Repository navigation
Completing the GPU Performance Picture: Understanding TAF Alongside Peak FLOPs and MAF — ROCm Blogs #600
Replies: 2 comments
|
Nice! |
0 replies
|
This is a really helpful breakdown of why a single FLOPs number never tells the whole story. The TAF approach with a fixed workload-derived GEMM set makes a lot of sense for repeatable comparisons. It is interesting to see the MI325X TAF values landing at 74 to 81 percent of MAF, which shows how much shape selection can skew results. Looking forward to seeing how this metric gets adopted across the industry. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Completing the GPU Performance Picture: Understanding TAF Alongside Peak FLOPs and MAF — ROCm Blogs
Learn how Typical Attained FLOPs complements Peak FLOPs and Max-Achievable FLOPs, including the methodology and MI325X results
https://rocm.blogs.amd.com/artificial-intelligence/taf-blog/README.html
All reactions