ROCm 4.2 - pp_od_clk_voltage file doesn't exist anymore

[valeriob01]

A new way to undervolt gpus is needed.

[dgcampea]

rocm-smi --setvc can be used to undervolt. It should also be a bit more robust in general for GPU tuning than messing with /sysfs.

[valeriob01]

ROCm 4.2.0 requires installing the rock-dkms package. You also might need to add amdgpu.ppfeaturemask=0xfffd7fff kernel option. (check kernel docs for more info about the values)
Personally I've noticed that starting from 4.0.0, a reboot seems required for ROCm to work every time it gets updated.

I have rock-dkms installed (version 4.2-21) and have set up grub commandline options to amdgpu.ppfeaturemask=0xfffd7fff , also tried with amdgpu.ppfeaturemask=0xffffffff , both times I got same result as before.

[dgcampea]

Skimming through ROCm docs it doesn't seem to mention any changes to rocm-smi, my guess is that its system config related.

• Is secureboot enabled?
• Is the rock-dkms module being loaded?
• Does ROCm 4.2 work for gpuowl?
• Can you list the contents of the directory where pp_od_clk_voltage is at?
• What Distro are you using?

[valeriob01]

Skimming through ROCm docs it doesn't seem to mention any changes to rocm-smi, my guess is that its system config related.

* Is secureboot enabled?

* Is the rock-dkms module being loaded?

* Does ROCm 4.2 work for gpuowl?

* Can you list the contents of the directory where `pp_od_clk_voltage` is at?

* What Distro are you using?

• Secureboot not enabled, and never have used it.
• rock-dkms not being loaded
• ROCm 4.2 working with gpuowl every day 24/7
• ls /sys/class/drm/card1/device/

# ls /sys/class/drm/card1/device/
aer_dev_correctable       msi_irqs
aer_dev_fatal             numa_node
aer_dev_nonfatal          pcie_bw
ari_enabled               pcie_replay_count
boot_vga                  power
broken_parity_status      power_dpm_force_performance_level
class                     power_dpm_state
config                    pp_cur_state
consistent_dma_mask_bits  pp_dpm_dcefclk
current_link_speed        pp_dpm_fclk
current_link_width        pp_dpm_mclk
d3cold_allowed            pp_dpm_pcie
device                    pp_dpm_sclk
df_cntr_avail             pp_dpm_socclk
dma_mask_bits             pp_features
driver                    pp_force_state
driver_override           pp_mclk_od
drm                       pp_num_states
enable                    pp_power_profile_mode
fw_version                pp_sclk_od
gpu_busy_percent          pp_table
gpu_metrics               product_name
hwmon                     product_number
i2c-10                    ras
i2c-11                    remove
i2c-5                     rescan
i2c-6                     reset
i2c-7                     resource
i2c-8                     resource0
i2c-9                     resource0_wc
irq                       resource2
link                      resource2_wc
local_cpulist             resource4
local_cpus                resource5
max_link_speed            revision
max_link_width            rom
mem_busy_percent          serial_number
mem_info_gtt_total        subsystem
mem_info_gtt_used         subsystem_device
mem_info_vis_vram_total   subsystem_vendor
mem_info_vis_vram_used    thermal_throttling_logging
mem_info_vram_total       uevent
mem_info_vram_used        unique_id
mem_info_vram_vendor      vbios_version
modalias                  vendor
msi_bus

• Ubuntu 20.04

[dgcampea]

I have rock-dkms installed (version 4.2-21) and have set up grub commandline options to amdgpu.ppfeaturemask=0xfffd7fff , also tried with amdgpu.ppfeaturemask=0xffffffff , both times I got same result as before.

Just to be sure, did you regenerate grub.cfg after editing the /etc/default/grub? (sudo update-grub)

[valeriob01]

I have rock-dkms installed (version 4.2-21) and have set up grub commandline options to amdgpu.ppfeaturemask=0xfffd7fff , also tried with amdgpu.ppfeaturemask=0xffffffff , both times I got same result as before.

Just to be sure, did you regenerate grub.cfg after editing the /etc/default/grub? (sudo update-grub)

To be absolutely sure I just redid update-grub ... same result as before.

[selroc]

I see by reading the forum that other have stumbled in this issue. I repeat that Radeon Pro VII can't be overclocked/undervolted due to restrictions.