Skip to content

Correct performance metrics! #537

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.

Sign up for GitHub

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

Merged
merged 1 commit into from
Jul 27, 2024
Merged

Correct performance metrics! #537

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
37 changes: 24 additions & 13 deletions docs/documentation/expectedPerformance.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,21 +6,32 @@ This page shows a summary of these results.
## Expected time-steps/hour

The following table outlines observed performance as nanoseconds per grid point (ns/GP) per equation (eq) per right-hand side (rhs) evaluation (lower is better).
We solve an example 3D, inviscid, 5-equation model problem with two advected species (a total of 8 PDEs).
The numerics are WENO5 and the HLLC approximate Riemann solver.
We solve an example 3D, inviscid, 5-equation model problem with two advected species (8 PDEs) and 8M grid points (158-cubed uniform grid).
The numerics are WENO5 finite volume reconstruction and HLLC approximate Riemann solver.
This case is located in `examples/3D_performance_test`.
We report results for various numbers of grid points per CPU die (or GPU device) and hardware.

| Hardware | | 1M GPs | 4M GPs | 8M GPs | Compiler | Computer |
| ---: | :----: | :----: | :---: | :---: | :----: | :--- |
| NVIDIA V100 | 1 device | 12.0 | 13.0 | 13.0 | NVHPC 22.11 | PACE Phoenix |
| NVIDIA V100 | 1 device | 12.6 | 13.0 | 13.0 | NVHPC 22.11 | OLCF Summit |
| NVIDIA A100 | 1 device | 8.9 | 7.0 | 7.4 | NVHPC 23.5 | Wingtip |
| AMD MI250X | 1 GCD | 13.5 | 11.3 | 12 | CCE 16.0.1 | OLCF Frontier |
| Intel Xeon Gold 6226 | 12 cores | 245 | 211 | 211 | GNU 10.3.0 | PACE Phoenix |
| Apple M2 | 6 cores | 365 | 306 | 563 | GNU 13.2.0 | N/A |

__All results are in nanoseconds (ns) per grid point (gp) per equation (eq) per right-hand side (rhs) evaluation, so X ns/gp/eq/rhs. Lower is better.__
Results are for MFC v4.9.3 (July 2024 release), though numbers have not changed meaningfully since then.
All results are for the compiler that gave the best performance.
CPU results may be performed on CPUs with more cores than reported in the table; we report results for the best performance given the full processor die by checking the performance for different core counts on that device.
GPU results on single-precision (SP) GPUs performed computation in double-precision via conversion in compiler/software; these numbers are _not_ for single-precision computation.
AMD MI250X GPUs have two graphics compute dies (GCDs) per MI250X device; we report results for one GCD, though one can quickly estimate full MI250X runtime by halving the single GCD grind time number.


| Hardware | | Grind Time | Compiler | Computer |
| ---: | ----: | :----: | :--- | :--- |
| NVIDIA GH200 (GPU only) | 1 GPU | 0.32 | NVHPC 24.1 | GT Rogues Gallery |
| NVIDIA H100 | 1 GPU | 0.45 | NVHPC 24.5 | GT Rogues Gallery |
| NVIDIA A100 | 1 GPU | 0.62 | NVHPC 22.11 | GT Phoenix |
| NVIDIA V100 | 1 GPU | 0.99 | NVHPC 22.11 | GT Phoenix |
| NVIDIA A30 | 1 GPU | 1.06 | NVHPC 24.1 | GT Rogues Gallery |
| AMD MI250X | 1 __GCD__ | 1.09 | CCE 16.0.1 | OLCF Frontier |
| NVIDIA A40 (SP GPU) | 1 GPU | 3.3 | NVHPC 22.11 | NCSA Delta |
| NVIDIA RTX6000 (SP GPU) | 1 GPU | 3.9 | NVHPC 22.11 | GT Phoenix |
| Apple M1 Max | 8 cores | 72 | GNU 14.1.0 | N/A |
| AMD EPYC 7713 | 32 cores | 137 | GNU 12.1.0 | GT Phoenix |
| Intel Xeon Gold 6226 | 12 cores | 152 | Intel oneAPI 2022.1 | GT Phoenix |

__All grind times are in nanoseconds (ns) per grid point (gp) per equation (eq) per right-hand side (rhs) evaluation, so X ns/gp/eq/rhs. Lower is better.__

## Weak scaling

Expand Down
Loading