|
Hi all, I would like to monitor the frequency of the iGPU of my TGL platform while running glmark2 in bursts which means glmark2 is 5sec
running and 5sec not running.
I disabled RC6 by echo 0 > /sys/class/drm/card0/gt_rc6_enable for my tests. Furthermore, I tried to lock the iGPU frequency to min, RP0 or RP1 by
setting e.g. echo 1300 > /sys/class/drm/card0/gt_min_freq_mhz echo 1300 > /sys/class/drm/card0/gt_max_freq_mhz echo 1300 > /sys/class/drm/card0/gt_boost_freq_mhz.
For frequency monitoring I am using
intel_gpu_top and
/sys/class/drm/card0/gt_cur_freq_mhz. I am a bit confused if I compare the output because
intel_gpu_top and
/sys/class/drm/card0/gt_cur_freq_mhz
show different results. These are my results if I lock gpu Frequency to 1300MHz
### /sys/class/drm/card0/gt_cur_freq_mhz Every 0.1s: cat /sys/class/drm/card0/gt_cur_freq_mhz node4-tglu-rvp: Mon Jan 10 20:22:59
2022 1300 ### intel_gpu_top root@node4-tglu-rvp:~/fGvPOC# intel_gpu_top -l Freq MHz IRQ RC6 RCS/0 BCS/0 VCS/0 VCS/1 VECS/0 req act /s % % se wa % se wa % se wa % se wa % se wa 0 0 15 100 0.08 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0 0 16 100 0.08 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0 0 10 100 0.05 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 188 172 6045 86 13.22 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 1300 1189 44907 0 99.25 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 1300 1201 44702 0 99.05 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 1300 1182 44873 0 99.23 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 1300 1193 44830 0 99.08 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 1066 988 36336 19 80.24 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 6 6 10 100 0.05 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 13 13 15 100 0.08 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 13 13 16 100 0.08 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0 0 13 100 0.05 0 0 0.00 0 0 0.00 0 0 0.00 0 0 0.00 0 0 Can someone help me to understand why
/sys/class/drm/card0/gt_cur_freq_mhz always shows 1300MHz and intel_gpu_top (req and act)is dropping to almost 0Hz when glmark2 is not running?
My final goal is to figure out if render p-state transitions can impact the timeliness of my real-time workload running on a RT core.
Kernel Info: Linux node4-tglu-rvp 5.10.41-rt42-intel-ese-standard-lts-rt #1 SMP PREEMPT_RT Thu Sep 23 10:21:35 UTC 2021 x86_64 x86_64 x86_64 GNU/Linux Intel Yocto BKC MR3 Thx, Markus Intel Deutschland GmbH |
