Age | Commit message (Collapse) | Author |
|
|
|
|
|
|
|
|
|
|
|
|
|
upstream release
Update rocm-smi to v3.3.0
|
|
|
|
|
|
|
|
[Changelog](https://github.com/RadeonOpenCompute/ROC-smi/commits/roc-2.9.x)
rocm-smi 2.8.0
[Changelog](https://github.com/RadeonOpenCompute/ROC-smi/commits/roc-2.8.x)
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
* Various fixes
* --showproductname uses the pci.ids file to report information, but this
file may be outdated or missing. Update the README to clarify this,since we won't force 'update-pciids' to be executed during SMI
execution, since that's invasive (and time-consuming)
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
Changes:
* Show Device SKU and Vendor
* Fix device info for mGPU & Vendor check
* Fix clock printout if DPM is disabled
* Add --showuniqueid flag to display Unique ID
* Fix finding marketing information on CentOS
* Clarify error message for PP table output
* Show GPU Memory Utilization
* Add --showpids flag to display PIDs for KFD processes
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
Minor fixes release
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
Minor fixes release
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
A Big release with version 2.3.0 this month, detailed changes can be
found on github
https://github.com/RadeonOpenCompute/ROC-smi/commits/master
1. Print PCIe bus location in --showhw flag
2. Change references of pclk to pcie
3. Clarify parameters in RAS help text
4. Add support to reset the GPU
5. README: Update README to reflect SMI help text
6. Allow for specifying multiple devices
7. Remove sudo requirement for concise output
8. SMI: Add --showvoltage flag to display GPU voltage
9. Add error reporting deprecation of pp_*clk_od
10. Fix getting Max Level
11. Fix setprofile for Vega20
12. Fix logging reference to device
13. Add initial RAS support
14. Add support for displaying memory usage information
15. Fix Power Profile set errors
16. Add logging to printLog function
17. Only load values if they exist
18. Overhaul logging
19. Check that a profile exists before parsing it
20. Handle race condition on getFanSpeed
21. Clean up concise output and header/footer
22. Add support for newer clock types
23. Don't throw an error if a file doesn't exist
24. Make clocktypes expandable, and use proper naming
25. Fix overdrive test
26. Use getFilePath instead of constantly using os.path.join
27. Don't show pcie_bw for --showallinfo, show voltages instead
28. Fix logic for checking max level
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
Add support for showing PCIe bandwidth
Add support for displaying the estimated usage of PCIe bandwidth over
the last second. This sysfs file will report the estimated bytes received
and sent over the last second, as well as the maximum packet size, and
the SMI will calculate estimated bandwidth using those values.
Properly allow setting more than one DPM mask
Previously, by concatenating all of the values for SetClocks
together without spaces between them, we would try to combine
multiple requested DPM states into a single string of integers.
For example, trying to set "--setsclk 0 2 4 6" should "turn on"
SCLK DPM states 0, 2, 4, and 6. This would try to set 0246 which
resulted in wrong masks being put into the driver.
We should still check that all the values being sent in are
integers, so this patch first makes one fully-concatenated
variable and ensures that it is still an integers. However, it
actually passes in to sysfs a value that is space separated.
Don't include non-AMD devices by default
Only list/control AMD devices unless the --alldevices flag is passed in,
this will avoid printing a bunch of unsupported fields, since it's not
up to our kernel to support them. And avoid a bunch of error message
regarding being unable to set values on devices that aren't AMD GPUs
Some things may be usable, depending on the driver code in the upstream
kernel (like temperature or fan speed on certain GPUs), but for the most
part, it's stuff that the SMI can't control, which just wastes screen space,
and most users don't care about trying to manage AMD and non-AMD devices in
the same tool. The ones that do can use --alldevices
Fix logic for checking max level
0 can be a max level, so check that the level is None, don't check
that level isn't "false" , since 0 is a valid return for the function
Since 0 is a valid return, change a failure to read from sysfs to None,
and check the getMaxLevel returns instead of assuming that they worked
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
New Release with several new additions, see the upstream git for
details:
https://github.com/RadeonOpenCompute/ROC-smi/
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
1. Temporary patch to fix --setpoweroverdrive feature until next official release
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
* Fix setting and resetting of OverDrive DPM tables
* Various PowerPlay/OD fixes
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
1. Removed empty variables
2. Used built in makepkg feature to extract sources instead of `tar`
3. Removed redundant variables
Thanks for the Tips Eli :)
Signed-off-by: Rigo Reddig <rigo.reddig@gmail.com>
|
|
|