Hw accelaration support with geforce [GeForce GT 710]

[ckruijntjens]

Describe the problem you are having

Hi guys,

Is it possible to get hardware acceleration with this gpu card? [GeForce GT 710] it is an older card. Max Cuda version 11.4.

I tried but i cant get it to work. Is the some way to do this? If so can you help me on this matter.

Version

Latest stable release.

Frigate config file

-

docker-compose file or Docker CLI command

-

Relevant Frigate log output

-

Relevant go2rtc log output

-

FFprobe output from your camera

-

Install method

Docker Compose

Object Detector

CPU (no coral)

Network connection

Wired

Camera make and model

Reolink

Screenshots of the Frigate UI's System metrics pages

Any other information that may be helpful

Yust want to know if it is possible or not.

[inkeep-ai-assistant[bot]]

The GeForce GT 710 has limited support for hardware acceleration in Frigate. Here's what you need to know:

CUDA Compatibility Requirements

For NVIDIA GPU hardware acceleration, Frigate requires specific minimum hardware support(1). The TensorRT detector uses the 12.x series of CUDA libraries and requires a minimum driver version of >=530(1). Additionally, the GPU must support a Compute Capability of 5.0 or greater, which generally correlates to a Maxwell-era GPU or newer(1).

The GeForce GT 710 is based on the older Kepler architecture and has a Compute Capability of 3.5, which falls below the minimum requirement of 5.0(1).

TensorRT Detector Limitations

The TensorRT detector is only available in images with the -tensorrt tag suffix, such as ghcr.io/blakeblackshear/frigate:stable-tensorrt(1). However, given your GPU's compute capability limitation, this detector would not work with your GT 710.

Alternative Options

Since your GeForce GT 710 doesn't meet the minimum requirements for CUDA hardware acceleration in Frigate, you have these alternatives:

1. CPU Detection: Continue using CPU-based object detection(2)
2. Google Coral TPU: Consider adding a Google Coral TPU, which provides excellent performance for object detection. A $60 device will outperform a $2000 CPU(2)
3. Intel-based hardware acceleration: If your system has an Intel CPU with integrated graphics, you might be able to use VAAPI or QuickSync acceleration instead(3)

The GT 710 unfortunately cannot provide hardware acceleration for Frigate due to its older architecture and insufficient compute capability.

📚 Sources:

1. Supported Hardware
2. Recommended hardware
3. Hardware Acceleration

---

Have another question? Just tag @inkeep.

[ckruijntjens]

@inkeep

What video card do you advice?

[inkeep-ai-assistant[bot]]

Based on the Frigate documentation, here are the recommended GPU options for hardware acceleration:

NVIDIA GPUs (TensorRT)

For NVIDIA GPUs, you need a Maxwell-era GPU or newer with Compute Capability of 5.0 or greater(1). The minimum driver version must be >=530(1). Some performance examples from the documentation include:

• RTX 3070 Mobile: ~5ms inference time(2)
• RTX 3050: 5-7ms inference time(2)
• GTX 1660 SUPER: ~4ms inference time(2)
• GTX 1070: ~6ms inference time(2)
• GTX 1060 6GB: ~7ms inference time(2)

Intel GPUs (OpenVINO)

Intel Arc GPUs offer excellent performance with OpenVINO(2):

• Intel Arc A750: ~4ms MobileNetV2 inference time(2)
• Intel Arc A380: ~6ms MobileNetV2 inference time(2)

Intel integrated GPUs also work well, supporting 6th Gen Intel platforms (Skylake) and newer(3).

AMD GPUs (ROCm)

AMD discrete GPUs are supported with the ROCm detector using the -rocm Frigate image(3). However, specific performance benchmarks aren't provided in the documentation.

My Recommendation

For your use case, I'd recommend considering an Intel Arc A380 or A750 as they offer excellent price-to-performance ratios with very low inference times(2). Alternatively, if you prefer NVIDIA, a GTX 1660 SUPER or RTX 3050 would provide good performance(2).

However, the documentation strongly emphasizes that a Google Coral TPU ($60) will outperform a $2000 CPU and is the most cost-effective option for object detection(2).

📚 Sources:

1. Supported Hardware
2. Recommended hardware
3. Supported Hardware

---

Have another question? Just tag @inkeep.

[ckruijntjens]

@inkeep when using google coral tpu do i need a gpu also?

[NickM-27]

The AI is wrong here. Having a gpu for hardware acceleration is highly encouraged and having a coral does not change that

[ckruijntjens]

The AI is wrong here. Having a gpu for hardware acceleration is highly encouraged and having a coral does not change that

Ok thanks for your reply. What does rhe coral do then? And what video card would you recommend now a days?

[NickM-27]

https://docs.frigate.video/frigate/hardware#do-hwaccel-args-help-if-i-am-using-a-coral

[inkeep-ai-assistant[bot]]

No, you do not need a GPU when using a Google Coral TPU(1). The Google Coral TPU handles the object detection independently and is designed to work with any system that can run Frigate(2).

Google Coral TPU as Standalone Solution

The Google Coral EdgeTPU is available in USB and M.2 formats, allowing for wide compatibility with devices(1). A single Coral can handle many cameras using the default model and will be sufficient for the majority of users(2). The documentation emphasizes that a $60 Coral device will outperform a $2000 CPU(2).

Multiple Detectors Cannot Be Mixed

It's important to note that multiple detectors cannot be mixed for object detection (for example, OpenVINO and Coral EdgeTPU cannot be used for object detection at the same time)(1)(3). This means you would use either the Coral TPU OR a GPU detector, not both simultaneously.

Configuration Example

Here's how you would configure a single USB Coral without any GPU:

detectors:
  coral:
    type: edgetpu
    device: usb

(1)

Performance Calculation

You can calculate the maximum performance of your Coral based on the inference speed reported by Frigate(2). With an inference speed of 10ms, your Coral will top out at 1000/10=100, or 100 frames per second(2). If your detection fps regularly approaches that limit, you should first consider tuning motion masks(2).

The Google Coral TPU is a complete object detection solution that doesn't require additional GPU hardware.

📚 Sources:

1. Supported Hardware
2. Recommended hardware
3. Supported Hardware

---

Have another question? Just tag @inkeep.