One thing I need you to understand first: you have to provide the database and the queries. This is a great way to ensure your system is working up to spec.So you wanna run a load test against SQL Server. You can use this to monitor the CPU, memory, I/O bandwidth, and GPU utilization of your system as well as the CPU and GPU temperatures. I like to use tmux to split my terminal and have htop, sudo iotop, watch sudo sensors, and watch nvidia-smi running simultaneously. gpu_burn 60 # will run gpu_burn for 60 seconds Monitor with htop, nvidia-smi, and sensors. Stress -cpu `nproc` -vm `nproc` -vm-bytes 1GB -io `nproc` -hdd `nproc` -hdd-bytes 1GB -timeout 60s Install and run gpu-burn git clone
# `nproc` workers writing to disk (write / unlink) # `nproc` Virtual Memory workers (malloc / free) # Run a stress test with `nproc` CPU workers ( sqrt) sudo apt-get install -y stress htop iotop lm-sensors On Ubuntu, you can install stress, htop, and iotop via apt-get. This article assumes you have the proper GPU drivers in place, check out Lambda Stack if you need to install CUDA, CuDNN, and the NVIDIA drivers. The easiest way to make your own monitoring tool with the output of a script is with the watch command. To stress test a system for simultaneous GPU and CPU loads, we'll use two stress tools: stress and gpu_burn, and three monitoring tools: htop, iotop and nvidia-smi. Often times you'll want to put a system through the paces after it's been set up.
CPU, GPU, and I/O utilization monitoring using tmux, htop, iotop, and nvidia-smi.
0 kommentar(er)
