difference between a reference and non-reference on the stock is also insignificant, you need to ask: Why make a non-reference for such GPUs if the performance and overclocked performance in comparison with stock is disappointing?

There is something peculiar about the fans on auto. Even when idle and on stock the fan tends to spin up (the noise is audible once you push it 69% fan speed) at certain intervals. I am not sure if this is something that Asus has embedded on, but it didn’t happen on HD7850 and 650Ti- both Direct CUII Editions.

The non-reference cooler is appreciated provided the manufacturer makes a better-than-reference card overall. It keeps the card cool- with ALLHD benchmark on 1440 preset (cat Zilla), it stayed 70 degrees on load with an ambient of around 25 degrees Celsius on an open test setup.

I’ve read ABOUT GTX 660Ti card and it does overclock nicely. Even the reference versions do a better job and give better overclocks- and also, you can go for tri/Quad SLI setup with 660Ti whereas 660 is restricted with dual card setup. You may want to look into that and/or see what the red team has.

Leave a Reply

Your email address will not be published. Required fields are marked *