GeForce GTX 660 vs 670

Place in performance ranking	353	405
Place by popularity	not in top-100	69
Cost-effectiveness evaluation	3.41	2.40
Architecture	Kepler (2012−2018)	Kepler (2012−2018)
GPU code name	GK104	GK106
Market segment	Desktop	Desktop
Release date	10 May 2012 (12 years ago)	6 September 2012 (11 years ago)
Launch price (MSRP)	$399	$229
Current price	$22 (0.1x MSRP)	$172 (0.8x MSRP)

Cost-effectiveness evaluation

Performance to price ratio. The higher, the better.

GTX 670 has 42% better value for money than GTX 660.

Detailed specifications

General performance parameters such as number of shaders, GPU core base clock and boost clock speeds, manufacturing process, texturing and calculation speed. These parameters indirectly speak of performance, but for precise assessment you have to consider their benchmark and gaming test results. Note that power consumption of some graphics cards can well exceed their nominal TDP, especially when overclocked.

Pipelines / CUDA cores	1344	960
CUDA cores	1344	960
Core clock speed	915 MHz	980 MHz
Boost clock speed	980 MHz	1033 MHz
Number of transistors	3,540 million	2,540 million
Manufacturing process technology	28 nm	28 nm
Power consumption (TDP)	170 Watt	140 Watt
Maximum GPU temperature	97 °C	no data
Texture fill rate	102.5 billion/sec	78.4 billion/sec
Floating-point performance	2,459.5 gflops	1,981 gflops

Form factor & compatibility

Information on compatibility with other computer components. Useful when choosing a future computer configuration or upgrading an existing one. For desktop graphics cards it's interface and bus (motherboard compatibility), additional power connectors (power supply compatibility).

Bus support	PCI Express 3.0	PCI Express 3.0
Interface	PCIe 3.0 x16	PCIe 3.0 x16
Length	9.5" (24.1 cm)	9.5" (24.1 cm)
Height	4.376" (11.1 cm)	4.376" (11.1 cm)
Width	2-slot	2-slot
Supplementary power connectors	Two 6-pin	One 6-pin
SLI options	+	+

VRAM capacity and type

Parameters of VRAM installed: its type, size, bus, clock and resulting bandwidth. Integrated GPUs have no dedicated video RAM and use a shared part of system RAM.

Memory type	GDDR5	GDDR5
Maximum RAM amount	2 GB	2 GB
Memory bus width	256-bit GDDR5	192-bit GDDR5
Memory clock speed	6.0 GB/s	6.0 GB/s
Memory bandwidth	192.2 GB/s	144.2 GB/s

Connectivity and outputs

Types and number of video connectors present on the reviewed GPUs. As a rule, data in this section is precise only for desktop reference ones (so-called Founders Edition for NVIDIA chips). OEM manufacturers may change the number and type of output ports, while for notebook cards availability of certain video outputs ports depends on the laptop model rather than on the card itself.

Display Connectors	One Dual Link DVI-I, One Dual Link DVI-D, One HDMI, One DisplayPort	One Dual Link DVI-I, One Dual Link DVI-D, One HDMI, One DisplayPort
Multi monitor support	4 displays	4 displays
HDMI	+	+
HDCP	+	+
Maximum VGA resolution	2048x1536	2048x1536
Audio input for HDMI	Internal	Internal

Supported technologies

Supported technological solutions. This information will prove useful if you need some particular technology for your purposes.

3D Blu-Ray	+	+
3D Gaming	+	+
3D Vision	+	+

API compatibility

List of supported graphics and general-purpose computing APIs, including their specific versions.

DirectX	12 (11_0)	12 (11_0)
Shader Model	5.1	5.1
OpenGL	4.2	4.3
OpenCL	1.2	1.2
Vulkan	1.1.126	1.1.126
CUDA	+	+

Synthetic benchmark performance

Non-gaming benchmark performance comparison. The combined score is measured on a 0-100 point scale.

Combined synthetic benchmark score

This is our combined benchmark performance score. We are regularly improving our combining algorithms, but if you find some perceived inconsistencies, feel free to speak up in comments section, we usually fix problems quickly.

GTX 670 13.80
+33.3%

GTX 660 10.35

670 outperforms 660 by 33% based on our aggregate benchmark results.

Passmark

This is the most ubiquitous GPU benchmark, part of Passmark PerformanceTest suite. It gives the graphics card a thorough evaluation under various types of load, providing four separate benchmarks for Direct3D versions 9, 10, 11 and 12 (the last being done in 4K resolution if possible), and few more tests engaging DirectCompute capabilities.

Benchmark coverage: 25%

GTX 670 5331
+33.3%

GTX 660 3999

670 outperforms 660 by 33% in Passmark.

3DMark Fire Strike Graphics

Fire Strike is a DirectX 11 benchmark for gaming PCs. It features two separate tests displaying a fight between a humanoid and a fiery creature made of lava. Using 1920x1080 resolution, Fire Strike shows off some realistic graphics and is quite taxing on hardware.

Benchmark coverage: 14%

GTX 670 7000
+38.9%

GTX 660 5040

670 outperforms 660 by 39% in 3DMark Fire Strike Graphics.

GeekBench 5 OpenCL

Geekbench 5 is a widespread graphics card benchmark combined from 11 different test scenarios. All these scenarios rely on direct usage of GPU's processing power, no 3D rendering is involved. This variation uses OpenCL API by Khronos Group.

Benchmark coverage: 9%

GTX 670 15436
+36.7%

GTX 660 11292

670 outperforms 660 by 37% in GeekBench 5 OpenCL.

GeekBench 5 Vulkan

Benchmark coverage: 5%

GTX 670 15583
+35.8%

GTX 660 11472

670 outperforms 660 by 36% in GeekBench 5 Vulkan.

GeekBench 5 CUDA

Benchmark coverage: 4%

GTX 670 11611
+35.3%

GTX 660 8583

670 outperforms 660 by 35% in GeekBench 5 CUDA.

Octane Render OctaneBench

This is a special benchmark measuring graphics card performance in OctaneRender, which is a realistic GPU rendering engine by OTOY Inc., available either as a standalone program, or as a plugin for 3DS Max, Cinema 4D and many other apps. It renders four different static scenes, then compares render times with a reference GPU which is currently GeForce GTX 980. This benchmark has nothing to do with gaming and is aimed at professional 3D graphics artists.

Benchmark coverage: 4%

GTX 670 49
+48.5%

GTX 660 33

670 outperforms 660 by 48% in Octane Render OctaneBench.