NVIDIA GeForce GTX 670

NVIDIA GeForce GTX 670

NVIDIA GeForce GTX 670: Outdated Warrior or Budget Option in 2025?

An analysis of the capabilities and limitations of this legendary card in modern realities


1. Architecture and Key Features

Kepler Architecture: The Beginning of a New Era

Released in 2012, the GTX 670 is based on the Kepler architecture (GK104 chip), which marked a breakthrough for NVIDIA thanks to its energy efficiency and support for DirectX 11. The manufacturing process is 28 nm, which was cutting edge at the time. The card has 1344 CUDA cores, dynamic overclocking via GPU Boost (base clock at 915 MHz, boost up to 980 MHz), and support for PCIe 3.0.

What’s Missing?

The GTX 670 does not support modern technologies such as ray tracing (RTX), DLSS, or FidelityFX. These features emerged in later generations of cards (starting from the RTX 20 series). Its capabilities are limited to basic graphical tasks without hardware acceleration for AI or complex effects.


2. Memory: Modest Resources for Modern Standards

GDDR5 and Bandwidth

The card is equipped with 2 GB of GDDR5 memory with a 256-bit bus. The bandwidth is 192 GB/s. This was sufficient for games made between 2012 and 2015, but by 2025, even at minimal settings in projects like Cyberpunk 2077 or Starfield, this amount becomes a critical limitation. High-resolution textures and complex shaders quickly drain resources, leading to lag and dropping FPS.

Advice: For comfortable performance in 2025, a minimum of 6–8 GB of VRAM is recommended. The GTX 670 will only be suitable for older games or less demanding indie projects.


3. Gaming Performance: Nostalgia Instead of Power

1080p: A Basic Level

In 2010s games like The Witcher 3 or GTA V, the GTX 670 achieves around 40–50 FPS on medium settings. However, in modern AAA titles, the situation is different:

- Cyberpunk 2077 (Low, 1080p): 15–20 FPS.

- Elden Ring (Low, 1080p): 20–25 FPS.

- Fortnite (Medium, 1080p): 30–35 FPS (without shadow effects enabled).

4K? Forget It

The card is not intended for resolutions above 1080p. Even 1440p in most games becomes a daunting task due to a lack of memory and computational power.


4. Professional Tasks: A Very Limited Niche

CUDA: A Plus for Old Applications

CUDA support allows the GTX 670 to be used in programs like Adobe Premiere Pro or Blender, but only for simple tasks. Rendering complex 3D scenes will take much longer compared to modern cards with Tensor cores.

Scientific Computing: Not Relevant

For machine learning or neural network tasks, at least 8 GB VRAM and architecture supporting FP16/INT8 are required. The GTX 670 is hopelessly outdated in this regard.


5. Power Consumption and Heat Output: Not the Worst Figures

TDP 170 W: Modest for 2025

By current standards, the card is relatively energy-efficient. A system with the GTX 670 requires a power supply of 450–500 W (including a safety margin).

Cooling: Noise vs. Temperatures

Reference models with a turbofan could reach temperatures of 80°C under load. It is recommended to use a case with good ventilation (2–3 fans) and regularly clean the system of dust.


6. Comparison with Competitors: Time Spares No One

Direct Competitors of 2012:

- AMD Radeon HD 7970: 3 GB GDDR5, slightly higher performance in DX11, but similar issues with modern games.

Modern Analogs (2025):

- NVIDIA RTX 3050 (8 GB GDDR6): Support for DLSS 3, RTX, new model prices — $250–300.

- AMD Radeon RX 6600 (8 GB GDDR6): Better performance at 1080p, $200–230.

The GTX 670 lags behind even budget new releases of 2025 by 3–4 times in speed.


7. Practical Advice: Caution, Retro Card!

Power Supply: Minimum 450 W (80+ Bronze). Ensure there is an 8-pin power connector.

Compatibility:

- Motherboards: PCIe 3.0 is backward compatible with PCIe 4.0/5.0, but performance will not improve.

- Drivers: Official NVIDIA support has been discontinued. The last versions are from 2021. There may be issues with Windows 11 and new games.

Tip: Consider the GTX 670 only as a temporary solution or for building a retro PC.


8. Pros and Cons

Pros:

- Low price on the secondary market ($30–50).

- Energy efficiency for its class.

- CUDA support for basic tasks.

Cons:

- 2 GB VRAM is critically low.

- No support for modern technologies (RTX, DLSS).

- Lack of drivers for new OS and games.


9. Final Conclusion: Who Is the GTX 670 Suitable for in 2025?

This card is a choice for:

1. Retro computer enthusiasts building PCs for games from the 2000s to 2010s.

2. Office systems where image output and working with 2D applications are required.

3. A temporary solution if the main card fails.

Why Not Buy a New One?

The GTX 670 has been discontinued, and any new equivalents (if found) will be overvalued. It’s better to look at budget models from 2023 to 2025 — they will ensure stable performance under current conditions.


Conclusion

The NVIDIA GeForce GTX 670 is a significant milestone in GPU history, but by 2025 its time has passed. Unless you are a collector or on a tight budget, invest in more modern solutions. Technology does not stand still, and even $200 today will grant access to cards with AI rendering support and 4K gaming capabilities.

Basic

Label Name
NVIDIA
Platform
Desktop
Launch Date
May 2012
Model Name
GeForce GTX 670
Generation
GeForce 600
Base Clock
915MHz
Boost Clock
980MHz
Bus Interface
PCIe 3.0 x16
Transistors
3,540 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.
112
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.
256bit
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.
192.3 GB/s

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.
27.44 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.
109.8 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.
109.8 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.
2.581 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.
1344
L1 Cache
16 KB (per SMX)
L2 Cache
512KB
TDP
170W
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
DirectX
12 (11_0)
CUDA
3.0
Power Connectors
2x 6-pin
Shader Model
5.1
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.
32
Suggested PSU
450W

Benchmarks

FP32 (float)
Score
2.581 TFLOPS
3DMark Time Spy
Score
1770
Blender
Score
217
OctaneBench
Score
37
Vulkan
Score
16062
OpenCL
Score
14826
Hashcat
Score
55110 H/s

Compared to Other GPU

FP32 (float) / TFLOPS
2.536 -1.7%
3DMark Time Spy
5182 +192.8%
3906 +120.7%
2755 +55.6%
Blender
1506.77 +594.4%
848 +290.8%
45.58 -79%
OctaneBench
123 +232.4%
69 +86.5%
Vulkan
98446 +512.9%
69708 +334%
40716 +153.5%
18660 +16.2%
OpenCL
62821 +323.7%
38843 +162%
21442 +44.6%
884 -94%
Hashcat / H/s
58476 +6.1%
55260 +0.3%
53248 -3.4%
52572 -4.6%