Wip/rx 7800 xt results #33
Open
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- AMD GPU benchmarks works on rocm platform (tested on rocm 6.1.1) and offers support of using CUDA api with AMD gpus - only code change is needed is to test.sh to check whether rocm-smi tool is available and whether it returns AMD GPUs - rocm-smi can be used to query the AMD gpu count - I had pytorch 2.3.0 and pytorch vision 0.18.0 installed with AMD rocm support enabled and all tests executed without errors on AMD RX 6800 Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
(cherry picked from commit 1d0d0aa)
- rename and rearragne resutls to be more consistently named and organized - separate multi-gpu results from single gpu results - use underscrore instead of space on names - each folder will only contain one set of benchmark results - separate amd and nvidia results to own subfolders Signed-off-by: Mika Laitio <[email protected]>
- allow specifying gpu-index parameter in addition of gpu-count parameter. - gpu index parameter can be used to request the benchmarks to be run only of certain gpu index in multi-gpu case - if more than one gpu, run benchmarks separately for each and then in the end run tests with all gpus used at a same time - fixes for: lamikr/rocm_sdk_builder#63 Signed-off-by: Mika Laitio <[email protected]>
use pytorch script to get device count instead of using amd or nvidia gpu specific commands and then execute the benchmark for each gpu found. Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
- defined minimal, medium and full set of models - use the amount of gpu memory as a criteria whether to run minimal. medium or full set of models (8GB device can run medium set without getting out of memory error) Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
- datafiles to be plotted must still be hardcoded in the code as a list - can show graphs that can be used to compare results from multiple amd and nvidia gpus - problem when showing the results is that the benchmarks that have been run on different GPU's have varied during the time. The list of benchmarks in full set is nowadays much bigger than on rx 1080/2060 release time. Signed-off-by: Mika Laitio <[email protected]>
- select more in detail which type of benchmarks are run depending from the amount of memery available (benchmarks executed for dounle type can be different than float type for example) Signed-off-by: Mika Laitio <[email protected]>
- add first benchmark results with the framework 16 laptop which has both the discrete 7700S gpu and internal M780M iGPU. - tests were executed with rocm sdk builder v6.1.2 version Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
- fixes the problem where the column resnet101 was by default before the resnet34 column Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
- latest rocm sdk builder has pytorch optimization in aotriton provided flash attenation algorithm for gfx1100/gfx1101/gfx1102 and gfx1103. Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
- improvements to logic that is used for selecting which models and precicious are benchmarked based on the amount of memory available on gpu Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
- trainign --> train - single --> float for couple of nvidia benchmarks which used different naming than other results Signed-off-by: Mika Laitio <[email protected]>
- execute at least 2 models also on small model case to allow showing line graphs easier Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
Signed-off-by: Mika Laitio <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
RX 7800 XT results from Bazza-63