Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Which GPUs support GDS #5

Open
zhileiz opened this issue Jul 20, 2021 · 2 comments
Open

Which GPUs support GDS #5

zhileiz opened this issue Jul 20, 2021 · 2 comments

Comments

@zhileiz
Copy link

zhileiz commented Jul 20, 2021

Hello,

The README says that:

For example GDS requires the latest Tesla, Volta or Ampere GPUs, 
ext4 mounting options, and GDS enabled storage systems (an NVMe drive at minimum). 
More on GDS setup in the NVIDIA GPUDirect Storage Installation and Troubleshooting Guide

However, the hardware/cuds-capability requirement is not mentioned in GDS setup in the NVIDIA GPUDirect Storage Installation and Troubleshooting Guide or anywhere else on the web.

Moreover, I was able to follow developer-onizuka's repo to set up GPS using a Quadro P400 card. However, I noticed that gdsio utility's GPUD mode has worse throughput and latency performance than CPU_GPU mode.

Would you please specify which GPUs (or cuda capabilities) are actually supported by GDS? Thank you!

@KiranModukuri
Copy link
Collaborator

GDS is supported only on NVIDIA graphics processing units (GPU) Tesla® or Quadro® models that support compute mode, and a compute major capability greater than or equal to 6.

Note: This includes A100, V100 and T4 cards.

The performance of GDS depends on multiple factors.

  1. The p2p performance between the NVMe device and the GPU. we have seen good performance on systems with GPU and NVMe devices connected under a PCIe switch.

  2. GDS is tuned for IO request sizes >= 4KB.

  3. offsets and IO aligned to 4KB size

@johnnynunez
Copy link

Why is it not compatible with the RTX Turing and RTX Ampere series? For university level it is clear that for students it is a long way to go to buy a quadro. If the rtx have this technology why is it not activated? In fact for gaming it is the same idea with these two series of graphics cards. RTX IO.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants