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

run TLB bench on RTX2080Ti #4

Open
qqwert0 opened this issue Mar 12, 2021 · 1 comment
Open

run TLB bench on RTX2080Ti #4

qqwert0 opened this issue Mar 12, 2021 · 1 comment

Comments

@qqwert0
Copy link

qqwert0 commented Mar 12, 2021

Hello,I run the TLB-bench on RTX2080Ti. The result is not a relatively smooth curve, but a constantly fluctuating curve.
I clone you git and follow the README. And I change the data size and stride, but the curves are always fluctuating. Do you know why? How can I test the TLB size of RTX2080Ti? thank you.

0

2

1

@tomas-karnagel
Copy link
Collaborator

tomas-karnagel commented Mar 12, 2021

Hi, thanks a lot for using our code. The work is 4 years old and I can't remember everything exactly without looking into the details.
But here are some pointers that might help.

  • We profiled the K80 (Kepler architecture) and P100 (Pascal architecture) and there it was possible to see the TLB boundaries clearly. However, we also tested a GPU with Maxwell architecture and there it was hard to see anything. So my guess is that Maxwell was completely different to the other two. Your GPU has the Turing architecture with a newer GDDR memory. So basically every architecture might be different and it can be hard to see these TLB boundaries.
  • For the P100, there was a small boundary (L1) at 32 MB with a 8 cycle difference. With newer architectures this is likely to be the same or higher MB with less cycles. Maybe in your last image, it looks like that there is a shift at 32 MB. If you change your y-axis of the image to 230 to 250 it might become more visible.
  • For the P100, the big shift (L2) (110 cycles) came at 2 GB. Maybe start looking in this area first, because the larger shift might be more visible.
    I hope this helps.
    Good luck.
    --
    Tomas

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

2 participants