Looks like something mislead about PCIe GEN

[NTMan]

How nvtop calculate PCIe GEN information? I ask because that I see in nvtop does not match what I see via /sys/bus/pci/devices

Generation	Released	Speed
PCIe 1.0	2003	2 GT/s
PCIe 2.0	2007	5 GT/s
PCIe 3.0	2010	8 GT/s
PCIe 4.0	2017	16 GT/s
PCIe 5.0	2019	32 GT/s
PCIe 6.0	2021	64 GT/s

Which means GPU with 16 GT/s should have GEN 4

Here is my HW specs: https://linux-hardware.org/?probe=dd5735f315

[Umio-Yasuno]

possible duplicate of: https://github.com/Syllo/nvtop/issues/138#issuecomment-1532288951 https://gitlab.freedesktop.org/drm/amd/-/issues/1967

[NTMan]

❯ cat /sys/bus/pci/devices/0000:03:00.0/pp_dpm_pcie
0: 2.5GT/s, x1 78Mhz 
1: 5.0GT/s, x4 156Mhz 
2: 16.0GT/s, x16 623Mhz

Hmm... What means absence of asterisk symbol?

[Umio-Yasuno]

❯ cat /sys/bus/pci/devices/0000:03:00.0/pp_dpm_pcie
0: 2.5GT/s, x1 78Mhz 
1: 5.0GT/s, x4 156Mhz 
2: 16.0GT/s, x16 623Mhz

Hmm... What means absence of asterisk symbol?

Oh, the same problem occurred on RX 6600.
Probably it is an AMDGPU driver issue, I will report it to drm/amd.

[Umio-Yasuno]

Reported: https://gitlab.freedesktop.org/drm/amd/-/issues/1447#note_2101382

[Syllo]

Hello, Nvtop extracts the PCIe bus from the device using libudev and does not rely on pp_dpm_pcie (code). The code walks the PCIe bus from the device up to the root and takes the smallest value for speed and width it sees on the path.

The link speed is translated similar to your table here.

It can be that the reporting is wrong somewhere through the PCIe path. Also recent GPUs have internal PCIe switches that dynamically lower/increase the PCIe speed to reduce power consumption. Do you see the same value while running heavy GPU workloads (games, AI)?

[NTMan]

Do you see the same value while running heavy GPU workloads (games, AI)?

Yes