Running AI Benchmark

 Here is how l got this useful AI benchmark python program to successfully run on my several machines.

 Note it seems to only work with Python 3.7

"AI Benchmark is an open source python library for evaluating AI performance of various hardware platforms, including CPUs, GPUs and TPUs."

AI-Benchmark website source link - click here .

Designed and Developed by Andrey Ignatov - click here for the creators project website. 

Link on installing Tensorflow and nvidia cudatoolkit [for GPU] , click here.

Using Anaconda conda

1)

Create an environment with the specific python version needed using:

conda create --name myenv37 python=3.7

2)

Activate it

conda activate myenv37

3)

Install tensorflow

pip install tensorflow

then run:

pip install ai-benchmark

Screenshots:

4)

Inside your conda environment session console:   run python

Enter the code below:

from ai_benchmark import AIBenchmark
benchmark = AIBenchmark()
results = benchmark.run()

Some system results below:

Laptop - Lenovo Legion 5 Intel i7 -Windows 11 - 32Gb RAM with RTX 4060 6Gb VRAM  [ benchmark result with NO GPU support]

For GPU, l ran below - see previous TensorFlow Link:

conda install -c conda-forge cudatoolkit=11.2 cudnn=8.1.0
# Anything above 2.10 is not supported on the GPU on Windows Native
python -m pip install "tensorflow<2.11"
# Verify the installation:
python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

I got:

Results [w/GPU on my Lenovo laptop]

So with CPU alone l got an AI benchmark of 1912, with my GPU l got a result of 19797, thats a difference of a factor by 10.35!

Results [w/CPU alone on my Workstation PC - Windows 11 - AMD w/64Gb RAM and the Nvidia RTX3090 GPU]

Results [w/GPU alone on my Workstation PC - RTX 3090 24Gb VRAM ]

So with CPU alone on my powerful desktop main ML workstation l got an AI benchmark of 1205, with my GPU l got a result of 34248, that's a difference of a factor by 28.4!

A difference also of almost 3X between my laptop and my core PC.

When l get hold/access to a PC with the top RTX 5090/5090ti l will try to run this benchmark again.

The RTX 3090 has significantly more CUDA cores than the RTX 4060. The RTX 3090 has 10,496 CUDA cores, while the RTX 4060 has 3,072 CUDA cores. 

This means the RTX 3090 has approximately 3.4 times the number of CUDA cores found in the RTX 4060.

The NVIDIA RTX 5090 has 21,760 CUDA cores, more than double the 10,496 CUDA cores of the RTX 3090, representing a massive generational leap in parallel processing power with its Blackwell architecture, alongside faster memory and enhanced RT/Tensor cores, leading to significantly higher overall performance. 

Key Takeaways:

• Massive Core Count: The RTX 5090's core count is significantly higher, enabling much faster parallel processing for tasks like rendering, AI, and simulations.
• New Architecture: The shift from Ampere (3090) to Blackwell (5090) means each core is also more powerful and efficient.
• Performance Boost: This huge core increase, combined with faster memory (GDDR7 vs GDDR6X) and better Tensor Cores, leads to substantial performance gains, not just an incremental update.
•