-
Notifications
You must be signed in to change notification settings - Fork 345
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
Occupancy Calculator for AMD in Excel (similar to CUDA) #1689
Comments
Hi @nikowleye, |
Thank you for sharing your document, @nikowleye. We are discussing internally how to proceed with your submission and hope to give you an update soon. Regards, |
@Rmalavally hi, just curious, are there plans to integrate this into ROCm? |
Hi, @nikl-i. I have some questions about the value of VGPR per CU. Is there any evidence that one CU have 65536 VGPRs ? Based on the information provided in amd-cdna-whitepaper, each CU of the MI100 has a 512KB-sized VGPR file. If there are 65536 VGPRs per CU, does that mean the VGPRs are all 64-bit in size? |
Hi, @ROCmSupport @Rmalavally . |
gfx9 "GCN" GPUs (i.e., not CDNA accelerators such as MI100 or MI200) have a compute unit architecture that:
You can learn a bit more about the general "GCN" architecture in this presentation: https://www.slideshare.net/DevCentralAMD/gs4106-the-amd-gcn-architecture-a-crash-course-by-layla-mah, e.g., slide 19 describes the "4x SIMD16" architecture of our GCN CUs and shows the VGPR size. See also https://www.olcf.ornl.gov/wp-content/uploads/2019/10/ORNL_Application_Readiness_Workshop-AMD_GPU_Basics.pdf. On the MI100 CDNA1 accelerator, the compute unit architecture:
On the MI200 CDNA2 accelerator, the compute unit architecture:
For a bit more coverage of the MI200 topic, you can see the media reports from our HotChips talk: https://chipsandcheese.com/2022/09/18/hot-chips-34-amds-instinct-mi200-architecture/ which gives some extra commentary from the Q&A session of the talk itself (https://hc34.hotchips.org/assets/program/conference/day1/GPU%20HPC/HC2022.AMD.AlanSmith.v14.Final.20220820.pdf), where this was covered in 1 bullet point on slide 8. |
Hi, @jlgreathouse. |
Hello! I've made recently an Occupancy Calculator for AMD GPUs similar to CUDA Occupancy Calculator](https://docs.nvidia.com/cuda/cuda-occupancy-calculator/index.html), and would like to share it somehow.
It's Excel file (like CUDA calculator) and includes several plots and summary information about occupancy factors with links to documentation and other materials.
If case you find useful, it would be great if you could suggest a way to make it more available to community (add to docs or something).
Thanks!
Screenshots:
Calculator inself (*.xlsx file in *.zip archive):
ROCm-Occupancy-Calculator.zip
The text was updated successfully, but these errors were encountered: