Thelio Mega has been specially designed for intensive graphical work tasks. Tests include Inception3, Resnet50, Resnet152, VGG16 and Inception4 models. I used Tensorman, available in Pop! _OS, to run tests. Tensorman is a tool that facilitates the management of Tensorflow tool chains.
The commands used to run these tests in Pop! _OS 20.04 LTS
Install Tensorman
sudo apt install tensorman sudo apt install nvidia-container-runtime sudo usermod -aG docker $ USER reboot
Clone the benchmarks
git clone https://github.com/tensorflow/benchmarks.git cd benchmarks
Perform the tests
Tensorflow night version is required for RTX 3090. Change –num_gpus = to match your setting and –batch_size = for the desired test.
tensorman pull nightly tensorman + nightly run –gpu python - ./scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py –batch_size = 128 –model = inception3 –variable_update = parameter_server –use_fp16 = True –num_gpus = 4 tensorman + nightly run –gpu python - ./scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py –batch_size = 64 –model = inception4 –variable_update = parameter_server –use_fp16 = True –num_gpus = 4 tensorman + nightly run –gpu python - ./scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py –batch_size = 128 –model = resnet50 –variable_update = parameter_server –use_fp16 = True –num_gpus = 4 tensorman + nightly run –gpu python - ./scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py –batch_size = 64 –model = resnet152 –variable_update = parameter_server –use_fp16 = True –num_gpus = 4 tensorman + nightly run –gpu python - ./scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py –batch_size = 128 –model = vgg16 –variable_update = parameter_server –use_fp16 = True –num_gpus = 4
The above tests are for measuring the performance of components, but not the computer and its heating system. Intense use of the GPU over an extended period of time demonstrates system performance in real-world scenarios. More details about that test and the results for Thelio Mega will appear soon.
Thelio Mega Configuration
AMD Ryzen Threadripper 3990X
250 GB Samsung 970 Evo Plus NVMe M.2 Drive
64GB 3200 MHz Kingston HyperX Memory
4 x NVIDIA GeForce RTX 2080ti (Gigabyte GV-N208TTURBO-11GC)
3 x NVIDIA GeForce RTX 3090 (Gigabyte G BO 24GD)